Using Mathematical Models to Inform Syphilis Control Strategies in Men Who Have Sex With Men

Transcription

1 Using Mathematical Models to Inform Syphilis Control Strategies in Men Who Have Sex With Men by Ashleigh Tuite A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Institute of Medical Science University of Toronto Copyright by Ashleigh Tuite 2015

2 Using Mathematical Models to Inform Syphilis Control Strategies in Men Who Have Sex with Men Abstract Ashleigh Tuite Doctor of Philosophy Institute of Medical Science University of Toronto 2015 Syphilis is resurgent in many high-income countries, disproportionately affecting urban men who have sex with men (MSM). Frequent screening of at-risk individuals remains the best available tool for syphilis control, but current public health efforts are not resulting in reduced disease burden. The aim of this thesis was to use mathematical modeling to understand the effect of different approaches to syphilis screening on epidemic dynamics and the health of MSM. An agent-based model of syphilis transmission in a core group of sexually active MSM was parameterized with data on the epidemiology of the current epidemic to evaluate plausible screening strategies that might be employed for epidemic control. Of the strategies evaluated, more frequent screening of at-risk MSM already accessing screening, rather than expanding outreach to provide screening to unscreened individuals, was found to be the most effective means of reducing syphilis incidence over a 10-year intervention period. A state-transition microsimulation model of syphilis natural history and medical care was developed to determine the cost-effectiveness of incorporating routine syphilis testing into the blood-work of MSM under care for HIV. When rates of syphilis acquisition were high, opt-out syphilis screening in HIV-infected MSM was projected to be a highly cost-effective intervention. A risk-structured deterministic compartmental mathematical model of syphilis transmission in MSM was used to examine the impact of sustained syphilis screening at varying levels of population coverage. Increasing screening in a population with initially low levels of coverage was shown to lead to increases in infection incidence. Although screening has the potential to ii

3 control syphilis outbreaks, suboptimal screening coverage may result in the establishment of higher equilibrium infection incidence than that observed in the absence of the intervention, possibly contributing to outbreak persistence. The results of this research suggest that current control efforts are not expected to reduce syphilis incidence, with more effective screening programs required to reduce syphilis burden in MSM. The work presented in this thesis provides some insight into factors that may lead to screening programs that both improve the health of individuals and reduce the overall population burden, ultimately resulting in improved epidemic control. iii

4 Acknowledgments This has been a long journey. I wouldn t have made it this far without my supervisor, Dr. David Fisman, and his tremendous encouragement, guidance, and enthusiasm. His willingness to give me a chance when I was just starting out and his unwavering belief in my abilities have made all of the difference. Thanks also to my committee members, Dr. Amy Greer and Dr. Rupert Kaul, for their expertise and for encouraging me to think critically. The initial ideas for this thesis arose from interactions with individuals working to control the ongoing syphilis outbreak in Ontario. Many thanks to Frank McGee and James Murray (Ontario AIDS Bureau), Dr. Robert Remis and Dr. Dionne Gesink (University of Toronto), Dr. Rita Shahin and Dana Al-Bargash (Toronto Public Health), Dara Friedman and Jacqueline Willmore (Ottawa Public Health), Dr. Ann Burchell (Ontario HIV Treatment Network) and Leo Mitterni (Hassle Free Clinic) for providing valuable input and data. Dr. Ahmed Bayoumi was extremely generous with his time, helping me learn the ropes of microsimulation modeling. Dr. Sharmistha Mishra has been an excellent mentor, introducing me to the wonderful world of syphilis and being a source of support, advice, and knowledge. Finally, thanks to my family for putting up with my seemingly endless schooling your support and patience mean the world to me. I was supported by generous funding by the Canadian Institutes of Health Research (Doctoral Research Award) and the Institute of Medical Science U of T Open Fellowship Award. iv

5 Contributions I was primarily responsible for the work presented in this thesis, including study design, model building, analysis, and drafting of manuscripts. However, other authors contributed as follows: Chapter 3: Sharmistha Mishra and David Fisman assisted with study conception, model design and parameterization, data acquisition, and critical revision of the manuscript for important intellectual content. Chapter 4: Ann Burchell contributed data and helped with study design. David Fisman helped with study design and helped revised the manuscript. Chapter 5: David Fisman helped with study design and revision of the manuscript. v

6 Table of Contents ACKNOWLEDGMENTS... iv CONTRIBUTIONS... v TABLE OF CONTENTS... vi LIST OF TABLES... ix LIST OF FIGURES...x LIST OF ABBREVIATIONS... xii CHAPTER 1: INTRODUCTION SYPHILIS NATURAL HISTORY Infection Transmission Stages of Infection EPIDEMIOLOGY Historical Trends Current Epidemiology Epidemiology of Syphilis in Toronto Risk Factors and Core Groups Syphilis and HIV Infection SYPHILIS DIAGNOSIS Diagnostic Tests Testing Algorithms SYPHILIS CONTROL Treatment Public Health Approaches to Control IMMUNITY Immune Response to Infection Protective Immunity with Untreated Infection Evidence of Protective Immunity Following Treatment SYPHILIS RESURGENCE Historical Context Proposed Explanations for Syphilis Resurgence Current Approaches to Syphilis Control in MSM vi

7 1.7 MATHEMATICAL MODELS OF SYPHILIS Types of Mathematical Models Key Concepts in Mathematical Modeling of Infectious Diseases Review of Existing Mathematical Models of Syphilis Models Examining Syphilis Natural History and Dynamics Models Examining Effectiveness of Syphilis Control Measures Models Examining Cost-Effectiveness of Public Health Control Measures Summary of Mathematical Modeling Studies CHAPTER 2: RESEARCH AIMS AND HYPOTHESIS RATIONALE HYPOTHESIS RESEARCH AIMS CHAPTER 3: SCREEN MORE OR SCREEN MORE OFTEN? USING MATHEMATICAL MODELS TO INFORM SYPHILIS CONTROL STRATEGIES ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION CHAPTER 4: COST-EFFECTIVENESS OF ENHANCED SYPHILIS SCREENING AMONG HIV- POSITIVE MEN WHO HAVE SEX WITH MEN: A MICROSIMULATION MODEL ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION CHAPTER 5: GO BIG OR GO HOME: IMPACT OF SCREENING COVERAGE ON SYPHILIS INFECTION DYNAMICS ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION vii

8 CHAPTER 6: DISCUSSION SUMMARY OF FINDINGS AND CONTRIBUTION TO THE LITERATURE METHODOLOGICAL CONSIDERATIONS GENERALIZABILITY OF FINDINGS CHAPTER 7: FUTURE DIRECTIONS AND CONCLUSIONS FUTURE DIRECTIONS Using Mathematical Modeling to Improve Syphilis Control and Forecast Likely Program Impacts Integrating Surveillance and Modeling to Evaluate Program Impact and Manage Uncertainty Modeling HIV Co-Transmission to Better Quantify the Risks and Benefits of Syphilis Control Interventions Incorporating Behaviour Change into Mathematical Models Investigating Repeat Infections to Improve the Understanding of Syphilis Natural History CONCLUSIONS REFERENCES APPENDIX A: DETAILS OF THE AGENT-BASED MODEL USED IN CHAPTER APPENDIX B: DETAILS OF THE COMPARTMENTAL MODEL USED IN CHAPTER COPYRIGHT ACKNOWLEDGEMENTS viii

9 List of Tables Table 1-1. Results of T. pallidum challenge experiment in cases with different histories of syphilis infection and treatment Table 1-2. Mathematical modeling studies examining syphilis natural history or control Table 3-1. Model parameters Table 3-2. Syphilis screening strategies evaluated in the model Table 4-1. Model variables and sources Table 4-2. Model costs and utilities Table 4-3. Discounted health and economic outcomes associated with different syphilis screening strategies Table 4-4. Preferred syphilis screening strategies under alternate model assumptions and for different willingness-to-pay thresholds Table 4-5. Undiscounted health and economic outcomes associated with different syphilis screening strategies Table 5-1. Model parameter values Table 5-2. Parameter values used for different model assumptions, derived by model calibration ix

10 List of Figures Figure 1-1. Overview of the natural history of untreated syphilis Figure 1-2. Reported syphilis cases and deaths in Canada Figure 1-3. Rates of early syphilis in Canada, by sex Figure 1-4. Rates of early syphilis in Toronto, Ontario Figure 1-5. A comparison of the traditional and reverse screening algorithms for the diagnosis of syphilis Figure 1-6. Estimated rates of primary and secondary syphilis in men who have sex with men. 29 Figure 1-7. Example of the importance of heterogeneity in models of sexually transmitted infections Figure 3-1. Schematic of the infection transmission component of the model Figure 3-2. Model-projected annual rates of reported early syphilis Figure 3-3. Model-projected annual rates of reported early syphilis under equivalent test volume strategies Figure 3-4. Expected reduction in syphilis cases following implementation of different intervention strategies Figure 4-1. Simplified overview of screening and treatment component of the decision analytic model Figure 4-2. Markov model overview Figure 4-3. Model validation and projections Figure 4-4. Strategy acceptability for different willingness-to-pay thresholds Figure 5-1. Outline of model structure, showing population flows between compartments x

11 Figure 5-2. Model calibration Figure 5-3. Model-projected syphilis trends with different population-level annual screening coverage Figure 5-4. Model-projected syphilis incidence at equilibrium Figure 5-5. Thresholds for local syphilis elimination under alternate model assumptions Figure 5-6. Impact of assumptions relating to the development of protective immunity on equilibrium syphilis incidence xi

12 List of Abbreviations ART CLIA CSF EIA FTA-ABS FSW HAART HIV ICER MHA-TP MSM PHAC POC QALY R0 Re RCT RPR SSA STI TPPA VDRL WHO Antiretroviral therapy Chemiluminescence assay Cerebrospinal fluid Enzyme immunoassay Fluorescent treponemal antibody absorption test Female sex worker Highly active antiretroviral therapy Human immunodeficiency virus Incremental cost-effectiveness ratio Microhemagglutination assay for T. pallidum Men who have sex with men Public Health Agency of Canada Point of care Quality-adjusted life year Basic reproduction number Effective reproduction number Randomized controlled trial Rapid Plasma Reagin Sub-Saharan Africa Sexually transmitted infection Treponema pallidum particle agglutination assay Venereal Disease Research Laboratory test World Health Organization xii

13 Chapter 1: Introduction 1

14 2 If I were asked which is the most destructive of all diseases I should unhesitatingly reply, it is that which for some years has been raging with impunity What contagion does thus invade the whole body, so much resist medical art, becomes inoculated so readily, and so cruelly tortures the patient? Desiderius Erasmus, 1520 [1]. Known as the great imitator for its protean clinical presentations, syphilis is a sexually transmitted infection caused by the spirochete Treponema pallidum subspecies pallidum. T. pallidum is an obligate human parasite, with no known animal or environmental reservoirs. The first recorded outbreak of syphilis occurred in Naples, Italy in 1494, where it subsequently spread rapidly across Europe and to India and China, ultimately establishing itself as an endemic infection globally [2]. Since that time, much progress has been made in our understanding of the pathogen, its interaction with human hosts, management and treatment of disease, and modes of controlling infection spread, to the point that syphilis elimination was considered an achievable goal at the end of the 20 th century [3]. Despite this optimism, syphilis control remains a global public health challenge. In this chapter, I will review our current understanding of syphilis and its control, as well as discuss how mathematical modeling can be applied to the challenge of syphilis control. 1.1 Syphilis Natural History Infection Transmission The primary mode of transmission is by direct sexual contact with lesions of individuals with primary or secondary syphilis. Data on transmissibility are derived from tracing of sexual contacts of infected cases and prospective studies of prophylactic treatment that included exposed placebo groups [4]. Infection rates in partners of known cases ranged from 18-88% in contact tracing studies [5-10]. Prospective studies are less subject to biases, and report between 9 and 63% of individuals with known exposure developing syphilis in the absence of treatment [11-13]. No data on frequency or types of sexual acts are presented in these studies, so these estimates are conservatively assumed to represent transmission probabilities per sexual partnership [4].

15 3 Vertical transmission from infected mother to fetus is the other major mode of transmission. Most infants with congenital infection are infected in utero, but contact with a genital lesion during delivery can also result in infection transmission [14]. The risk of transmission depends on infection stage in the mother during pregnancy and may be a function of spirochete burden in the blood. Risk decreases as maternal disease progresses: it is estimated at % for primary or secondary syphilis, 40% for early latent syphilis, and 10% for late latent syphilis [15, 16]. Although rare, vertical transmission with tertiary syphilis has been reported [2] Stages of Infection Infected individuals progress through different clinical stages of infection, although delineations between stages are not necessarily definitive and may vary from individual to individual. For instance, individuals may have signs of both primary and secondary syphilis [17]. A simplified overview of the natural history of syphilis infection is presented in Figure 1-1. Following an incubation period of approximately 21 days, infected individuals enter the primary stage of infection, which is typically characterized by the development of a painless chancre at the site of inoculation [18]. Chancres typically heal without treatment. The signs of secondary syphilis usually occur 4 to 10 weeks after the appearance of the chancre, involving hematogenous dissemination [18]. Manifestations of secondary syphilis include maculopapular rash on flank, shoulder, arm, chest, back, hands and soles of feet, alopecia, condylomata lata, and mucous patches [19]. Rash is the primary symptom among those seeking medical care and is found on physical examination in greater than 90% of patients [18, 20]. Sore throat, malaise, headache, and lymphadenopathy are other common symptoms associated with secondary syphilis [21]. Although clinical symptoms resolve without treatment, organisms remain detectable, usually in the lymph nodes and spleen [22]. The primary and secondary stages are typically considered the infectious stages of disease, during which transmission can occur via sexual contact. Following secondary syphilis, individuals enter the latent stage of syphilis infection. This stage is generally divided into early and late stages. Early latent syphilis describes asymptomatic infection during the first year following infection.

16 4 Recurrent syphilis days 4 10 weeks 25% Infection Primary syphilis Secondary syphilis Early Latent syphilis Late ( 1 year postinfection) (>1 year postinfection) Central nervous system invasion 25 60% Asymptomatic early neurosyphilis 5% Symptomatic early neurosyphilis Tertiary syphilis 2-5% 2-9% 10% 15% Late neurosyphilis General paresis Tabes dorsalis Cardiovascular syphilis Gummatous disease (2 30 years post-infection) (3 50 years post-infection) (20 30 years post-infection) (1 46 years post-infection) Figure 1-1. Overview of the natural history of untreated syphilis. Primary, secondary, and recurrent syphilis are considered infectious via sexual transmission. Individuals with central nervous system invasion may develop symptomatic early neurosyphilis or late neurosyphilis, or remain asymptomatic. Adapted from Golden et al. [18].

17 5 Recurrences, in the form of mucocutaneous lesions, have been noted to occur in approximately 25% of untreated patients with latent infection, most commonly in the first year of infection [17]. Due to the risk of recurrence to secondary syphilis, some guidelines classify early latent syphilis as an infectious stage [14]. The term late latent describes asymptomatic syphilis infection after the first year of infection. This period is associated with relative immunity to recurrence of active disease and increased resistance to re-infection with homologous treponemal strains [22]. Individuals with untreated latent syphilis may develop sequelae years or decades after initial infection, as a result of treponemes invading the skin, eyes, central nervous system, cardiovascular system, or other internal organs, producing damage as a result of invasion and inflammation [22]. Tertiary syphilis is rarely seen today, likely due to the effectiveness of syphilis control programs and prevalence of antibiotic treatment [21]. However, in the era before antibiotics, tertiary syphilis was a major contributor to cardiovascular and neurological morbidity. In the 1920s, it was estimated that greater than 20% of all patients in US mental institutions had neurosyphilis [23]. In the Oslo Study, which began in 1891, nearly 2000 patients with early syphilis were observed for 20 years or more. Approximately one-third of cases developed tertiary syphilis [17]. Tertiary syphilis is not transmissible by sexual contact, and vertical transmission is rare [22]. The different manifestations of tertiary syphilis depend on the site of treponemal invasion. Cardiovascular symptoms, including aneurysm, aortitis, and aortic endocarditis may result from replication of treponemes in the wall of the aorta [22]. Gummas are destructive lesions in the skin, bones, or viscera, that can range in size from microscopic to large tumour-like masses [8]. Central nervous system involvement can occur any time after primary infection, and is often asymptomatic [21]. Diagnosis of symptomatic neurosyphilis is based on both clinical manifestations and cerebrospinal fluid (CSF) abnormalities [24]. Symptomatic neurosyphilis can be subdivided into clinical, occasionally overlapping, syndromes [21]: meningeal syphilis, which usually occurs within the first year of infection and is characterized by headache, stiff neck, nausea, and vomiting, and may involve the cranial nerves; meningovascular syphilis, commonly occurring 4-7 years after infection and presenting as focal central nervous system ischemia or stroke; and parenchymatous neurosyphilis, occurring decades after infection and presenting as general paresis or tabes dorsalis.

18 6 Adverse outcomes associated with untreated syphilis during pregnancy include late abortion or stillbirth, neonatal death, prematurity or low birth weight, and syndromes of congenital syphilis [25]. 1.2 Epidemiology Historical Trends Before the discovery that syphilis was easily treatable with penicillin, it was a ubiquitous sexually transmitted infection. The pervasiveness of syphilis is highlighted by its association with the demise of many notable historical figures [26-28]. Although a diagnostic test for syphilis was not available until 1906, general paresis was noted to be on the rise in Canada during the last quarter of the 19th century [29]. A number of prevalence estimates were generated once a means of accurate serological diagnosis became available. At Toronto s Hospital for the Insane, 25 percent of male admissions in 1910 were diagnosed with general paresis (in contrast with 2 percent of female admissions), while 12 percent of admissions at Toronto s Hospital for Incurables were due to syphilis [29]. In 1916, 12 percent of all patients (regardless of reason for admission) at the Toronto General Hospital tested positive for syphilis [29]. Based on these and other data, it was estimated that between 5 and 15 percent of the general Canadian population was infected at the beginning of the twentieth century [29]. These estimates were in agreement with those from other countries, including the United States, England, and Germany [29-31]. Syphilis was also a major cause of mortality (Figure 1-2). In the United States, syphilis mortality rates in the early to mid-20 th century reached over 15 per 100,000, comparable to the AIDS mortality rates in the mid-1990s [32]. Syphilis has been a notifiable disease in Canada since With the introduction of penicillin and the implementation of screening, diagnoses of syphilis (all stages, including late latent) in Canada declined from 138 cases per 100,000 in 1944 to a low of 1.7 cases per 100,000 in 1997 (Figure 1-2) [33]. Syphilis-attributable mortality decreased from a high of 8.2 per 100,000 in 1942 to less than 0.2 per 100,000 by 1970 (the last year with available data) [34]. Despite the overall downward trend in diagnoses over this time period, transient upsurges in syphilis rates were observed in the mid-1970s and mid-1980s.

19 7 Figure 1-2. Reported syphilis cases and deaths in Canada. All reported syphilis cases are included, regardless of stage, for the years Syphilis-attributable deaths are available for the years Sources: Public Health Agency of Canada Notifiable Diseases On-Line [35] and Venereal Disease in Canada Annual Report [34].

20 Current Epidemiology Beginning in the early 2000s, syphilis diagnoses across Canada began to increase. In 2011, there were 1757 early syphilis cases reported in Canada, compared with 115 cases in 1997 [36]. In contrast to what was observed in the previous decade, cases occurred disproportionately in males, with men accounting for 93.5% of cases in 2011 (Figure 1-3). In the absence of data on the sex of partners of cases, male-to-female rate ratios can be used as a marker of male-tomale transmission. The male-to-female rate ratio for early syphilis in 2011 was 13.7, compared to 1.3 in 1997, when syphilis reached its nadir [36]. Increased diagnoses of early syphilis have been reported in all Canadian provinces and territories in recent years. Although the affected population subgroups have varied, many jurisdictions have noted the high prevalence of men reporting sex partners of the same sex among cases, especially in urban centres [37-48]. From 2002 to 2007, rates of early syphilis in Canada were highest in males aged 30-39, although more recently, high rates in year old males have also been observed [36]. Despite recent increases in syphilis in adults, congenital syphilis is relatively rare in Canada, where universal screening of all pregnant women remains the standard of care [14]. However, rates of congenital syphilis did increase in the mid-2000s, particularly in the province of Alberta, which had experienced a predominantly heterosexual syphilis outbreak [36, 49]. Since 2011, rates have declined to less than 1 per 100,000 live births [36]. In 2012 there were three reported cases of congenital syphilis [35]. Recent trends in Canada mirror what has been observed in other high-income countries; syphilis re-emergence in men who have sex with men (MSM) has been reported in urban settings across the United States, Western Europe, and Australasia [18, 50-57]. The most striking epidemiological feature in high-income countries has been the disproportionate representation of men who have sex with men among cases, with incidence also varying by geography and ethnicity [22, 58-60]. There is also a strong association with HIV co-infection (discussed below).

21 9 Figure 1-3. Rates of early syphilis in Canada, by sex. Reported rates of primary, secondary, and early latent syphilis are included for the years The rate ratio of male to female cases is plotted on the right-hand axis. Source: Public Health Agency of Canada [36].

22 10 Despite the focus of this review on recent syphilis resurgence in high-income countries, it should be noted that the bulk of syphilis disease burden continues to occur in low-income countries. Globally, an estimated 10.6 million new cases of syphilis occurred in 2008, with greater than 60% of cases occurring in Africa and South-East Asia [61]. In low-income countries, heterosexual intercourse is the primary mode of transmission, although vertical transmission is also common. Although the incidence of congenital syphilis has been dramatically reduced in highincome countries, it still remains the most common infectious cause of fetal loss or stillbirth worldwide [62]. An estimated 520,000 cases of congenital syphilis occur annually, resulting in 215,000 stillbirths or fetal deaths, 90,000 neonatal deaths, 65,000 preterm or low birth weight infants, and 150,000 infants with congenital disease [63] Epidemiology of Syphilis in Toronto As with many other urban centres, the Canadian city of Toronto has experienced elevated rates of early syphilis since the early 2000s (Figure 1-4). The outbreak in Toronto is of major importance and worthy of direct study for several reasons. First, the city is Canada s largest population centre; the Greater Toronto Area contains approximately 17% of the Canadian population, and is highly connected via air, road and rail travel to other North American and international centres. Secondly, thanks to extensive epidemiological and behavioural research conducted in Toronto, the city s sexually active MSM community, which is disproportionately affected by the recent surge in syphilis incidence, is relatively well characterized with respect to behavioural parameters necessary for realistic modeling of syphilis dynamics [64]. Thirdly, the city of Toronto has been actively engaged in syphilis control efforts since the epidemic was first recognized, meaning that data have been collected both on the epidemic itself and on efforts instituted to control it. These factors combine to make modeling of syphilis in Toronto, with an aim to understand and ultimately control the spread of this disease in the city, both critically important, and possible.

23 11 Figure 1-4. Rates of early syphilis in Toronto, Ontario. Rates of primary, secondary, and early latent syphilis, including infectious neurosyphilis, for males and females are show for the years 2002 to 2013, The rate ratio of male to females is plotted on the right-hand axis (note logarithmic scale). Source: Toronto Public Health [65]

24 12 In 2013, the most recent year for which data are available, a total of 458 cases of early (primary, secondary, early latent and infectious neurosyphilis) syphilis were reported, for an overall rate of 16.4 cases per 100,000, compared with a mean of 504 cases per year occurring in the 5-year period between 2008 and 2012, for an overall rate of 18.5 per 100,000 per year [65]. The mean age of cases was 39 years [65]. Greater than 95% of cases were reported in males in 2013; among these male cases, 79% were years of age [65]. Rates in males declined from 39.6 cases per 100,000 in 2012 to 32.3 cases per 100,000 in 2013 [65]. Between 2009 and 2013, 27% of cases were diagnosed with primary syphilis, 38% with secondary syphilis, 33% with early latent syphilis, and 2% with infectious neurosyphilis. The percentage of reported early syphilis cases with HIV co-infection has averaged 47% per year over the past 5 years, although in 2013, the percentage was 43%, the lowest reported since The proportion of HIV co-infected early syphilis cases diagnosed during the early latent stage of infection has been increasing over the past 5 years; from 2009 to 2011, early latent cases represented approximately 32% of early syphilis cases diagnosed in HIV-infected individuals, in 2012, 44% of cases were diagnosed with early latent infection, while in 2013, early latent cases represented 51% of cases [65]. This trend is less apparent in the overall case data, but early latent syphilis cases have represented an increasing proportion of early syphilis diagnoses in recent years [65]. Of 6 reported cases of early neurosyphilis in 2013, 4 occurred in HIV-positive individuals [65]. Sex with same sex is consistently the most common risk factor reported during this ongoing outbreak and was reported by 87% of male cases in 2013 [65]. Failure to use a condom was reported as a risk factor by 63% of cases [65] Risk Factors and Core Groups Among MSM, the risk of syphilis acquisition is associated with certain behaviours, including having anonymous and/or multiple sex partners, use of illicit drugs, and unprotected oral sex [51, 66-70]. Venues where men meet sex partners are also an important aspect of the ongoing epidemic in MSM [71, 72]; meeting partners via the internet or mobile applications, bathhouses, and public spaces facilitates anonymous sex, and makes partner notification, a critical component of syphilis control, challenging. Significant proportions of MSM with syphilis report sex partners who were not local [68, 71], and travel may lead to synchronization of outbreaks across cities [73]. Links between the adoption of behaviours to reduce HIV

25 13 transmission or acquisition risk and increased syphilis incidence have been noted [74]. In particular, serosorting, where individuals with concordant HIV status intentionally engage in unprotected sex, reduces HIV risk but leaves men at risk for other sexually transmitted infections (STIs)[75]. A survey of MSM in several Canadian cities identified serosorting as a common practice [76]. The concept of the core group ties together sexual behaviours that place individuals at increased risk of syphilis infection and community-level differences in syphilis transmission and persistence. A core group describes an interconnected network of people that contribute disproportionately to the spread of STIs [77]. The idea of the core group was first conceptualized to explain the observation that despite the absence of immunity to re-infection, transmission of gonorrhea among sexually active populations is not as widespread as would be predicted [78]. Yorke et al. [78] proposed that there are subgroups within communities that do not interact randomly with each other, and that this non-random mixing is what limits transmission, while at the same time permitting persistence of endemic infection among core group members. Members of core groups may variously be identified based on occurrence of repeat infections, having high numbers of sexual partners, engaging in high-risk occupations, such as commercial sex work, or residing in geographic areas with large numbers of cases [77]. Core groups are believed to be of strategic importance for sustaining STI transmission, and there is great interest in identifying characteristics that can be used to target core group members for epidemic surveillance and interventions [42, 77, 79-82] Syphilis and HIV Infection HIV-infected individuals are disproportionately represented among syphilis cases, and syphilis-infected individuals are at increased risk of HIV infection. For example, approximately 47% of early syphilis cases occurring between 2009 and 2013 in Toronto reported HIV coinfection [65]. Studies in Toronto and Seattle have estimated the annual incidence of syphilis infection among HIV-infected MSM to be 4% [57, 83]. In an HIV pre-exposure prophylaxis trial in HIV-negative MSM, incident syphilis infection was shown to be associated with HIV acquisition (hazard ratio: 2.6, 95% confidence interval: ) [84]. Syphilis infection may enhance HIV acquisition risk, with genital ulcers acting as portals of entry for HIV by recruiting HIV susceptible inflammatory cells to the genital tract and disrupting mucosal barriers to

26 14 infection [85]. Syphilis infection may also increase HIV transmission risk by facilitating HIV shedding in the genital tract [85, 86]. Interestingly, in Canada and the United States, the trend of increasing syphilis incidence in MSM does not appear to be associated with a concomitant rise in HIV incidence [22], possibly due to the prevalence of seroadaptive behaviours. There is also a connection between syphilis and HIV at the ecological level [87]. Syphilis incidence in MSM began to decline in the early 1980s, contemporaneously with the recognition of the HIV epidemic. The introduction of highly active antiretroviral therapy (HAART) for the treatment of HIV and its marked impact on mortality among HIV-infected individuals was also coincident with an increase in syphilis rates in MSM in the late 1990s. Although it is not possible to demonstrate causality (and would likely be an overly simplistic explanation), these trends are suggestive of HIV-driven behaviour change in MSM contributing to changes in syphilis incidence. Evidence from case reports and retrospective studies suggests that HIV co-infection may alter the clinical presentation and natural history of syphilis infection [18]. Reported atypical presentations of early syphilis in HIV co-infected individuals include multiple chancres, delayed chancre healing, abnormal CSF findings, and more frequent and accelerated development of neurosyphilis [18, 20, 88, 89]. Serologic response to treatment may also be suboptimal [90-94]. Current treatment recommendations are the same for HIV-infected and uninfected individuals, although some experts recommend treating early infection the same as late latent infection (i.e., 3 weekly injections of penicillin G instead of a single dose) in HIV-positive individuals [14]. The observation that HIV co-infected individuals are more likely to develop neurologic complications has led to recommendations for CSF examination of all HIV co-infected cases with a CD4 cell count less than 350 cells/ml, even in the absence of neurologic symptoms or signs and regardless of the stage of syphilis infection [14]. 1.3 Syphilis Diagnosis Diagnostic Tests The causative agent of syphilis, the spirochete Treponema pallidum subspecies pallidum, was first discovered in 1905 and the first blood test was developed in 1906 [95]. Because T. pallidum cannot be cultured continuously in the laboratory, diagnosis is primarily achieved using

27 15 serologic testing [96]. A definitive laboratory diagnosis of early syphilis infection is made by dark-field microscopic examinations and direct fluorescent antibody test of lesion exudate or tissue [22]. Polymerase chain reaction (PCR) tests can also be used for direct detection of T. pallidum DNA in skin and mucous membrane lesions, but has low sensitivity in non-ulcerative specimens [97]. Dark-field microscopy is currently infrequently used for diagnosis, and presumptive diagnosis using two types of serological tests is the usual approach. Serological tests are divided into two categories: non-treponemal and treponemal tests. The two test types have different advantages and disadvantages, and must be interpreted together, along with other clinical information to ensure accurate diagnosis and monitoring of response to treatment. Non-treponemal assays detect non-treponemal antibodies directed against cardiolipin, which is released by damaged host cells and from the treponemes themselves [98]. The most commonly used non-treponemal tests include the rapid plasma reagin (RPR) and venereal disease research laboratory (VDRL) tests. False positive reactions may occur in the presence of various medical conditions, including unrelated infections and autoimmune disorders [99]; test specificities are generally greater than 95% [100]. False negative results may occur very early after infection, as these tests typically become positive four to six weeks after infection [99]. The sensitivities of the RPR and VDRL tests are 86% and 78% for primary syphilis [99]. Nontreponemal test sensitivity is also low during late latent infection; approximately 20% of late latent cases will have non-reactive tests [99]. Test sensitivities for secondary and early latent syphilis are % [99]. Non-treponemal tests are widely used because they are inexpensive, simple to perform, and quantitative titres can be used to evaluate treatment response [100]. The test is quantified by evaluating serum dilutions. The presence of non-treponemal antibodies is indicative of an active syphilis infection, and changes in antibody titres can be used to monitor the course of infection and effectiveness of treatment [99]. Antibody titres usually correlate with disease activity, with high or rising titres predictive of a recent infection. Effective treatment results in a rapid titre decline and non-treponemal tests frequently become non-reactive some time after treatment. However, in some treated patients, antibodies persist at low but detectable levels. This is referred to as a serofast state. In the absence of treatment, a decline in antibody titres will still be observed, resulting in a low titre in the late latent and tertiary stages of disease [98]. A four-fold decrease in antibody titre following treatment is considered necessary to

28 16 indicate successful treatment, while a four-fold rise in titre is associated with re-infection or treatment failure. Treponemal-specific assays are based on antibody reactivity toward T. pallidum antigens. As these tests may remain active for a patient s lifetime, even with effective treatment, they are not used to evaluate response to therapy and cannot distinguish between active and past, treated infection [99]. Both treponemal antibodies and non-treponemal reagins will be detected in the blood of an individual with active infection, while those with a successfully treated past infection may still have a positive treponemal test with a negative or low titre non-treponemal test [101]. An advantage of treponemal tests is that they generally have much higher sensitivities for detecting primary syphilis than the non-treponemal tests [100]. Some of the commonly used assays include: microhemagglutination assay for T. pallidum (MHA-TP); T. pallidum particle agglutination test (TPPA); enzyme immunoassay (EIA); fluorescent treponemal antibody absorption (FTA-ABS) test; and chemiluminescence assay (CLIA). EIAs are increasingly used as an initial screening test due to high sensitivity and specificity, and their amenability to automation [22]. Serological testing is typically conducted in a laboratory setting, and it can take days to weeks to return test results to patients. As with other diseases where a timely diagnosis is desirable, point-of care (POC) tests have been developed for syphilis. These tests can be performed outside of a laboratory setting and provide results rapidly enough to affect immediate patient management, generally 20 minutes or less [102]. The availability of rapid results reduces the time between testing and treatment, during which transmission may be ongoing, and also reduces the potential for loss to follow-up. Most currently available tests use treponemal antigens, preventing discrimination between current and past infection, although dual treponemal and non-treponemal tests are in development [102]. Although these tests are widely available for use in low-income countries, none are currently approved for use in Canada [102]. A recent research study conducted in an outreach setting in Alberta, Canada, showed reasonable acceptance of testing (80% uptake) and diagnostic performance (85.3% sensitivity, 100% specificity) of a POC test, although the benefit of using the test was limited, as the targeted population had high prevalence of previous syphilis infection [103]. In high-income countries, current POC tests are likely best suited for use in hard-to-reach populations or in non-traditional

29 17 venues, and in populations with high syphilis incidence and low prevalence of previous treated infection [102] Testing Algorithms Different testing algorithms can be used to diagnose syphilis (Figure 1-5). Regardless of the approach used, both a non-treponemal and a treponemal test are used, but the order of the tests differs. The traditional approach is to screen using a non-treponemal test, followed by confirmatory testing of reactive samples using a treponemal assay. This approach has the advantage of avoiding the detection (and need for follow-up) of previously treated cases, unless a case has been re-infected or is in the serofast state. The major disadvantage of the traditional screening algorithm is the potential to miss cases with RPR-negative, untreated disease, which typically occurs in those with early primary or late latent infection. A reverse algorithm, using a treponemal test for screening, followed by a confirmatory non-treponemal test, is increasingly used, as treponemal tests can be run using automated systems, allowing for rapid processing of large numbers of samples and detection of patients with all stages of infection, including early infection [98]. As it is estimated that only 15-25% of patients who are treated during early syphilis will have a serologically non-reactive treponemal test 2 to 3 years post-infection [104, 105], confirmatory tests will be required in screened individuals with a previous history of syphilis. Compared to the traditional algorithms, the number of discrepant results between screening and confirmatory tests is expected to increase, requiring a second treponemal test to validate the screening test [104]. This approach may result in over-treatment and increased need for follow-up, particularly in low prevalence settings [106, 107]. Current Canadian laboratory testing guidelines recommend both the traditional and reverse screening algorithms, with choice guided by local disease epidemiology, test volumes, and available resources [101]. Additionally, if clinical assessment suggests possible or probable syphilis, particularly primary disease, treatment should be offered even if an initial nontreponemal test is negative, with a follow-up test repeated in two to four weeks [101].

30 18 Figure 1-5. A comparison of the traditional and reverse screening algorithms for the diagnosis of syphilis. Treatment is generally indicated when syphilis is considered likely, based on test results, and when an individual has not been treated previously. Previously treated individuals with likely syphilis and a four-fold or more increase in titre are recommended for retreatment.

31 19 Once a person is diagnosed with syphilis, additional testing is required if neurosyphilis is suspected (reviewed in [108]). Different algorithms are used for HIVuninfected and HIV-infected cases, as HIV-infected cases with low CD4 counts may present with different serological and CSF findings than immunocompetent hosts [108, 109]. Lumbar puncture is recommended for cases presenting with neurologic or ophthalmic manifestations, tertiary syphilis, or congenital syphilis, as well as patients who do not achieve an adequate response to treatment [14]. As discussed previously, HIV-infected individuals are also recommended to undergo lumbar puncture if CD4 cell count is less than 350 cells/ml, they have late latent syphilis, or an RPR titre of 1:32 dilutions [14]. The diagnosis of neurosyphilis is usually made based on a combination of reactive serologic results, abnormalities of CSF cell count or protein, and/or a reactive CSF-VDRL [14]. Following diagnosis and treatment of syphilis, non-treponemal test titres should be monitored until they are seronegative or reach a stable low titre [14]. The timing and frequency of follow-up testing depends on the stage at treatment and an individual s HIV status, with more frequent follow-up recommended for HIV-positive individuals [14]. Lumbar puncture should be repeated in patients with neurosyphilis at 6-month intervals until CSF parameters normalize [14]. 1.4 Syphilis Control Treatment Beginning with the first documented emergence of syphilis as an epidemic disease at the end of the 15 th century, mercury was used as a treatment of syphilis infection [110]. Mercury had been used to treat other skin diseases and appeared effective in treating syphilitic lesions. In addition to being used in ointments, mercury was administered internally and via inhalation of vapours [110]. Treatment would typically be administered for years, leading to the aphorism a night with Venus, and a lifetime with mercury [111]. Despite the notable adverse effects associated with treatment, including neuropathies, kidney failure, severe mouth ulcers and loss of teeth,

32 20 and death caused by poisoning, mercury remained the principle treatment for syphilis until 1909 [111]. That was the year Paul Ehrlich discovered an arsenic compound that successfully treated syphilis in rabbits; this discovery ultimately led to the manufacture of arsphenamine, which became know as the drug Salvarsan [111]. Although this drug was effective at curing syphilis, treatment was complex and produced toxic side effects. Despite these drawbacks, administration of arsenic in combination with bismuth or mercury became the primary syphilis treatment [111]. In 1917, the observation that febrile illness appeared to reduce the symptoms of neurosyphilis led to the development of malariotherapy, wherein patients were infected with malaria to induce a fever, with the malaria infection subsequently treated with quinine [111]. A turning point in the treatment of syphilis occurred in 1943, when penicillin was first shown to cure syphilis [112]. Penicillin remains the primary treatment to this day. Its impact was immediate and dramatic, with the decline in syphilis mortality coinciding with the introduction of penicillin [32]. As Douglas notes [113], the relationship between syphilis antimicrobial therapy is unique among the major infectious diseases of the early 20 th century, as no other infectious disease problem was reduced to such an extent without a major structural intervention, such as improved sanitation, or a vaccine. T. pallidum has a generation time of approximately 30 hours, making long-acting preparations, such as benzathine penicillin G, the preferred therapy [21]. Treatment is used to halt progression to tertiary syphilis. When administered during the infectious stages of infection, it also prevents transmission in the population. Primary, secondary, and early latent syphilis are treated with a single injection of benzathine penicillin G, while late latent syphilis requires 3 weekly injections [14]. Neurosyphilis is treated with intravenous penicillin G for 10 to 14 days [114]. Cure is based on serological response, defined as a four-fold decline in titres using a non-treponemal test at 6 months for infectious syphilis and one year for early latent syphilis [14]. Distinguishing between true treatment failures and re-infection is challenging, but it has been estimated that approximately 5-10% of cases require retreatment [115]. The most common adverse event associated with treatment is the Jarisch-Herxheimer reaction, an acute febrile reaction frequently accompanied by headache, myalgia, and other symptoms [22]. It

33 21 occurs most frequently in patients with early syphilis and usually resolves within 24 hours [22]. In penicillin allergic patients, macrolides and cephalosporins are effective alternative therapies. Although azithromycin is an effective alternative agent for treatment of early syphilis [116], macrolide-resistant syphilis strains have been identified in some regions [117], and azithromycin is not currently recommended for routine use [14] Public Health Approaches to Control Syphilis is a disease that should be eradicable: humans are the only host; it has an incubation period of several weeks, facilitating interruption of transmission with rapid prophylactic treatment of contacts; infectiousness is limited to less than one year in the absence of treatment; it is readily diagnosed; and its infectious stage is treatable with a single dose of antibiotics [60]. Despite these characteristics, syphilis control remains a challenge. In his role as Commissioner of Health for the State of New York, and subsequently as Surgeon General of the United States, Thomas Parran recognized the tremendous societal costs of syphilis and the importance of stigma and lack of public understanding as barriers to achieving STI prevention and control [118]. He developed the foundations for syphilis control in the 1930s, before the introduction of penicillin. Initially, the primary objective of syphilis control programs was the prevention of downstream sequelae of untreated infection [119]. Over time, as the burden of tertiary syphilis declined, programs focused on interrupting transmission [119]. The major elements of STI control programs as outlined by Parran remain essentially unchanged today: timely diagnosis and treatment, screening for asymptomatic or undetected infection, partner notification, and public education and engagement [118, 120]. Although adapted to address the unique characteristics of different pathogens, these components continue to guide control efforts for many STIs and other communicable diseases to this day [118].

34 22 Individuals may be diagnosed with syphilis by a number of mechanisms. Those with symptoms may directly seek medical care and undergo testing and subsequent treatment. Partners of diagnosed cases may be identified through the contact tracing and partner notification process. Suggested trace back periods for identifying at-risk contacts depend on the stage at which an index case is diagnosed, ranging from 3 months for primary syphilis to 1 year for early latent syphilis [14]. Partners of syphilis cases who can be traced are encouraged to seek testing and receive treatment if found to be infected. Novel approaches to partner notification have become necessary with the rise in popularity of using the Internet and mobile apps to meet sex partners [121], including posting notifications in online chat rooms [122] and using online social networking platforms to reach out to partners of index cases [123]. Individuals may also be diagnosed through routine STI screening services provided through medical care providers, including at sexual health clinics. Canadian guidelines for syphilis screening vary, depending on an individual s perceived risk. It is recommended that persons at ongoing risk of syphilis infection undergo screening at 3- month intervals, and all MSM with an STI risk, regardless of HIV status, are advised to be screened at least annually [14]. Venue-based screening blitzes have been implemented as a way to screen individuals who may not be engaged in routine medical care [124]. These short-term programs provide testing services at venues frequented by at-risk populations, with the aim of increasing screening uptake and reaching those who have not previously been tested. Mass treatment of at-risk populations has been used as a syphilis control measure, although the effectiveness of such an approach appears to be limited to short-term reductions in infection incidence and it may ultimately lead to a rebound in incidence rates to levels as high or higher than rates observed prior to the intervention [125, 126]. Presumptive treatment of individuals for syphilis, as has been reported previously [126], will prevent reporting of cases and downstream contract tracing activities, which may create a challenge for disease surveillance programs.

35 Immunity Immune Response to Infection Strong humoral and cell-mediated immune responses are induced early in the course of infection. The lesions of primary and secondary syphilis are initially infiltrated primarily by T lymphocytes, followed by macrophages [62]. Resolution of the symptoms associated with the primary and secondary stages of infection correlates with the development of these cellular immune responses [19]. Although most treponemes are cleared from the lesions, some are able to persist [62]. The humoral immune response results in the production of antibodies required for opsonization and complementmediated immobilization or neutralization [24, 127, 128]. Circulating antibodies to T. pallidum are detected early in primary infection, reach high titres as infection disseminates during the secondary stage, and remain detectable throughout the course of infection [19, 114, 129]. Despite this host response to infection, T. pallidum can survive and replicate, leading to the establishment of latent infection, and in some cases, the development of tertiary disease. Given that humans develop a robust immune response, how is it that: (i) individuals can be infected multiples times and (ii) T. pallidum can persist in the body, causing tertiary disease? Antigenic diversity and variation may be key for T. pallidum s ability to evade the immune detection [24]. Experimental evidence suggests that antigenic variation of bacterial surface proteins may allow T. pallidum to evade the acquired immune response [24]. T. pallidum has few integral outer membrane proteins [130, 131]; among these, T. pallidum repeat K (TprK) is the best studied [24]. Strong antibody and T-cell immune responses are elicited against TprK, and in the rabbit model, immunization with recombinant TprK provides partial protection against infectious challenge [ ]. TprK sequences differ substantially not just between strains, but within individual strains [136]. Diversity is localized to 7 discrete variable regions of the protein, which are predicted to be surface exposed [24]. New variants are produced by segmental gene conversion, with segments of new sequence at donor sites located elsewhere on the chromosome replacing portions of the 7 variable regions in the tprk

36 24 open reading frame [137]. In the rabbit model, sequence diversity accumulates after acquired immunity develops [138, 139]. The resulting changes in the exposed variable regions of TprK allow the variant treponemes to evade antibody binding and opsonophagocytosis, thereby persisting during chronic latent infection [24] Protective Immunity with Untreated Infection Patients with untreated syphilis appear to experience protective immunity to repeated infection. Colles law derives from the observation of a 19th century physician that wet nurses who breast-fed infants with congenital syphilis often developed chancres of the nipple, while the mothers of infected infants did not, suggesting that they were protected from repeated infection [19]. Prisoners with untreated syphilis infection who were experimentally challenged with T. pallidum displayed protection from developing chancres at the site of cutaneous inoculation [140]. A similar phenomenon has also been observed in rabbit models; immunization of rabbits with irradiated T. pallidum has been shown to induce protective immunity to repeat infection [141] Evidence of Protective Immunity Following Treatment Data supporting the existence of protective immunity to re-infection among patients successfully treated for syphilis infection is less clear. Repeat infections among previously treated individuals are common [81, ]. However, there are experimental data suggesting there is a degree of transient acquired immunity among those treated for latent syphilis (Table 1-1). An unethical experimental challenge study conducted in prisoners in 1953 showed that acquired immunity appears to be stagedependent [4, 140]. Although not all participants with previously treated late latent infection were protected from re-infection (as determined by chancre development at the site of inoculation), symptoms of re-infection were less severe; of 10 cases with prior late latent infection who were re-infected, only one had a dark-field positive lesion [140]. By contrast, all individuals with treated early syphilis displayed some signs of re-infection following experimental challenge.

37 25 Table 1-1. Results of T. pallidum challenge experiment in cases with different histories of syphilis infection and treatment. Status of Cases Prior to Challenge a Number Inoculated % (Number) with Dark-field Positive Lesions and Increased Titre % (Number) with Dark-Field Negative Lesions and Increased Titre % (Number) Uninfected Never infected Untreated late latent syphilis Treated early syphilis Treated late latent syphilis (8) 0 (0) 0 (0) 5 0 (0) 0 (0) 100 (5) (9) 18 (2) 0 (0) 26 4 (1) 35 (9) 61 (16) Table adapted from: Magnuson et al. [140]. a Late latent syphilis refers to syphilis of more than two years duration

38 26 It should be noted that study participants were challenged with an experimental dose much larger than that typically experienced during natural infection, at 2000 times the infectious dose [140]. Although the study was not designed to examine the impact of time since treatment and susceptibility to re-infection, the authors note a trend suggestive of a relationship between longer interval after treatment and increased likelihood of reinfection [140]. 1.6 Syphilis Resurgence Historical Context With the introduction of penicillin in the 1940s, and in countries with adequate health care access and public health infrastructure, syphilis has transitioned from a highmorbidity pandemic to a low-level, largely concentrated endemic infection, which displays periodic outbreaks [113]. The resurgence of syphilis that has recently been observed in MSM is not a novel phenomenon. In 1965, Cannefax noted that the belief that the discovery of a one shot treatment would lead to the elimination of syphilis as a public health problem had not been borne out, and that beginning in 1957 there had been an international resurgence of syphilis morbidity [146]. An analysis of long-term syphilis trends in the United States has shown that rates tend to oscillate with an approximately 10-year periodicity [73] Proposed Explanations for Syphilis Resurgence Two major mechanisms for periodic increases in syphilis incidence have been proposed: endogenous oscillations predicted by the natural dynamics of an infection to which there is partial protective immunity [73], or sexual behaviour changes driving changes in infection risk [147]. The argument that partial protective immunity contributes to cycles in syphilis incidence is controversial and is based on an analysis of long-term trends in syphilis and gonorrhea in the United States [73]. The authors showed that gonorrhea, which lacks

39 27 temporary immunity, fails to display the same cycles in incidence that are observed for syphilis, and used mathematical modeling to show how the endogenous characteristics of syphilis could cause these observed long-term cycles. These cycles occurred without the need for invoking changes in population behaviour or demographics. A more detailed description of this model is presented in the mathematical modeling section below. It is possible to induce rapid declines followed by increases in syphilis incidence by introducing one-time broad coverage mass treatment programs into populations with a large pool of prevalent infections [126, 148]. Under these circumstances, there is no need to invoke the existence of treatment-induced interruption of protective immunity to produce the observed dynamics. The treatment program has the effect of treating incubating and infectious cases of syphilis, resulting in a transient decrease in infection transmission and incidence. However, mass treatment also returns individuals with latent infection, who were no longer infectious to others and presumably protected from reinfection, to the pool of susceptible individuals. In the presence of ongoing syphilis transmission in the population, these individuals are now at risk for re-infection, driving an increase in incidence. The existence of a degree of transient immunity following treatment affects how rapidly and how large the rebound effect will be, but is not a determinant of its occurrence [149]. However, in a population where screening is ongoing and large shifts in screening practice are not observed, it is necessary to include a period of protective immunity following treatment of latent infection to generate observed periodic resurgences in incidence, if attempting to describe these dynamics in the absence of other exogenous factors [73]. Over time, as immunity wanes in treated individuals, the pool of susceptible individuals reaches a critical size, allowing for sustained transmission and syphilis resurgence. If this period of protective immunity did not exist, treated individuals would become immediately susceptible to re-infection, with this flow of previously infected individuals leading to the establishment of stable infection incidence, preventing the boom-and-bust dynamics observed with syphilis [150]. Others have proposed that the observed changes in syphilis epidemiology are explained by changes in high-risk sexual behaviour [60, 147]. Although it is clear that

40 28 engaging in certain sexual behaviours increases an individual s syphilis acquisition risk, demonstrating that changes in sexual behaviour at the population level are responsible for periodic syphilis epidemics is more challenging. Supporting this argument is the observation that resurgences of syphilis in the United States have tended to occur in specific subpopulations; within these subgroups, the periodicity of syphilis resurgences is heterogeneous [147]. A recent attempt to analyze syphilis trends in MSM in the United States has shown that changes in syphilis diagnoses in MSM correlate temporally with events that led to changes in sexual behaviour, including the Stonewall riots, the onset of the HIV epidemic, and the introduction of highly active antiretroviral therapy (Figure 1-6) [87]. It has been proposed that increases in AIDS-associated mortality may have contributed to the decline in syphilis incidence in men that occurred in the early 1990s in the United States [151], and that current syphilis resurgence in MSM would not have occurred had AIDS mortality rates remained at peak levels [152]. Increasing rates of certain high risk sexual behaviours in communities of MSM have been noted since the late 1990s, possibly contributed to by improvements in HIV treatment, demographic changes in MSM populations, including a growing proportion of MSM with prevalent HIV infection, and technology-driven expansions of social and sexual networks that facilitate rapid partner change [74, ] A recent mathematical modeling study has suggested that adaptive changes in sexual behaviour alone can lead to periodic resurgences in syphilis infections [161], although another modeling study examining the cycling phenomenon reached the conclusion that a combination of temporary immunity and changes in sexual behaviour are likely responsible for recurrent syphilis epidemics [162]. These studies are discussed in greater detail in the mathematical modeling section below

41 29 Primary and secondary syphilis cases per 100,000 population Stonewall riots: June 1969 First report of AIDS: June 1981 HAART: July Year Figure 1-6. Estimated rates of primary and secondary syphilis in men who have sex with men. Rates of primary and secondary syphilis in MSM are estimated from national data for the United States, Adapted from Peterman et al. [87].

42 Current Approaches to Syphilis Control in MSM The ongoing syphilis epidemic has led to the implementation of novel approaches to increase screening in MSM. Clinic-based programs to increase screening have been implemented in some jurisdictions (reviewed in [163]). Most of these programs focus on incorporating screening into the routine medical care of HIV-infected MSM who are already engaged in care, thereby minimizing the need for men to make additional visits to their health care providers [ ]. An evaluation of different approaches to enhanced screening in two Australian cities found that opt-out or opt-in strategies for all HIV-positive MSM appeared to be more effective than risk-based screening (with high risk individuals identified as those reporting multiple partners, attending sex-on-premises venues, or having an STI on previous testing) [164]. Although opt-out screening programs have demonstrated increases in syphilis testing in HIV-infected MSM, and a corresponding increase in the number of asymptomatic syphilis infections detected, screening did not reach the targeted 3- monthly frequency [ ]. This likely relates to the fact that HIV viral load testing is occurring in many men less frequently than every 3 months. Another program was able to increase STI screening rates in HIV-positive MSM by coupling it with an anal cytological screening program [168]. Other approaches to enhance screening uptake include reminders to health care providers or men considered at risk of syphilis infection. Use of a computer alert in a sexual health clinic reminding clinicians to test men at high risk of syphilis (MSM reporting 10 or more partners in the past 12 months) increased screening rates and detection of asymptomatic syphilis [169]. Sending automated test reminders via text or to at-risk MSM has been found to increase screening uptake [170, 171]. Implementation of guidelines specifying annual testing of men reporting at least one male sexual partner in the previous year was found to increase syphilis screening rates in a sexual health clinic in Sydney, Australia [172]. Although all of these programs using reminders achieved improvements in screening uptake, the evaluation periods were of limited duration (9 to 12 months), and the average interval between tests was not reported, such that it is not possible to conclude if such approaches result in sustained

43 31 increases in testing at the frequency required to interrupt syphilis transmission in the broader population of MSM. A limitation of clinic-based programs is that they exclude MSM not engaged in medical care. We do not know what proportion of the MSM population has regular contact with the health care system or how MSM engaging in care differ from those who do not. Although Canada-wide estimates are lacking, recent estimates from British Columbia suggest that only 57% of HIV-infected individuals have been diagnosed with infection and are retained in medical care [173], although this percentage may be higher among MSM [174]. In the United States, it is estimated that 40% of HIV-positive individuals are engaged in care [175]. These numbers suggest that clinic-based interventions will only reach a proportion of at-risk men. However, given the apparent importance of treatment and subsequent re-infection in sustaining syphilis transmission, it may make sense, from a disease transmission perspective, to target men already engaged with the health care system for enhanced screening. Treatment is making syphilis-infected men vulnerable to re-infection, so it is important to ensure that they are being screened regularly to prevent an ongoing cycle of re-infection. The yield of screening programs in non-medical settings, such as MSM-oriented venues, has typically been low [124, 176], although other potential benefits, such as increasing awareness of the need to be screened for syphilis, cannot be ignored. 1.7 Mathematical Models of Syphilis Types of Mathematical Models A model is a simplified representation of a complex system, the goal of which is to explain or predict features associated with the system, in order to gain insight into the underlying processes that govern the system [177]. Mathematical models use mathematical equations to represent complex systems, providing a quantitative language to describe the relation between variables and changes in state [178]. Mathematical models of infectious diseases use equations or set of rules to describe the spread of infection or the development of disease [178]. Unlike statistical models, which describe

44 32 associations between variables, mathematical models are based on a theoretical framework that represents causal pathways and mechanisms linking exposures, interventions, and infection or disease [178]. Mathematical models of infectious diseases can provide a rational framework for understanding disease dynamics and delivering insights into the potential impacts of disease control interventions, both on individuals who undergo such interventions and also on the wider community. Mathematical models are increasingly used in public health to assist policy decisions, guide resource allocation, and inform strategic planning for prevention of infectious diseases [178, 179]. Models are not a substitute for randomized controlled trials (RCT) or other population-based studies but they can be useful tools for synthesizing evidence and evaluating health policy when RCT-based evidence is absent or insufficient, or when conducting a trial is unfeasible or unethical [179]. They also facilitate the evaluation of indirect health and economic outcomes accruing to individuals not directly receiving the intervention of interest [180]. This incorporation of indirect effects of interventions, particularly the downstream prevention of cases following treatment of a single individual, is useful for aiding the interpretation of surveillance data and determining the cost-effectiveness of different strategies [180]. Modeling also provides a means for the estimation of the impact of a program, by allowing us to compare what might have happened in the absence of such a program (the counterfactual) [178]. Mathematical models of infectious diseases can be classified based on some key characteristics [178]. A primary distinguishing feature is whether a model is static or dynamic. In a static or linear model, infection incidence is described by a constant or linear function. Linear models are frequently used in health economic analyses [180]. Static models ignore indirect effects, which can lead to under- or overestimations of the effects of interventions [178]. A dynamic or non-linear model expresses infection transmission as a function of prevalence of infectious individuals, number of susceptible individuals, and the risk of infection. Changes in infection incidence (force of infection) over time reflect both positive and negative feedbacks that can result from changes in infection prevalence, immunity, or mortality [177]. Dynamic models are generally the

45 33 most appropriate choice for modeling communicable diseases, as these feedbacks are often key for capturing infection dynamics in populations and for understanding unexpected consequences of interventions that may arise when implementing changes in a complex dynamic system [180]. The role of chance determines whether a model is deterministic or stochastic. In deterministic models, for a given set of parameter values and initial conditions, every realization is the same. Chance is not incorporated into the model, and outputs approximate the average behaviour of a system [178]. Deterministic models are generally less computationally intensive and easier to analyze than stochastic models [178]. However, in systems where chance is important for determining a system s behaviour, such as examining infection emergence, infection spread in small populations, or rare diseases, deterministic models may not be appropriate [178]. Stochastic models incorporate chance. Events, such as infection transmission, are assumed to occur with a certain probability, such that different results are generated each time a model is run with a given set of parameters [178]. To account for stochastic variation, these models are typically run over multiple simulations to generate estimates of average model behaviour and variation around the average [177]. Models may be population based or individual based. Population-based models divide individuals into groups or compartments and track changes happening to these compartments. Model outputs represent population-level processes and trends, without specifying which individuals are involved [178]. It is possible to incorporate population heterogeneity, such as age or sex, by dividing the population into additional subgroups [178]. By contrast, individual-based (also referred to as agent-based or microsimulation) models explicitly model and track individuals over time [178]. Events, health states, treatment histories, and behaviours are associated with an individual and connections between individuals are modeled. Although not all stochastic models are individualbased, all individual-based models are stochastic [178]. An advantage of individualbased models is that they allow for the explicit representation of social and sexual networks. Individual-based models tend to be computationally intensive and difficult to analyze, and require substantial data to parameterize [178].

46 Key Concepts in Mathematical Modeling of Infectious Diseases The first documented use of a mathematical approach to quantify the risks and benefits of a public health intervention to control a communicable disease was by Daniel Bernoulli in 1760, when he formulated a mathematical model to support the adoption of universal variolation with smallpox [181]. Infectious disease modeling as a field began to be established in the early 20 th century, as mathematical approaches were applied to a range of disease and questions (reviewed in [182]). Early mathematical models were formulated to understand disease recurrence and host-vector disease dynamics. Beginning in 1927, Kermack and McKendrick published epidemic models demonstrating that the density of susceptible individuals in a population must exceed a critical value for an outbreak to occur [ ]. These models are the source of the commonly used Susceptible-Infectious-Recovered (SIR)-type deterministic compartmental models. Since that time, the number and types of mathematical models used to study a range of infectious diseases has grown tremendously. Some key concepts in mathematical modeling and the insights derived from mathematical models will be reviewed below. The basic reproduction number (R0) describes the average number of secondary infections generated by a single infectious case in a completely susceptible population [186]. The value of R0 determines whether a communicable disease will spread in a population. When R0 is less than 1, each infectious case will generate less than one new infection and the disease will eventually die out. When R0 is greater than 1, each case will produce more than one new infection and the disease has the potential to spread through the population. Early in an outbreak, in an infection-naïve population and in the absence of any interventions, the number of new cases will grow exponentially. When R0 is 1, each case produces one new case and an infection is endemic in the population. In a homogenously mixing population, R0 can be calculated as: R 0 = βcd where β is the probability of transmission per contact, c is the number of contacts per unit of time, and d is the duration of infectiousness.

47 35 Many infectious disease control interventions act on one of the components of R0 to reduce transmission. In the case of syphilis, treatment acts to reduce the duration of infectiousness, while condom use reduces the probability of transmission per sexual contact. Advising cases to abstain from sex upon diagnosis of an STI aims to reduce the number of sexual contacts an infectious individual has. R0 is a unitless measure of the potential for an infectious disease to spread in a population. Another important characteristic of communicable diseases is the time between infection of a primary case and a secondary case. Typically, the serial interval refers to the time from symptom onset in a primary case and symptom onset in his or her secondary case, while generation time refers to the time from infection of an initial case to the infection of his or her secondary case [187]. Thus, serial interval is observable while generation time is generally not. Two diseases with similar values of R0 may have very different generation times, resulting in different epidemic time courses. For example, both measles and tuberculosis have estimated R0 values in the range of 10-15, but the serial interval for measles is measured in days or weeks, while that of tuberculosis is measured in years [188]. Consequently, epidemic dynamics are quite different for these two pathogens. The doubling time for the number of cases in a growing outbreak can be calculated as: ln( 2)T g R 0 1 ( ) where Tg is the generation time. In the pre-vaccination era, biennial measles outbreaks were the norm [189], while it has been projected that it takes one to several hundred years for a tuberculosis outbreak to rise, fall, and reach a stable endemic level [190]. Although R0 and the generation time facilitate estimation of whether and how quickly a communicable disease will spread in a population early in an outbreak, these values alone do not determine whether an infectious disease will be controllable (assuming the appropriate interventions exist). Another contributing factor is the proportion of transmission that occurs prior to symptom onset or asymptomatically [191]. For diseases where transmission can occur in the absence of symptoms, control can be

48 36 challenging, even if R0 is small, as is the case for HIV [191], since most transmission occurs in asymptomatic individuals. Conversely, although small pox has a high estimated R0, asymptomatic transmission is rare, such that infectious cases can be isolated, once infection is recognized, to prevent ongoing transmission [191]. The absence of asymptomatic transmission was likely critical for smallpox eradication. Similar approaches have been applied to the West African Ebola virus epidemic, as infection is believed to be transmitted overwhelmingly by symptomatic (and gravely ill) individuals [192]. As mentioned above, Kermack and McKendrick first articulated the idea that susceptible individuals are critical for infection transmission. The effective reproduction number (Re) captures this phenomenon. Re is calculated as: R e =R 0 S where S represents the proportion of the population that is susceptible at a given point time. In the context of the introduction of a novel infectious disease into a completely susceptible population (i.e., S=1 initially), when R0 is greater than 1, infection spreads, the number of infectious cases grows (determined initially by R0) and the number of susceptible individuals decreases. Consequently, the effective reproduction number declines over time. When Re equals 1, the outbreak begins to decline, as each infectious case generates less than one new case. Once the outbreak ends, a new outbreak will not occur in this population until the proportion of the population that is susceptible increases to the point that the effective reproduction number has increased above 1, either through loss of immunity in previously infected individuals or entry of new individuals into the population through birth or immigration. Thus, although the introduction of an infectious case has the potential to spark an outbreak, having sufficient population susceptibility is the ultimate determinant of whether the outbreak will occur. When an infectious disease is endemic in a population, Re is equal to 1. The fraction of the population susceptible at the endemic steady state (S * ) is therefore: S * = 1 R 0

49 37 This relation can be used to determine the proportion of the population that must be immune, either through immunity or vaccination, to prevent infection spread [186]. Some important insights from this relation are that even in the absence of intervention, infection transmission will stop without infecting the entire population, and that a control intervention such as vaccination does not need to achieve 100% coverage to interrupt infection transmission; so long as the proportion of susceptible individuals is below 1/R0, an outbreak is unlikely to occur. It is important to note that the basic reproduction number is not a static value. It represents an average value, and can change over time, from population to population, and even from individual to individual [193]. In the case of sexually transmitted infections, heterogeneity in risk is particularly important, and the relation presented above for population susceptibility at endemic steady state is an oversimplification. For STIs, general population prevalence tends to equilibrate at levels lower than that predicted based on R0, while prevalence in groups at high risk of acquiring and transmitting infection is higher [193]. Similarly, patterns of sexual mixing - specifically, the likelihood that individuals with high numbers of sexual partners have sex with others who also have high numbers of sexual partners influence infection spread and prevalence. Hethcote and Yorke explored the importance of variability in behaviour for STI persistence [194], specifically applied to gonorrhea. However, many of the insights provided by this modeling work can be extended to other STIs. Using a simple deterministic compartmental model of gonorrhea transmission in the general population and assuming reasonable values of partner change (2 partners/year), transmission probability per partnership (0.75), and duration of infection in the presence of treatment (0.167 years), the value of R0 in this example would be 0.25, and consequently, gonorrhea would not be expected to spread in the population (Figure 1-7). To achieve sustained transmission of this STI using this model structure, the average rate of partner change in the entire sexually active population would have to be 8 partners per year [186]. The failure to accurately model infection transmission relates not to using unrealistic parameter estimates, but rather to ignoring heterogeneity.

50 38 Figure 1-7. Example of the importance of heterogeneity in models of sexually transmitted infections. The expected prevalence of gonorrhea in a population with (A) no heterogeneity in rates of sexual partner acquisition, and (B) high and low rates of partner change. Both models assume a mean rate of partner change of 2, transmissibility of 0.75 per partner, and duration of infectiousness of years. In (B) 2% of the population is in the high activity group and there is proportionate mixing between groups. Example is adapted from Vynnycky and White [186] and based on work by Hethcote and Yorke [194].

51 39 Sexual behaviour data demonstrate that over a given period of time, most individuals report relatively low number of sexual partners, while fewer individuals report higher numbers of partners. For example, a survey of the sexual activity patterns of the heterosexual population in Quebec found that males reported a mean of 11 lifetime sexual partners, but the variance associated with this estimate was 163 [195]. For females, the reported mean and variance were 6 and 72, respectively [195]. Both the large variance in partner numbers and the discrepancies in reported values between males and females have been observed in other high-income countries [196, 197]. A rightskewed distribution in reported partner numbers is also reported in surveys of MSM [76, 198]. It is this heterogeneity that is important for the spread and persistence of STIs. One way to capture this heterogeneity is through the effective partner change rate (ĉ), where ĉ describes the epidemiologically relevant average number of sexual partners [186, 199]. When variance is large, this value may be much larger than the arithmetic mean. The effective partner change rate can be approximated as: ĉ =c + σ2 c where c is the arithmetic mean and σ 2 is the variance [199]. The simple model of gonorrhea described above can be modified to allow for two groups, one with a low average rate of partner change, which contains 98% of the population, and the other with a high average rate of partner change, which contains 2% of the population and represents a core group [186]. If the overall average rate of partner change is held constant (2 partners/year), and the partner change rate is assumed to be 1.4 per year in the low activity group, this results in a rate of partner change in the high activity group of 31.4 per year and a variance of 17.6 [186]. This gives an effective partner change rate of 10.8, and an estimated value of R0 of R0 is now greater than 1 and gonorrhea is able to spread and persist in the population (Figure 1-7). Prevalence in the high activity group is much higher than that in the overall population. In addition to heterogeneity in population sexual behaviour, mixing patterns may also influence transmission dynamics of STIs. Sexual mixing describes the types of

52 40 sexual contact patterns that occur within and across population subgroups. Unlike rates of partner acquisition, quantifying mixing patterns empirically is less straightforward [200]. The two major types of mixing typically considered in models of STIs are: proportionate mixing and assortative mixing [186]. In proportionate mixing, sexual partners are chosen based on the number of potential partnerships supplied by individuals in a given sexual activity group [201]. Note that mixing is based on the number of partnerships in a given group, not the number of individuals in a group [186]. This type of mixing may also be called random mixing. Assortative mixing refers to individuals preferentially forming partnerships with individuals in the same population subgroup [186]. For example, individuals with high rates of partner change may preferentially choose partners with high rates of partner change. Disassortative mixing is another form of mixing, with individuals preferentially forming partnerships with those who have characteristics unlike their own [186]. In fact, heterosexual partnerships could be considered a type of disassortative mixing. Groups, and mixing between them, can be defined by variables other than rates of partner acquisition, including gender, age, and socio-economic status. The gonorrhea model described above assumes proportionate mixing, but empirical studies suggest weakly assortative mixing by sexual behaviour is more typical [202]. More assortative mixing results in less widespread but more persistent infection transmission, whereas disassortative mixing results in slower initial spread but more widespread transmission [193]. Network models allow for more explicit representations of sexual partnerships than can be described using compartmental models [186]. Networks of contacts between individuals are represented by graphs with nodes representing individuals and edges joining nodes, representing contacts between individuals. In the case of STIs, edges represent sexual partnerships that have the potential to lead to the transmission of infection. Networks can range from very simple to highly complex and realistic, and may be measured empirically or represented by idealized networks. Data for parameterization of

53 41 network models of sexually transmitted infections are often difficult to obtain. Empirical networks are constructed using social network analysis techniques, with data collected by infection tracing, contact tracing, or diary-based studies [203]. Idealized networks are computer-generated networks that are defined by how individuals are distributed in space and how connections are formed (reviewed in [203]). The type of idealized network used to study infectious disease transmission will depend on the characteristics of the pathogen and the population under study. In random networks, individuals are connected randomly, regardless of spatial or other defining characteristics, representing homogenous behaviour and similar to proportionate mixing in a compartmental model. The initial growth rate and number of infected individuals is lower in a random network than predicted in a compartmental model, as a result of saturation of local partners to infect [186]. In a lattice network, individuals are placed on a regular grid in space and linked to neighbouring individuals, resulting in a highly clustered network. This results in lower rates of infection transmission than in random networks [186]. When a small number of long-range partnerships are added to a lattice model, a small world network is created. Infection is able to reach all parts of a network quickly, resulting in more rapid spread than in a lattice network. This type of network is thought to most accurately model human social networks [203]. Human sexual behaviour data seem to be best described by scale-free networks [186]. Scale-free networks rely on preferential attachment, whereby connections are preferentially made with individuals with large numbers of pre-existing contacts, resulting in a power-law distribution in the number of contacts per individual [203]. In other words, most individuals in the network have few partners, while a few individuals have many partners [186]. This heterogeneity results in the emergence of core groups of highly connected individuals in these types of networks. Interestingly, in a true scale-free network where partner numbers follow a power law distribution, variance tends to infinity, and consequently, so does R0. Under these circumstances, the only way to eliminate an STI would be via behavioural interventions that modify network structure [186]. In reality, there is expected to be a finite upper limit to the distribution of sexual partner number. Finally, individual-based simulation network models explicitly describe the entire network of individuals and their partnerships [186]. Specific partnership formation and dissolution rules are applied and

54 42 typically modeled as stochastic processes. These networks tend to be more realistic and intuitive but are also generally computationally expensive and not analytically tractable [186]. All of the different types of network models described above can be used to understand how network structure influences both the transmission of different pathogens and the potential impact of control measures. One key insight from scale-free network models is that strategic use of vaccination, through the targeting of highly-connected nodes, can result in interruption of infection transmission at much lower thresholds than would be predicted assuming a homogenously mixing population [203] Review of Existing Mathematical Models of Syphilis The above section discussed general concepts that are important for mathematical modeling of infectious diseases and some key findings that shape how we think about communicable disease transmission and control, particularly as applied to sexually transmitted infections. The following sections focus on mathematical models of syphilis. Existing modeling studies have provided important qualitative and quantitative insights into infection dynamics and the impact of control measures. An overview of mathematical models of syphilis is presented below, with the general characteristics of each model summarized in Table 1-2. This review excludes models that focused primarily on congenital syphilis. A number of modeling studies have been conducted in which syphilis is modeled as a biological cofactor for HIV transmission, with the aim of quantifying syphilisattributable HIV burden [ ] or the impact of interventions to control syphilis transmission on HIV incidence [ ]. A variety of approaches have been used to model syphilis transmission to address these questions, ranging from Bernoulli models of HIV transmission that include syphilis as a modifying risk factor [207], to agent-based models that incorporate the natural history of HIV and syphilis [212]. However, as these studies did not explicitly evaluate syphilis dynamics or the impact of interventions on syphilis burden, they have been excluded from this review.

55 43 Table 1-2. Mathematical modeling studies examining syphilis natural history or control. Model Characteristics Study Setting Population Purpose Feedback in System Role of Chance Representation of Individuals Althouse 2014 [161] USA Network of sexually active individuals Understand disease dynamics and intervention evaluation Non-linear Deterministic and stochastic Individual and population Dobay 2013 [218] Switzerland General population Intervention evaluation Non-linear Deterministic Population Mitchell 2013 [219] Down 2012 [220] Gray 2011 [221] China FSW and clients Intervention evaluation Non-linear Deterministic Population Australia Gay men Intervention evaluation Non-linear Stochastic Individual Australia Gay men Intervention evaluation Non-linear Stochastic Individual Owusu-Edusei 2011 a [107] USA General population low syphilis prevalence Cost-effectiveness analysis Linear Deterministic Population Owusu-Edusei 2011 [106] USA General population low syphilis prevalence Cost-effectiveness analysis Linear Deterministic Population

56 44 Model Characteristics Study Setting Population Purpose Feedback in System Role of Chance Representation of Individuals Johnson 2011 [222] South Africa General population Intervention evaluation Non-linear Deterministic Population Wilson 2011 [223] Gray 2010 [224] Australia Gay men Intervention evaluation Non-linear Stochastic Individual Australia Gay men Intervention evaluation Non-linear Stochastic Individual Johnson 2010 [225] South Africa General population Understand disease dynamics and intervention evaluation Non-linear Deterministic Population Wilson 2010 [226] Australia Male clients of FSW Cost-effectiveness analysis Linear Deterministic Population Litvak-Hinenzon 2009 [162] Generic Small-world network Understand disease dynamics and intervention evaluation Non-linear Deterministic and stochastic Individual and population Tuli 2009 [227] USA Incarcerated MSM Cost-effectiveness analysis Non-linear Deterministic Population Breban 2008 [147] USA Core group of sexually active individuals Understand disease dynamics Non-linear Stochastic Individual Chuck 2008 b [228] Canada General population Cost-effectiveness analysis Linear Deterministic Population

57 45 Model Characteristics Study Setting Population Purpose Feedback in System Role of Chance Representation of Individuals Kourbatova 2008 [229] Russia Occupational groups undergoing mandatory STI screening Cost-effectiveness analysis Linear Deterministic Population Volz 2007 [230] USA High school students Understand disease dynamics Non-linear Deterministic Population Grassly 2005 [73] USA Core group of sexually active individuals Understand disease dynamics Non-linear Stochastic and deterministic Population Blandford 2003 [231] USA Sex partners exposed to early syphilis Cost-effectiveness analysis Linear Deterministic Population Pourbohloul 2003 [149] Canada General population Intervention evaluation Non-linear Deterministic Population Korenromp 2002 [232] SSA City with advanced HIV epidemic Intervention evaluation Non-linear Stochastic Individual Marseille 2001 [233] South Africa FSW and clients Cost-effectiveness analysis Linear Deterministic Population Korenromp 2000 [148] SSA Rural population Intervention evaluation Non-linear Stochastic Individual

58 46 Model Characteristics Study Setting Population Purpose Feedback in System Role of Chance Representation of Individuals Garnett 1997 [4] Generic General population Understand disease dynamics and intervention evaluation Non-linear Deterministic Population Cates 1996 [119] USA General population Understand disease dynamics Non-linear Deterministic Population Oxman 1996 [234] USA Heterosexual population Intervention evaluation Non-linear Deterministic Population Abbreviations: FSW, female sex workers; MSM, men who have sex with men; SSA, sub-saharan Africa; STI, sexually transmitted infection a Study also included an analysis for a population with high syphilis prevalence in sub-saharan Africa. As the focus of that analysis was screening pregnant women to prevent adverse birth outcomes, it was excluded from the review. b Study also evaluated the cost-effectiveness of screening in a pre-natal population, which was outside of the scope of this review.

59 47 Mathematical models of syphilis can be broadly characterized as those developed to understand disease natural history and those whose primary aim is to evaluate the economic and/or health impact of interventions for syphilis control. Each of these applications, and the insights gained, will be discussed in turn Models Examining Syphilis Natural History and Dynamics Several studies have used mathematical models of varying complexity to understand the transmission of syphilis in populations [4, 73, 119, 147, 162, 225, 230, 234]. A simple model made use of the components of the basic reproductive number to understand the basis of syphilis control programs [119]. Cates et al. [119] used historical data to estimate an R0 of 1.15 for syphilis. Oxman et al. [234] showed the importance of core groups with high rates of sexual partner change for sustaining syphilis transmission in a population. This study also investigated how assumptions of immunity to reinfection (either biological or behavioural) following treatment impact infection transmission, with protection from re-infection leading to attenuation of transmission in the short-term [234]. Garnett et al. [4] used a deterministic compartmental model to demonstrate that syphilis incidence in populations is expected to show damped oscillations in the absence of treatment, mimicking the expected dynamics of a Susceptible-Infected-Removed - type system where immunity is lifelong. With the introduction of treatment, the absence of any post-treatment immunity would result in more stable infection dynamics, similar to what is observed with gonorrhea, whereas the presence of immunity would make periodic epidemics more likely [4]. Johnson et al. [225] used Bayesian methods to make inferences about syphilis transmission and natural history parameters and showed that commonly used parameter values were consistent with those estimated through model fitting. Recognizing the importance of the dynamic nature of sexual contacts in syphilis transmission, Votz and Meyers [230] applied a novel mathematical modeling approach incorporating dynamic contact patterns to a high school syphilis outbreak. The model was able to incorporate contact tracing data, although the relatively small size of the

60 48 dataset and associated parameter uncertainty limited the ability to draw meaningful conclusions about the outbreak [230]. As mentioned briefly above, several models have been developed to investigate the drivers of periodic reemergence of syphilis epidemics [73, 147, 161, 162]. Distinguishing between epidemic cycling behaviour driven by host immunity and occasional outbreaks driven by changes in sexual behaviour has been the focus of these models, as it has important implications for syphilis control; in particular, if epidemic cycling is an inherent feature of syphilis dynamics, then intensification of control efforts during specific stages of the cycle may be warranted, and behavioural interventions are likely to be ineffective [147]. A stochastic dynamic model was developed to investigate the periodic resurgence of syphilis observed in the United States [73]. The authors concluded that periodic increases in syphilis incidence are to be expected when there is a partially protective host immune response, due to the build up of non-immune individuals. They argue against behaviour changes driving these dynamics, as concomitant oscillations in gonorrhea, an infection without a transient immune state, have not been observed, despite the fact that changes in sexual behaviour would be expected to impact transmission of both infections [73]. This model also demonstrated the importance of travel networks, with sexual contacts between individuals from different cities resulting in the synchronization of syphilis epidemics across the country [73]. However, a subsequent analysis using a similar model structure re-investigated the conclusion that cycling syphilis epidemics are an endogenous feature of the system, driven by immunity [147]. The authors found that cycling only occurred under specific conditions, and required unrealistically high rates of partner change of at least 10 partners per year, or an average of 323 lifetime sex partners. Finally, a network model was developed to investigate both of these potential mechanisms [162]. The model assumed a small-world sexual network, where most individuals were connected to spatially close neighbours, with occasional long-distance connections between individuals [203]. The base case model identified recurring spatial

61 49 waves of infections that occurred with a periodicity of approximately 10 years, similar to the results presented by Grassly et al. [73]. They identified pacemaker centres, regions of the network that were highly connected and consistently at the centre of new waves of infection, which subsequently spread throughout the population [162]. Based on these dynamics, the authors concluded that both sexual behaviour and immunity were important for recurrent syphilis outbreaks. In the absence of transient immunity, or when a significant proportion of the population was immediately resusceptible to infection following treatment, no oscillations occurred. The structure of the sexual network was also critical; re-infection of the pacemaker centres relied on short-cut connections with distant regions of the network for infection reintroduction. The authors did not consider other possible types of networks, although scale-free networks may be more appropriate for modeling sexually transmitted infections [203]. None of the models examining syphilis periodicity described above explicitly examined the impact of treatment on the development of protective immunity. In the base case, a period of protection from re-infection was assumed for all individuals following infection, and sensitivity analyses explored the effect of returning a proportion of infected individuals directly to the susceptible pool following recovery [73, 162]. As the proportion of infected individuals not developing protective immunity increased, the periodicity of oscillations in syphilis prevalence decreased, with the models ultimately displaying Susceptible-Infected-Susceptible dynamics and no oscillations [73]. A recent network-based model showed that periodicity in syphilis epidemics can be generated in the complete absence of sterilizing immunity, that is, using a Susceptible- Infected-Susceptible framework [161]. The model assumed an adaptive behavioural response whereby susceptible partners of infectious syphilis cases preferentially severed their sexual connections with cases. As infection spread in the network, individuals were more likely to be connected to an infected individual, resulting in more disconnections and decreased infection transmission. By contrast, when syphilis prevalence was low, partnerships were of longer duration, as fewer partnerships were dissolved due to infection status. A major assumption of this model was that syphilis cases were symptomatic throughout the course of their infection, such that partners of cases were

62 50 aware of their partner s infection status and could make an informed decision to sever connections. Although treatment was not explicitly modeled, in the base case it was assumed that all infected cases became resusceptible to infection 79 days after initial infection, which would be structurally equivalent to a population where all cases received treatment relatively early in the course of infection Models Examining Effectiveness of Syphilis Control Measures All of the models developed to examine the effectiveness of syphilis control measures incorporated non-linear effects, and varied in complexity from deterministic compartmental models to stochastic individual-based models. Garnett et al. [4] explored how treatment at relatively high rates impacted syphilis dynamics in a population with high rates of partner acquisition. This model assumed a period of immunity to re-infection for individuals treated during latent infection. When treatment was introduced concomitantly with infection in an infection-naïve population, treatment was shown to reduce overall syphilis prevalence and the effective reproductive number, relative to what was observed in the absence of treatment. Once syphilis reached an endemic steady state, prevalence of primary and secondary disease was higher than that observed in the absence of treatment, as individuals with treated latent infection eventually became resusceptible to infection and a smaller proportion of the population acquired immunity to re-infection, due to high rates of treatment during the earlier stages of infection. By contrast, if treatment was introduced into a population where syphilis had reached a steady state, the same level of treatment as used in the previous scenario resulted in elimination of infection over the short term. This occurred because syphilis was endemic, with a large proportion of the population already infected, yielding an effective reproduction number of 1. Screening reduced this value to less than 1, leading to elimination. Over time, with waning immunity in those with treated latent infection, the proportion of the population susceptible to infection would increase and an outbreak could occur if an infectious individual was introduced into the population.

63 51 Two models have suggested possible mechanisms for the occurrence of rebound in cases following mass treatment [148, 149]. Both models demonstrated that a single mass treatment intervention was expected to lead to a transient decrease in infection, followed by a rapid increase to incidence at or beyond rates observed prior to the intervention. More rapid rebound was observed when individuals were assumed to be immediately re-susceptible to infection [148]. In a population with a high burden of latent syphilis, mass treatment had the effect of returning these infected but no longer infectious individuals to the pool of susceptible individuals, where they could become reinfected and drive increases in syphilis incidence [148]. The use of ongoing syphilis chemoprophylaxis has been proposed as a novel means of controlling syphilis in MSM [235]. Mathematical modeling has suggested that targeted chemoprophylaxis in MSM reporting relatively high rates of sexual partnerships could prevent significant numbers of new infections in the modeled population, with overall impact varying based on adherence and effectiveness, as well as the risk-groups targeted [223]. Duration of the intervention was a key driver of estimated effectiveness; when chemoprophylaxis was used for only one year, syphilis diagnoses rapidly returned to pre-intervention levels [223]. Syndromic management programs were projected to have a relatively large impact on reducing syphilis prevalence in South Africa [222], although estimates of the impact of these types of STI control programs were shown to be subject to a great deal of uncertainty [225]. Variability in model projections came both from uncertainty related to the cost and efficacy of proposed interventions, as well as uncertainty in parameters describing STI natural history [225]. Several modeling studies have demonstrated the importance of screening for epidemic control [219, 220, 224]. A compartmental model evaluated the impact of implementing a point-of-care screening program among female sex workers (FSW) in China, over a 5-year intervention period and a 10-year post-intervention follow-up period [219]. Frequent screening of both FSW and their clients was required to reduce syphilis prevalence and avert infections. A relatively small rebound effect following intervention

64 52 cessation was projected. An agent-based model of syphilis transmission among MSM in Australia evaluated screen and treat interventions at different population coverage levels and frequencies [220, 224]. Increasing frequency of screening among men with a history of syphilis testing was expected to be far more impactful than expanding screening programs to increase the proportion of the population screened annually. With screening every 3-months among those with a history of seeking testing, syphilis diagnoses were expected to decline to zero after 10 years [224]. More complete contact tracing was shown to have a relatively minor impact on syphilis dynamics in this population [224]. Targeted screening in men engaging in group sex or reporting more than 10 partners per year was also shown to be an efficient approach to reducing syphilis transmission [224]. Although more frequent and targeted screening were identified as effective approaches for reducing syphilis transmission, the number of tests required per infection averted was relatively high, compared to the yields with improved contact tracing [224], highlighting the possible divergence between the population-level and individual-level impact of public health interventions. When the model was extended to include HIV transmission, the qualitative results remained unchanged [220]. An additional finding was that increased frequency of screening among HIV-positive men, among whom there was a high syphilis prevalence, could limit the outbreak, despite this group representing less than 10% of the modeled population [220]. Targeting interventions to specific risk groups within a population was the focus of a small-world network model that assessed the impact of a hypothetical syphilis vaccine [162]. The authors showed that by identifying and targeting higher risk regions of a sexual network (individuals presenting with repeat infections over time and their contacts), a much smaller proportion of the population would need to be vaccinated to achieve syphilis elimination than would be predicted using traditional calculations based on the basic reproductive number [162]. Behaviour change, including reduced rates of sexual partner acquisition and increased condom use, has been modeled in the context of reducing syphilis transmission. Declining syphilis diagnoses in a modeled population representing a sub- Saharan African city with a severe and advanced HIV epidemic were estimated to be

65 53 largely driven by reductions in men visiting FSW and partnership rates [232]. A hypothetical intervention leading to short-term reductions in rates of partner acquisition and increases in condom use was not projected to have a large impact on reducing syphilis transmission among MSM in Australia [221]. Large and sustained changes in behaviour would have to be widely adopted to reduce syphilis transmission. Althouse and Hébert-Dufresne showed that, depending on the stage of the epidemic when interventions were introduced, behaviour change, which was modeled as severing sexual relations with infected partners, could be more effective than treatment for reducing syphilis transmission [161]. A compartmental model was used to examine whether the approach of isolation and quarantine used by the city of Zurich during 16th century syphilis epidemics was an appropriate response [218]. These investigators concluded that such an approach could be effective, when the carrying capacity of the facility used for isolation was sufficiently large and cases were appropriately identified and removed from the population Models Examining Cost-Effectiveness of Public Health Control Measures A number of mathematical models have examined the cost-effectiveness of interventions to prevent syphilis and other STIs. The majority of these models did not incorporate indirect effects associated with averted infections [107, 226, 228, 229, 233, 236] and only one used quality-adjusted life years (QALY) as an outcome measure [226]. Most evaluations compared infections averted or correct diagnoses as the outcome of interest. Several studies assessed the impact of alternate syphilis screening algorithms on cost-effectiveness using cohort decision analytic models [106, 107, 228]. An evaluation comparing a treponemal test first (reverse) algorithm to a non-treponemal test first (traditional) algorithm in the general population of Alberta, Canada, found that the reverse algorithm was projected to be cost-saving (i.e., less costly and more effective) in this population, using correct diagnoses as a measure of effectiveness. A study of

66 54 treponemal-first versus non-treponemal first algorithms in the general population of the United States projected that the treponemal-first algorithms would be more costly and more effective (incremental cost-effectiveness ratio of $2042 per additional case treated) than the traditional algorithm [106], with increased follow-up and overtreatment of false positives driving the increased costs. An additional analysis that considered different testing algorithms in high and low syphilis prevalence settings confirmed these qualitative findings [107]. The fact that these evaluations did not use scaled, generic measures of health as outcomes makes it challenging to draw conclusions about the cost-effectiveness of these approaches. First, this is because in the absence of generic measures of health, it is difficult to derive meaningful comparisons of the relative value gained by preventing a given health state versus prevention of an alternate state. The QALY is widely used in health economics as a summary measure of health outcome, as it allows comparisons across different diseases and populations, facilitating healthcare resource allocation decisions [237]. For example, it would be difficult to appreciate the relative benefit associated with a dollar spent to prevent an incident case of syphilis, as opposed to spending the same dollar to prevent a case of pneumonia, diabetes, or heart disease. Secondly, recommended World Health Organization (WHO) benchmarks for determining cost-effectiveness depend on health being measured using instruments that are interpretable in terms of (quality- or disability-adjusted) life years gained. For example, WHO identifies programs as highly cost-effective in a given country if health is purchased at a cost of less than per capita GDP for a life year [238]. Analyses that consider clinical delivery of services to individuals commonly express health gains in terms of quality-adjusted survival time gained (e.g., QALY) [239]. QALY are explicitly preference-based, scaled, generic measures of health, and can be conceptualized as quality weights applied to survival [239]. Analyses that consider disease control or elimination on a population level commonly use the disability-adjusted life year (DALY) metric, which considers both disutility for individuals and productivity losses associated with disease, which may be age-specific [239]. The application of both QALY and DALY to health economic evaluations of syphilis control has been extremely limited.

67 55 Cost-effectiveness analysis has also been used to evaluate interventions applied to specific population groups. Cost-effectiveness analyses of occupation-based STI screening programs [229] and frequent STI screening of female sex workers in a highincome country [226] have suggested that based on current incidence or prevalence estimates, such programs are unlikely to be cost-effective. A program distributing female condoms to sex workers in South Africa was projected to be cost-saving, with most of the gains achieved from averted HIV infections [233]. A dynamic compartmental model found that a program screening and treating incarcerated men who have sex with men was expected to be cost-saving, even under conservative assumptions around number of sexual partnerships [227]. Finally, a decision analytic cohort model showed that azithromycin was expected to be a cost-saving alternative treatment for incubating syphilis in partners of infectious cases [231], although current Canadian guidelines do not recommend its routine use [14] and the static model used in this analysis would lack the ability to project downstream impacts of antibiotic resistance that might accrue with widespread azithromycin use Summary of Mathematical Modeling Studies Mathematical models have proven to be a useful tool for understanding syphilis dynamics, epidemiology, and control. The modeling studies summarized above highlight some of the challenges associated with syphilis control and suggest some reasons why current public health efforts have not been effective in reducing infection burden in MSM at the population level. Behavioural interventions, such as increasing condom use (particularly during oral sex) and reducing partner number, are likely to face low acceptability and need to be widely adopted to be impactful [221, 240]. The effectiveness of partner notification is limited for anonymous sexual partnerships. Given these challenges, high frequency of screening is likely required in the absence of behaviour change.

68 Chapter 2: Research Aims and Hypothesis 56

69 Rationale Many urban centres in high-income countries are experiencing a resurgence of syphilis, with the epidemic concentrated in men who have sex with men. Although past epidemics have subsided, likely due to a combination of public health control efforts and sexual behaviour change, the current epidemic persists. Since the introduction of effective antibiotic treatment, sustained elevated rates of early syphilis in a population group over the period of a decade or more is a unique occurrence that has not been observed previously in high-income settings. This phenomenon of persistence in the presence of ongoing efforts to screen and treat at-risk men was the motivating factor for the work presented in this thesis. Frequent screening of at-risk individuals remains the best available tool for syphilis control. I sought to understand why current control efforts are not resulting in reduced syphilis incidence and to investigate how screening might be used to greater effect. Mathematical models of infectious diseases provide a useful framework for understanding disease dynamics and the impacts of disease control interventions. In settings were resources are finite, maximizing the impact of interventions is critical, and mathematical modeling can provide a rational framework to guide this decision-making process. The overarching objective of this thesis was therefore to use mathematical modeling to understand the effect of different approaches to syphilis screening on epidemic dynamics and the health of MSM. 2.2 Hypothesis The central hypothesis of the work presented in this thesis is that in the absence of more effective screening programs, and in the face of unchanging sexual behaviour at the population level, syphilis will persist in MSM. I sought to address this hypothesis by evaluating the impact of different screening approaches in modeled populations of men who have sex with men. Different mathematical modeling approaches were used, with the methodology used guided by the specific aims and hypotheses of the research questions.

70 Research Aims The first aim of my research was to evaluate whether expanding population coverage or increasing test frequency among individuals already undergoing routine syphilis screening would be more effective in a population of high-risk men who have sex with men. I hypothesized that different approaches to syphilis screening in MSM would have different population-level effects on epidemic control, as measured by incident and diagnosed syphilis cases. Although previous mathematical modeling studies have demonstrated that screening is an effective means of reducing syphilis transmission in populations, how screening is applied, in terms of targeted populations and frequency of testing, can result in very different, and sometimes surprising outcomes, and appears to be context dependent [149, 220, 224]. I tested this hypothesis by constructing an agent-based model of syphilis transmission in a core group of sexually active MSM that captured complex sexual behaviours and the HIV status of individuals. The modeled population reflected the demographics of MSM residing in the city of Toronto, Canada, which has experienced elevated rates of early syphilis since the early 2000s [241]. The aim of the second study was to determine the cost-effectiveness of incorporating routine syphilis testing into the blood-work of MSM under medical care for HIV. The first study showed that more frequent screening was likely to be an effective approach for epidemic control, but I hypothesized that opt-out screening leading to overtreatment and increased medical intervention in a cohort of men with a high prevalence of previously treated syphilis might make this approach economically unattractive. To investigate this possibility, I constructed a microsimulation model that incorporated syphilis natural history, screening and treatment, and the associated costs. The modeled population was based on the participants of the Ontario HIV Treatment Network Cohort Study [242], in which a similar approach had been proposed (and is currently undergoing evaluation [243]). The previous studies considered screening strategies for short-term syphilis control. The final aim of my research was twofold: (i) to examine the impact of sustained syphilis screening at varying levels of population coverage, and (ii) to investigate how

71 59 different potential syphilis epidemic drivers interact with treatment to alter population infection dynamics. I hypothesized that screening at levels insufficient for syphilis elimination may be contributing to infection persistence, by acting to replenish the pool of individuals susceptible to infection, and that assumptions around protective immunity following treatment would alter the effectiveness of screening programs. I constructed a deterministic compartmental mathematical describing syphilis transmission in a population of sexually active MSM to investigate these questions. The population in this study represented a general population of sexually active MSM, but adopted some characteristics of MSM residing in Toronto, Canada.

72 Chapter 3: Screen More or Screen More Often? Using Mathematical Models to Inform Syphilis Control Strategies The work described in this chapter is published as: Tuite AR, Fisman DN, Mishra S. Screen more or screen more often? Using mathematical models to inform syphilis control strategies. BMC Public Health 2013;13:

73 Abstract Background: Syphilis incidence among men who have sex with men (MSM) continues to rise despite attempts to increase screening and treatment uptake. We examined the marginal effect of increased frequency versus increased coverage of screening on syphilis incidence in Toronto, Canada. Methods: We developed an agent-based, network model of syphilis transmission, representing a core population of 2,000 high-risk MSM. Epidemiological and biological parameters were drawn from regional surveillance data and literature-derived estimates. The pre-intervention period of the model was calibrated using surveillance data to identify 1000 credible simulations per strategy. Evaluated strategies included: annual syphilis screening at baseline coverage, increased screening frequency at baseline coverage, and increased coverage of annual screening. Intervention impact was measured as annual prevalence of detected early (primary, secondary, and early latent) syphilis cases and syphilis incidence per year over 10 years. Results: Of the strategies evaluated, increasing the frequency of syphilis screening to every three months was most effective in reducing reported and incident syphilis infections. Increasing the fraction of individuals tested, without increasing test frequency, resulted in a smaller decline in incidence, because reductions in cases via treatment were counterbalanced by increased incident syphilis among individuals with prior latent syphilis. For an equivalent number of additional tests performed annually, increased test frequency was consistently more effective than improved coverage. Conclusions: Strategies that focus on higher frequency of testing in smaller fractions of the population were more effective in reducing syphilis incidence in a simulated MSM population. The findings highlight how treatment-induced loss of immunity can create unexpected results in screening-based control strategies.

74 Introduction Urban centres in high-income countries have witnessed a re-emergence of syphilis in recent years, with the epidemic concentrated among men who have sex with men (MSM) and HIV-infected individuals [22, 244, 245]. Although the reason for this reemergence remains unclear, proposed explanations include: increased survival of HIVinfected individuals [151, 152] and changes in sexual behaviour, such as increased prevalence of serosorting [74] or unprotected oral sex [51, 246]. Despite educational campaigns promoting safer sex and serological screening, the rate of diagnosed infectious cases in many North American cities continues to increase [43, 68, 145, 247, 248]. Between 2000 and 2010, there was a greater than 20-fold increase in early (primary, secondary, and early latent) syphilis cases in the Canadian city of Toronto, with annual rates increasing from 1.9 to 38.3 cases per 100,000 males [249]. Individuals who acquire syphilis may be asymptomatic, but untreated infection can lead to ocular, auditory, and neurological complications, even during the early stages of infection [250, 251]. Syphilis has also been associated with acquisition and transmission of HIV [85, 252]. Challenges that make syphilis difficult to control include: the development of vague or no symptoms during early infection, such that infected individuals may not seek treatment [166]; low serologic test sensitivity during the first 4-6 weeks of infection [100]; and the failure to develop protective immunity following successful treatment of infectious syphilis, such that in the presence of ongoing high-risk behaviours, individuals are at risk of syphilis re-infection [4, 81, 144, 145]. In the face of these challenges, frequent syphilis screening in at-risk individuals remains the best available tool for syphilis control, by reducing the length of time an individual with syphilis remains infectious and untreated. Clinical studies have shown that the adoption of more frequent syphilis screening among high-risk groups is feasible and increases the detection of early (primary, secondary, and early latent) syphilis [166, 167, 169]. A recent mathematical modeling

75 63 study suggested that more frequent testing may be more effective at reducing syphilis incidence than expanding the proportion of the population receiving annual screening [224]. However, these conclusions were based on a small number of model realizations of an epidemic in a large MSM population [224]. Further evaluation is required to determine whether the findings are robust to parameter uncertainty (using a larger number of simulations), model uncertainty (using a different model structure), and are applicable to a smaller, core population of high-risk MSM at greatest risk of infection and re-infection. Given the burden of syphilis in MSM and the clear need for more impactful interventions to reduce incidence and achieve long-term epidemic control, we created a mathematical model of syphilis transmission dynamics in the Canadian city of Toronto to estimate the potential effectiveness of different syphilis screening strategies. Specifically, we evaluated whether expanding population coverage or increasing test frequency among individuals already undergoing routine syphilis screening would be more effective in a population of high-risk MSM.

76 Methods Transmission Model Overview We modeled the dynamics of syphilis in MSM in Toronto, Canada, using an agent-based approach. We modeled Toronto because it is one of the cities most affected by the recent syphilis resurgence in the MSM population [249], and sexual behaviour data are readily available for high-risk MSM [64]. Of the approximately 500 men diagnosed with early syphilis in Toronto in 2010, 89% reported sex with men as a risk factor, and up to 48% were co-infected with HIV [249]. Data used to model the behavioural component of the model (discussed in more detail below) were derived from a second generation surveillance study that used a venue-based recruiting strategy to enroll MSM from major urban centres, such that study participants were more likely to identify with the gay community [64]. We use the term high-risk to emphasize that the modeled population likely differs from the general population of MSM in Canada with respect to frequency of unprotected anal intercourse and other high-risk behaviours, and has an incidence of HIV infection that is approximately five times higher than that observed among all MSM in this region of Canada [64, 253]. An agent-based approach allowed for the explicit modeling of complex sexual networks and recording of sexual histories and health states of discrete individuals over time. The model was constructed using the AnyLogic software package ( The model represented a population of 2,000 high-risk MSM, with individuals forming a network of sexual contacts along which the transmission of syphilis occurred. Based on local HIV prevalence estimates in MSM, 20% of the population was assumed to be HIV positive [254]. Each modeled individual was described by an infection transmission component and a partnership component, as described in greater detail below. Model parameters were drawn from the literature and were Toronto-specific, wherever possible, or were based on expert opinion in the absence of data. Additional model details are provided in Appendix A.

77 65 Infection Transmission Component The infection transmission component of the model (Figure 3-1) represented an individual s health state and incorporated the stages of syphilis infection: susceptible, incubating, primary, secondary, early latent, late latent, and immune. Uninfected individuals became infected via a transmission event following sexual contact with an infected partner. Individuals progressed through the various stages of syphilis, with parameters describing the transitions between states listed in Table 3-1. Men were tested for syphilis if they sought medical care for symptoms or accepted screening, with a likelihood of correct diagnosis dependent on test sensitivity [99, 100]. We did not include the diagnosis of false positive cases. A proportion of regular and casual contacts of diagnosed cases were identified and screened via contact tracing, with trace back periods dependent on disease stage, as per Canadian guidelines (3 months, 6 months, and 1 year, for primary, secondary, and latent syphilis, respectively) [14]. At baseline, HIV-positive men were more likely to undergo regular screening than HIV-negative men [83, 254]. Successful treatment of infectious individuals was assumed to abort the development of protective immunity for men with primary or secondary syphilis, such that they returned directly to the susceptible state following effective treatment [140]. Individuals with latent syphilis who received treatment were assumed to be immune from re-infection for 5 years [4]. We assumed that an individual could not become `superinfected with syphilis if he was already infected or had recently received treatment for latent infection and was transiently immune. Partnership Component The partnership component described an individual s sexual network and was based on Toronto-specific data from the Lambda survey [64]. The biological and behavioural survey was conducted on a convenience sample of 2,500 gay and bisexual men in Toronto and Ottawa, who were enrolled through targeted gay venues [64].

78 66 Figure 3-1. Schematic of the infection transmission component of the model. Each box represents a discrete health state for an individual, with transition times between health states defined in Table 3-1. Dashed lines represent transitions that occur following successful treatment of syphilis infection.

79 67 Table 3-1. Model parameters. Parameter Details Value Distribution Source Population Characteristics Population size 2,000 Assumption Time spent in model (years) (min, max, mode) Proportion of MSM who are HIV positive 1, 34, 17 Triangular Remis et al. [64]; Assumption 0.2 Remis et al. [254]; Assumption Syphilis Natural History Probability of transmission (per act) (min, max, mode) Penileanal/Penileoral 0.01, 0.05, Triangular Gray et al. [224]; Assumption Incubation period (days) Uniform discrete Infection/infectious period (days) Tramont [255] Clark and Danbolt [256]; Schrijvers et al. [5]; Singh and Romanowski [251] Primary Uniform discrete Secondary Uniform discrete Early latent* 365 Recurrent 90 Late Latent* Until end of time in model Probability of recurrent syphilis 0.25 Stamm [257]

80 68 Parameter Details Value Distribution Source Duration of protective immunity following treatment (years) Garnett et al. [4]; Grassly et al. [73]; Magnuson et al. [140] Primary or secondary syphilis 0 Partnership Characteristics Latent syphilis 5 Number of partners in past 6 months (proportion of population in each category) Remis et al. [64] Maximum number of partners in past 6 months (by partner number category) (min, max, mode) 1 1 Assumption 2-9 2,9,8 Triangular ,29,25 Triangular ,75,50 Triangular

81 69 Parameter Details Value Distribution Source Duration of partnership (days) (min, max, mode) Remis et al. [64]; Assumption Casual 1,2,1 Triangular Regular 7,3000, Triangular 365 Behavioural Characteristics Frequency of anal sex (per day) Gray et al. [224] Casual partnership Regular partnership Frequency of oral sex (per day) Gray et al. [224] Casual partnership Regular partnership Probability of condom use (anal sex) Aral et al. [244]; Remis et al. [64] HIVconcordant 0.5 HIV-discordant 0.8 Condom efficacy Assume condom use for anal intercourse only 0.9 Fitch et al. [258]

82 70 Parameter Details Value Distribution Source Test and Treatment Characteristics Probability of seeking medical care for symptoms Bissessor et al. [166]; Buchacz et al. [259] Primary 0.25 Secondary 0.60 Time to treat (days) Bissessor et al. [260]; Assumption Primary 3-56 Uniform Secondary 1-57 Uniform Proportion of population screened routinely for syphilis Bissessor et al. [166]; Botes et al. [168]; Burchell et al. [261]; Teague et al. [262]; Assumption HIV positive 0.5 HIV negative 0.2 Test sensitivity Probability of partner notification Treponemalspecific screening test 0.95 Martin et al. [263]; Nessa et al. [264] Assumption Casual partner 0.1 Regular partner 0.6

83 71 Parameter Details Value Distribution Source Trace-back period for partner notification (months) Public Health Agency of Canada [14] Primary 3 Secondary 6 Early latent 12 Time from index case identification to screening of named partner(s) (days) 3-21 Uniform Assumption *Not infectious

84 72 The survey collected data on sexual and health-seeking behaviours, including condom use, known HIV status, and syphilis testing [64]. Behavioural parameters were estimated for men reporting one or more sexual partners in the past six months. In the model, each individual was assigned a number of desired partners per sixmonth period, and formed partnerships with other partner-seeking individuals in the simulated population. Partnerships could be concurrent or serial. Casual and regular partnerships differed by frequency of sexual contact and partnership duration. We differentiated between casual and regular partnerships for the sake of applying contact tracing (i.e., we assumed different probabilities of identification and treatment of regular versus casual partners) and for the application of different behavioural characteristics, such as condom use and frequency of sexual contact. Condom use was assumed to be nil during oral sex [64]. We assumed a greater likelihood of condom use for partnerships involving men with discordant HIV status, compared to partnerships involving HIV concordant men [64]. Model Calibration Each stochastic model realization represents one result out of many possible epidemic trajectories. We selected and analyzed simulated epidemics that reproduced the annual case detection rate of primary, secondary, and early latent syphilis among highrisk MSM in Toronto between 2006 and We used the reported number of cases of early syphilis among males in Toronto during this time period divided by the estimated size of the Toronto MSM population [253, 254, 265] to provide lower bound estimates of the expected burden of detected disease in the modeled population of 2000 men over a five year period (~50-75)[249]. The upper bound was based on annual number of cases of early syphilis diagnosed in sentinel surveillance clinics (Leo Mitterni, personal communication), which are expected to serve higher-risk MSM; using these data, we expected to see approximately 10-fold more cases in the modeled population over the five-year pre-intervention period than the number estimated using city of Toronto surveillance data (i.e., ~750 cases). This is comparable to estimates for other STIs in epidemiological core groups [266]. We therefore regarded model realizations as

85 73 credible if they produced detected case counts that fell within the uncertainty bounds defined above during the pre-intervention period (i.e. between reported infections over the five year pre-intervention period in our population of 2000 men). A total of 1000 of these credible model realizations were used for the analysis of each intervention. During the calibration process, parameters were varied within plausible limits (as described in Table 3-1). Individual parameter values in the selected well-calibrated model realizations covered this same range and distribution for key input parameter values, but were used jointly in individual realizations in a manner that optimized model fit to observed data. Control Strategies and Analysis We evaluated the impact of increasing either the coverage or frequency of syphilis screening in high-risk MSM (Table 3-2) by comparing the projected number of reported and incident syphilis cases under these interventions to the base case of annual screening at current coverage (Strategy A). Increasing annual population coverage by 10% (Strategy B) and increasing frequency of screening in those already being screened (Strategy C) were considered feasible, pragmatic interventions. However, because the increased frequency strategies required many more tests than the expanded coverage scenario, we also evaluated the impact of dramatically increasing population coverage for annual testing (Strategy D), such that the total number of tests performed per year was approximately equivalent to the number of test performed under the increased frequency scenarios. Note that screening 100% of the population annually resulted in 80 fewer tests per year than screening individuals already seeking screening every 3 months. All interventions were initiated in 2011, with immediate scale-up, and were sustained over a 10-year period. Intervention impact was measured as the prevalence of detected early (primary, secondary, and early latent) syphilis cases per year over the intervention period. We also estimated the annual incidence of syphilis. To characterize the uncertainty around these estimates, we constructed 95% bootstrap confidence intervals with 1000 replications, using sampling with replacement from the 1000 runs conducted for each model scenario.

86 74 Table 3-2. Syphilis screening strategies evaluated in the model. Intervention Description Details (A) Base case Screen every 12 months 20% of HIV-negative individuals screened 50% of HIV-positive individuals screened 60% of regular and 10% of casual partners of infectious index cases treated (B) Increase coverage of screening Increase coverage by 10% 520 tests performed annually Same as (A), but: 30% of HIV-negative individuals screened 60% of HIV-positive individuals screened 720 tests performed annually (C) Increase frequency of screening Screen every 6 or every 3 months Same as (A), but: Frequency of screening in population is increased to every 6 (1040 tests annually) or 3 (2080 tests annually) months*

87 75 Intervention Description Details (D) Equivalent number of tests Screen a proportion of the population every 12 months such that the total number tests performed is equivalent to (C) To equal every 6 months: 100% of HIV-positive individuals screened and 40% of HIV-negative individuals screened (1040 tests annually) To equal every 3 months: 100% of the population screened (2000 tests annually)* *Note that 80 extra tests are required annually for the screen every 3 months strategy, compared to the equivalent number of tests strategy with 100% annual coverage.

88 Results In total, approximately 2500 simulations were required to produce 1000 credible epidemics for each scenario. Of the discarded simulations, 37% fell below the lower limit and 63% produced detected cases above the upper limit. None of the credible epidemics achieved local elimination over the subsequent 10 years. Using model realizations that produced reported case counts within our target calibration range for the time period between 2006 and 2010, we examined the effect of increasing the frequency or coverage of screening in the model population for a 10-year period beginning in All strategies were projected to reduce the number of reported early syphilis cases, compared to the base case scenario of continuing to screen a fixed proportion of the population annually (Figure 3-2). Results were similar for incident syphilis cases (data not shown). Increasing the frequency of screening to every 3 months in men already undergoing testing was the most effective strategy for reducing syphilis cases in this population; over the 10-year intervention period, screening every 3 months was projected to avert approximately 650 incident syphilis cases relative to the base case, compared with 300 and 125 cases averted with 6 monthly or expanded annual coverage, respectively. Comparing strategies that required approximately the same annual number of tests, more frequent testing in men already accessing screening was projected to be more effective than expanding the proportion of the population that received annual testing (Figure 3-3). The proportions of cases averted, relative to the base case, were greater for the more frequent screening strategies than when the equivalent number of tests was applied to expanded coverage over the intervention period (Figure 3-4).

89 77 Figure 3-2. Model-projected annual rates of reported early syphilis. Results are based on 1000 realizations of each intervention scenario and are presented as mean values with corresponding 95% uncertainty bounds. Prior to 2011, all scenarios included annual screening only, with the specified interventions implemented at the start of 2011 (indicated by a dashed line).

90 78 Figure 3-3. Model-projected annual rates of reported early syphilis under equivalent test volume strategies. Results are based on 1000 realizations of each intervention scenario and are presented as mean values with corresponding 95% uncertainty bounds. Prior to 2011, all scenarios included annual screening only, with the specified interventions implemented at the start of 2011 (indicated by dashed vertical line). Screening every 3 months and 100% annual screening (black lines) required approximately the same number of screening tests annually, as did the semiannual and 52% annual screening strategies (grey lines).

91 79 Figure 3-4. Expected reduction in syphilis cases following implementation of different intervention strategies. The proportion of cases averted was calculated relative to the expected number of cases in the base case scenario (no increase in frequency or coverage of screening). Proportion of cases averted is presented for both diagnosed early syphilis cases and incident cases (reported and unreported) and is calculated using the mean value of 1000 realizations for each intervention scenario. Error bars represent 95% uncertainty bounds. Strategies requiring the same number of annual tests are indicated.

92 Discussion Using a dynamical model of syphilis transmission in a core group of MSM that was parameterized with the best available data on the epidemiology of the current epidemic and disease natural history, we evaluated plausible screening strategies that might be employed for epidemic control. We found that increasing test frequency in atrisk MSM who already access screening, rather than expanding outreach to provide screening to under-screened individuals, would be the optimal means of reducing syphilis incidence. While the finding may appear counterintuitive, it is consistent both with the observed rebound in syphilis rates that occurred in Vancouver in the context of dramatic expansion of empirical treatment efforts [149], and with the hypothesis that recurrent syphilis epidemics occur as a result of emergence of large populations of susceptible, rather than infectious, individuals [73, 162]. Our findings are expected under a disease natural history model that includes immunity to super-infection in individuals with late syphilis, and which allows infected individuals to become re-infected (and infectious) after treatment. Rebound in such a system occurs because creation of a large population of newly syphilis-susceptible individuals, via treatment, provides the necessary population at risk for a marked increase in new infections. By contrast, screening and treating a limited subset of the population more frequently and rapidly continually truncates infectiousness in the population, allowing case numbers to decline. By incorporating sexual network data and capturing the indirect effects of interventions, such as the downstream prevention of cases following treatment of a single individual, we are able to evaluate the potential impact of changes in syphilis screening on disease dynamics. Previous modeling studies of syphilis transmission have provided important qualitative and quantitative insights for epidemic control by exploring the influence of key parameters on syphilis rebound [4, 22, 148, 149]. These studies demonstrated that a rapid loss of immunity following treatment, faster turnover in highrisk groups, and higher rates of partner exchange would lead to an earlier and larger rebounds in syphilis incidence [22, 148, 149]. The observed transient drop in cases

93 81 observed prior to initiation of the interventions (in Figures 3-2 and 3-3) likely represents damped oscillations occurring in the system, as would be expected to occur in a dynamic model with replacement of removed individuals with susceptible individuals. To our knowledge, only one previous study has performed a similar examination using an agent-based model of syphilis in Australian MSM, and demonstrated the potential impact of more frequent screening in MSM, compared to increasing the proportion of MSM screened [224]. Our model differs from the Australian study, in that we focused on a core group of high-risk MSM, with higher expected rates of partner change and connectivity. Our results therefore confirm that the Australian findings are robust to parameter and model uncertainty, and are generalizable to a smaller, core population of high-risk MSM, in a different epidemic context. Our findings lend epidemiological support to current STI guidelines that recommend frequent syphilis screening in MSM [14, 104]. Our model is subject to limitations, including uncertainty in model parameters. When the appropriate data were available, parameters were drawn from distributions to account for this uncertainty. Although we included results from a large number of model realizations for each strategy, we did not explicitly evaluate the impact of stochastic uncertainty on model results within each parameter set. However, the impact of stochasticity on the results appears to be small, likely due to the reasonably large number of simulations performed. The partnership component of the model was parameterized using data from a cross-sectional study, and thus may not capture changes in behaviour over time, and their consequent impact on the syphilis epidemic in Toronto s MSM population. Although our model included HIV status, we did not attempt to evaluate the synergistic impact of syphilis testing and treatment on syphilis-hiv co-infection. An important next step is to incorporate costs and evaluate the cost-effectiveness of these strategies. In summary, a model that incorporates the best available data on syphilis transmission in MSM core groups in a large urban centre suggests that syphilis screening campaigns will be most successful if they focus on reducing the interval between tests in

94 82 high-risk MSM, rather than focusing on outreach to increase the proportion of the population screened.

95 Chapter 4: Cost-effectiveness of enhanced syphilis screening among HIV-positive men who have sex with men: a microsimulation model The work described in this chapter is published as: Tuite AR, Burchell AN, Fisman DN. Cost-effectiveness of enhanced syphilis screening among HIV-positive men who have sex with men: a microsimulation model. PLoS One 2014;9:e

96 Abstract Background: Syphilis co-infection risk has increased substantially among HIV-infected men who have sex with men (MSM). Frequent screening for syphilis and treatment of men who test positive might be a practical means of controlling the risk of infection and disease sequelae in this population. We evaluated the cost-effectiveness of strategies that increased the frequency and population coverage of syphilis screening in HIV-infected MSM receiving HIV care, relative to current standard of care. Methods: We developed a state-transition microsimulation model of syphilis natural history and medical care in HIV-infected MSM receiving care for HIV. We performed Monte Carlo simulations using input data derived from a large observational cohort in Ontario, Canada, and from published biomedical literature. Simulations compared usual care (57% of the population screened annually) to different combinations of more frequent (3- or 6-monthly) screening and higher coverage (100% screened). We estimated expected disease-specific outcomes, quality-adjusted survival, costs, and costeffectiveness associated with each strategy from the perspective of a public health care payer. Results: Usual care was more costly and less effective than strategies with more frequent or higher coverage screening. Higher coverage strategies (with screening frequency of 3 or 6 months) were expected to be cost-effective based on usually cited willingness-to-pay thresholds. These findings were robust in the face of probabilistic sensitivity analyses, alternate cost-effectiveness thresholds, and alternate assumptions about duration of risk, program characteristics, and management of underlying HIV. Conclusions: We project that higher coverage and more frequent syphilis screening of HIV-infected MSM would be a highly cost-effective health intervention, with many potentially viable screening strategies projected to both save costs and improve health when compared to usual care. The baseline requirement for regular blood testing in this group (i.e., for viral load monitoring) makes intensification of syphilis screening appear readily practicable.

97 Introduction Major urban centers in the developed world have witnessed a dramatic reemergence of syphilis in recent years, with the epidemic concentrated among men who have sex with men (MSM) and HIV-infected individuals [22, 241, 244, 245]. Rates of syphilis acquisition among HIV-positive MSM up to 300-fold higher than those observed in the general male population have been reported [267]. Early detection and treatment of infected individuals both prevents the development of neurological and other complications of infection [250, 251] and reduces the duration of infectiousness, thereby limiting syphilis transmission to other individuals [4]. Canadian guidelines recommend that individuals at ongoing risk for syphilis be screened at 3-month intervals and that all MSM with STI risk, regardless of HIV status, undergo screening at least annually [14]. Annual syphilis testing rates were approximately 57% in HIV-infected MSM enrolled in the Ontario HIV Treatment Network Cohort Study (OCS) in 2009 [83]. The disproportionate and increasing burden of syphilis in MSM and HIV-infected individuals has led to calls for novel syphilis control strategies targeting these populations [224, 260, 268]. Men who are currently under medical care for HIV typically undergo routine blood work every 3 to 6 months [14]. Implementing routine syphilis serologic testing in this group presents a practical and inexpensive opportunity for increasing screening frequency, allowing for timely detection and treatment [166]. The current syphilis testing algorithm in Ontario, Canada includes a treponemal screening test followed by a confirmatory non-treponemal test in individuals who test positive in the initial screening test [269]. A second confirmatory treponemal test may be used to help interpret findings in individuals with a positive screening test and nonreactive non-treponemal test. Although the treponemal test is more sensitive than the non-treponemal test for detecting early infection, once the test is positive, it often remains so for life, resulting in low specificity in individuals with previously treated syphilis [99, 100]. Follow-up non-treponemal tests can distinguish current from past (treated) infection, although with imperfect sensitivity and specificity. Increasing the frequency of syphilis screening in a population with high prevalence of previous syphilis

98 86 infection, as is seen in HIV-positive MSM in Ontario [83], may thus result in a high number of false positives, leading to unnecessary treatment [106, 107] and diversion of limited public health resources [270]. Given the current burden of syphilis in HIV-positive MSM, it is unclear if the benefits associated with preventing new cases of neurosyphilis and tertiary syphilis via more frequent screening would outweigh the costs and health consequences associated with unnecessary treatment. Our objective was to evaluate whether strategies of enhanced (more frequent and/or higher coverage) routine syphilis screening in HIVpositive MSM receiving HIV medical care would be effective and economically attractive, relative to the current standard of care.

99 Methods Model Overview We developed an individual-level state-transition ( microsimulation ) model [271] that follows simulated patients from time of model entry until death. The model included syphilis screening, the natural history of syphilis, and costs and consequences of syphilis screening and treated and untreated syphilis. We synthesized health and economic data related to syphilis in HIV-positive MSM; probability, cost, and quality-oflife estimates were derived from the published literature wherever possible. Model parameters are presented in Table 4-1. The model was used to simulate 500,000 individual men similar to those enrolled in the Ontario HIV Treatment Network Cohort Study (OCS), an observational, open dynamic cohort of people receiving medical care for HIV infection in Ontario, Canada [242]. OCS data were used to estimate the proportion of men in the cohort with previous treated infection [83]. In our base case, we compared usual care to more frequent screening, and screening with higher population coverage. For the usual care strategy, 57% of the population received syphilis screening annually [83]. In the usual coverage 3 and -6 months strategies, 57% of the population was screened every 3 or 6 months. In higher coverage strategies, 100% of the population was screened annually, or every 3 or 6 months. In all usual coverage strategies, individuals who had been screened once were more likely to be tested in the future [83]. Upon model initiation, 50% of individuals who were screened in the first round of screening were assigned to the regular screening group and were screened at all subsequent screening events; individuals in the nonregular screening group were selected at random at each screening event, such that the total proportion of the population screened equaled 57%.

100 88 Table 4-1. Model variables and sources. Variable Details Base Case Value Range Source Population Characteristics Age (yr, SD) Burchell et al. [83] CD4 count (cells/ L)(SD) Burchell et al. [83] Population with previous treated syphilis infection (%) a 21 Burchell et al. [83] Annual incidence (%) First infection 4.0 Burchell et al. [83] Re-infection 4.8 Burchell et al. [83] Probability of death (per yr) Baseline probability of death Age-specific estimates Statistics Canada [272] Excess mortality hazard in HIV-positive individuals Bhaskaran et al. [273] Syphilis Natural History Average duration of syphilis stage (mo) Primary Garnett et al. [4]; PHAC [14] Secondary Garnett et al. [4]; PHAC [14] Early latent Garnett et al. [4]; PHAC [14]

101 89 Variable Details Base Case Value Range Source Duration of immunity after treatment of late latent syphilis (yr) Time to develop late neurosyphilis (yr) Probability of developing symptomatic neurosyphilis Probability of neurologic involvement Garnett et al. [4] Golden et al. [18] Golden et al. [18]; Rolfs et al. [94]; Zetola and Klausner [273] Probability of developing early neurosyphilis b (among those with neurological involvement) Probability of developing late neurosyphilis (among those with neurologic involvement) Golden et al. [18] Golden et al. [18] Probability of recovery from symptomatic early neurosyphilis without disability, following treatment CDC [274] Odds of developing neurosyphilis if CD4 count <350 cells/ml (relative to CD4 350) Time to develop tertiary syphilis (yr) Ghanem et al. [275]; Marra et al. [276] PHAC [14]

102 90 Variable Details Base Case Value Range Source Probability of developing tertiary syphilis (gummatous and cardiovascular) Test Characteristics Larsen [99]; Golden et al. [18] Screening test (EIA) sensitivity Primary Sena et al. [100] Secondary Owusu-Edusei et al. [106]; Sena et al. [100] Early latent Owusu-Edusei et al. [106]; Sena et al. [100] Late latent Owusu-Edusei et al. [106] Screening test (EIA) specificity No prior syphilis infection Sena et al. [100] Previous treated syphilis infection Blandford et al. [277]; Owusu-Edusei et al. [107]

103 91 Variable Details Base Case Value Range Source Confirmatory test (RPR) sensitivity Primary Sena et al. [100] Secondary Owusu-Edusei et al. [107]; Sena et al. [100] Early latent Sena et al. [100] Confirmatory test (RPR) specificity Late latent No prior syphilis infection Owusu-Edusei et al. [107]; Sena et al. [100] Sena et al. [100] Probability of receiving lumbar puncture Previous treated syphilis infection Late latent, tertiary, neurosyphilis or any state with CD4 count 350 cells/ L CD4 count >350 cells/ L and not in late latent, tertiary or neurosyphilis stage Owusu-Edusei et al. [107] 1 PHAC [14] 0 PHAC [14] Probability of post-dural headache following lumbar puncture Turnbull and Shepherd [278]

104 92 Variable Details Base Case Value Range Source Treatment Characteristics Probability of seeking treatment for syphilis symptoms Primary Bissessor et al. [166]; Kourbatova et al. [229] Secondary Bissessor et al. [166]; Kourbatova et al. [229] Early latent Bissessor et al. [166]; Kourbatova et al. [229] Probability of treating individual identified as syphilis infected True positive Blandford et al. [231]; Assumption False positive, no prior history of syphilis infection False positive, history of treated syphilis infection Blandford et al. [231]; Assumption Assumption Probability of treatment failure Early syphilis c Late syphilis Blank et al. [279]; Riedner et al. [280] Blank et al. [279]; Riedner et al. [280] Probability of anaphylaxis following treatment Tsevat et al. [281]

105 93 Variable Details Base Case Value Range Source HIV Natural History and Treatment Set point viral load (log copies/ml) 4.6 Deeks et al. [282]; Mocroft et al. [283]; Rhone et al. [284] Increase in CD4 count with initiation of ART (cells/ L) 535 initialcd ( )0.98 initialcd4 ( ) Drusano et al.[285]; Sanders et al. [286] Decline in CD4 count with detectable viral load (cells/ L) x log viral load Cook et al. [287] Strategies Achieved coverage Higher coverage screening 1 Assumption Usual care 0.57 Burchell et al. [83] a We assumed no prevalent infection at baseline. b Early neurosyphilis refers to neurosyphilis that develops during the primary, secondary, or early latent stages of syphilis infection. c Early syphilis refers to primary, secondary, or early latent syphilis infection.

106 94 We assumed a standard screening algorithm consisting of a treponemal-specific assay (such as EIA) followed by a non-treponemal assay (such as RPR), which is now standard practice in Ontario. Test sensitivities were based on stage of infection, and specificities depended on prior infection history. After testing, men could be correctly or incorrectly classified as infected or uninfected, with incorrect classification resulting in the inappropriate use of resources (for false positives and false negatives). Among men with a history of previous treated syphilis, we assumed different probabilities of treatment for true vs. false positive cases, to capture the incorporation of additional information (such as reported sexual risk behaviour) in a clinician s decision to treat. This assumption was tested in sensitivity analyses. Screening programs were conducted for 20 years in the base case. Each month, individuals could seek testing for syphilis symptoms. All individuals were eligible for screening, with frequency and uptake dependent on the strategy. We assumed that higher screening would be implemented as an opt-out policy with perfect compliance in our base case. We assumed that lumbar puncture was performed for all men diagnosed with late latent syphilis, neurosyphilis, or tertiary syphilis, and for those testing positive for syphilis with CD4 count < 350 cells/ml [14]. An overview of the decisions involved in screening and treatment is presented in Figure 4-1. The Markov model consisted of the following mutually exclusive and collectively exhaustive health states: uninfected; primary syphilis, secondary syphilis and early latent syphilis ( early syphilis ); late latent syphilis; neurosyphilis; tertiary syphilis; treated late syphilis; and death (Figure 4-2). The uninfected state was subdivided into uninfected with no prior syphilis infection and uninfected with previously treated syphilis infection, to capture different syphilis incidence [83], test specificities, and probabilities of treatment following positive test results in these two groups. The neurosyphilis state was divided into early symptomatic neurosyphilis and late neurosyphilis, to reflect the fact that individuals treated for early neurosyphilis could recover without long-term disability [274].

107 95 Figure 4-1. Simplified overview of screening and treatment component of the decision analytic model. A decision is made to screen or not screen HIV-positive MSM at risk of syphilis acquisition. If screening is performed, men can be correctly or incorrectly classified, resulting in appropriate or inappropriate use of resources, with associated costs and health consequences. Syphilis is treated according to the current Canadian guidelines. Chance nodes indicate points at which probabilities (described in Table 4-1) are applied. Individuals progress to the appropriate health state (outlined in Figure 4-2) following progression through the screening and treatment decision tree. Note that men who seek treatment for symptomatic syphilis infection follow the same set of decisions.

108 96 Figure 4-2. Markov model overview. The model has a total of eleven health states, with the neurosyphilis state further subdivided into early symptomatic neurosyphilis and late neurosyphilis. Arrows indicate allowed transitions between states. Men have a chance of remaining uninfected or acquiring syphilis and progressing through the disease states. Syphilis infection is characterized by four stages: primary, secondary, early latent, and late latent, which may develop into tertiary syphilis. Men only exit the tertiary syphilis state via death, even if they receive treatment. Men may develop neurosyphilis at any stage of their syphilis infection. Men with early neurosyphilis may receive treatment and recover without disability or have lifetime disability (indicated by entry into the tertiary syphilis state). All men with late neurosyphilis are assumed to have lifetime disability and enter the tertiary syphilis state. Men may transition to the death state from any model state. Treatment results in men returning to the previously infected and treated state. Movement through the model health states depends on transition probabilities identified from the literature. Disutilities are associated with the primary, secondary, tertiary, and neurosyphilis states, and long-term healthcare costs are associated with tertiary and neurosyphilis.

109 97 In the base case, incidence of syphilis infection was based on estimated rates observed in the OCS for MSM, with infection rates higher in men with prior syphilis [83]. As surveillance data suggest that risk of syphilis infection is low in individuals aged 65 [241], we assumed that men were at risk of syphilis acquisition for a period of 20 years after model entry, although this was varied in sensitivity analyses. Monthly transitions between stages of infection (i.e., primary to secondary, secondary to early latent, early latent to late latent) and return to the uninfected state following treatment for late syphilis, were derived from mean estimates of time spent in each stage and were converted to probabilities assuming an exponential distribution [288]. Individuals treated for primary, secondary, or early latent syphilis were assumed to return immediately to the uninfected state. Individuals successfully treated for late latent or early symptomatic neurosyphilis with no long-term disability entered a transient immune state prior to returning to the uninfected state [4]. Individuals with early symptomatic neurosyphilis were assumed to either recover after six months, or transition to the tertiary syphilis state, to capture the long-term health consequences associated with their health state. All individuals with late neurosyphilis were assumed to have long-term health consequences and transitioned to the tertiary state following treatment, to capture the associated lifetime costs and disutility. Tertiary syphilis was considered to be associated with long-term sequelae whether or not treatment was received (i.e., transitions back to morbidity-free health states were disallowed). Mortality rates were based on Canadian life tables [289], the excess mortality observed in HIV-infected individuals compared to the general population [273], and the excess mortality associated with untreated tertiary syphilis [290]. In all strategies, syphilis infection was managed according to the current Canadian guidelines with early syphilis treated with a single dose of benzathine penicillin, late latent syphilis treated with three once-weekly doses, and neurosyphilis treated with high dose intravenous penicillin [14]. To account for enhanced risk of neurosyphilis with lower CD4 counts [275, 276] and CD4 count-based recommendations

110 98 for lumbar puncture upon syphilis diagnosis [14], we modeled progression of HIV infection, as measured by changes in viral load and CD4 level. Antiretroviral therapy (ART) was initiated when CD4 count was 500 cells/ml [291]. We did not model interruptions in ART. In the absence of ART, men were assumed to experience a decline in CD4 count. Costs and Health Outcomes Costs are presented in Table 4-2, and included testing, treatment and follow-up costs, as well as lifetime costs for untreated individuals who progressed to tertiary disease. We used Ontario-specific testing and treatment costs [292, 293]. Treatment costs depended on disease stage and included medication and physician costs, and costs of follow-up visits and testing, with the number of follow-up tests based on Canadian treatment guidelines [14]. Adverse event costs included a general practitioner visit for post-dural headache and hospitalization for anaphylaxis. All costs were converted to year 2011 Canadian dollars using the health and personal care component of the Canadian Consumer Price Index [294]. In the absence of Canadian data about lifetime costs associated with tertiary syphilis, we used costs derived from the United States [295], which were adjusted for inflation and converted to 2011 Canadian dollars. Health outcomes measured intermediate disease-specific outcomes and changes in quality adjusted survival (Table 4-2). We used previously published estimates of the quality-of-life of individuals with HIV [286], disutility associated with primary, secondary, tertiary, and neurosyphilis, and the adverse outcomes associated with lumbar puncture and treatment [ ].

111 99 Table 4-2. Model costs and utilities. Variable Details Base Case Value Range Source Costs (2011 CDN$) Diagnostic tests (including labour) Screening test Public Health Ontario Laboratories (PHOL); Ontario Ministry of Health and Long- Term Care (MHLTC)[293] Confirmatory test PHOL; MHLTC [293] Lumbar puncture PHOL; MHLTC [293] Treatment a Early syphilis MHLTC [292]; Assumption Late latent syphilis MHLTC [292]; Assumption Neurosyphilis MHLTC [292]; Assumption Tertiary MHLTC [292]; Assumption Adverse events b Post-dural headache Fisman et al. [300]; Assumption Anaphylaxis Fisman et al. [300]; Assumption Lifetime cost of tertiary or neurosyphilis (excluding treatment) Owusu-Edusei et al. [295]

112 100 Variable Details Base Case Value Range Source Utilities Base case, HIV-infected individual on ART Base case, HIV-infected individual, asymptomatic, not on ART Sanders et al. [286] Sanders et al. [286] Syphilis disutility Primary syphilis Kwong et al. [296]; WHO [297] Secondary syphilis Kwong et al. [296]; WHO [297] Neurosyphilis and tertiary syphilis (per year) Kwong et al. [296]; WHO [297] Lumbar puncture disutility Procedure Ward et al. [301]; Assumption Post-dural headache Ward et al. [301]; Assumption Disutility of treatmentassociated anaphylaxis Pepper and Owens [298]

113 101 Variable Details Base Case Value Range Source Other Variables Cycle length (mo) 1 Assumption Discount rate (%) CADTH [302] a Base case treatment cost for neurosyphilis was calculated as the weighted average of inpatient and outpatient treatment, assuming 95% of cases require inpatient treatment. Base case treatment cost for tertiary syphilis was calculated as the weighted average of gummatous and cardiovascular syphilis, assuming 62% of tertiary syphilis cases are gummatous and 38% are cardiovascular [99]. Cardiovascular cases were further subdivided into those requiring surgery (20%) and those not requiring surgery [303]. b Post-dural headache was assumed to require a general practitioner visit. Anaphylaxis was assumed to require hospitalization.

114 102 Model Validation and Simulations To ensure that our model was reproducing observed trends in reported syphilis infection, we compared the expected incidence of diagnosed early neurosyphilis in our modeled cohort under the usual care scenario to estimated rates of neurosyphilis between 2008 and 2012 [241] in Toronto s HIV-infected male population [254]. Our base case analysis was performed as a first-order Monte Carlo simulation of 500,000 individuals assigned to each strategy. To account for uncertainty surrounding parameter estimates we performed probabilistic sensitivity analyses using second-order Monte Carlo simulations, such that parameters were sampled from distributions for each of 1000 simulated trials; each trial included 1000 identical individuals with costs and outcomes projected in stochastic (first-order) simulations. We used gamma distributions for costs and beta distributions for probabilities and utilities [304], with plausible ranges based on 95% confidence intervals, lowest and highest published values, or by varying inputs by ±50%, depending on available data. We performed additional sensitivity analyses with alternate assumptions around treatment probabilities, syphilis incidence, infection risk period, program duration, compliance with higher coverage strategies, and CD4 count at initiation of ART.

115 Results Model Validation and Projected Effectiveness In the absence of enhanced (more frequent and/or higher coverage) screening, approximately 2.9 per 10,000 men were projected to be diagnosed with neurosyphilis annually, which is within the range of rates of (mean 4.5) per 10,000 HIVinfected males in Toronto between Compared to usual care, higher (100%) coverage 3-monthly screening was projected to reduce incidence of diagnosed early neurosyphilis (0.9 per 10,000 person-years), while increasing the diagnosis of early syphilis (from 3.7 to 4.7 per 100 person-years). Unnecessary treatment of uninfected men due to false positive test results also increased (from 0.5 to 3.5 per 100 personyears) (Figure 4-3). The additional strategies resulted in infection incidence intermediate between usual care and higher coverage 3-monthly strategies. Cost Effectiveness Compared to the usual care (57% coverage, annual) strategy, higher (100%) coverage 3-monthly screening was projected to cost more ($98.69) and be more effective (0.015 QALY) (Table 4-3). All intermediate strategies cost less than the usual care strategy, but provided fewer QALY gains than higher coverage 3-monthly screening. The differences in QALYs between strategies were minimal, reflecting the fact that most health consequences of latent syphilis occur years or decades after initial infection. To capture second-order uncertainty, we performed 1000 simulated trials, each with 1000 participants randomly assigned to each screening strategy. The preferred strategy varied depended on the willingness-to-pay threshold (Figure 4-4). Assuming a willingness-to-pay (WTP) of $0 per QALY, higher coverage screening every 3 or 6 months was most frequently preferred. At a WTP threshold of $50,000 or $150,000/QALY, preferences for individual strategies were less clear, although in general higher coverage strategies were preferred to lower coverage strategies of the same frequency, and there was an approximately linear relationship between increased screening frequency and likelihood that a strategy would be preferred.

116 104 Figure 4-3. Model validation and projections. Model estimated diagnoses of early neurosyphilis, early (primary, secondary, and early latent) syphilis, and false positive cases. Reported values represent the average rates in the modeled cohort over a 20-year period for the different strategies evaluated. Neurosyphilis infections are plotted x100 for comparability. Usual care annual represents model projections based on current estimates of screening coverage and frequency among HIV-infected MSM under medical care. Toronto HIV-infected men represents estimated rates of diagnosed early neurosyphilis among HIV infected men living in Toronto (average for the years , error bars represent 95% confidence intervals). Usual care, usual 6 months, and usual 3 months refer to screening 57% of the population every 12, 6, or 3 months, respectively. Higher coverage annual, 6, months and 3 months refer to screening 100% of the population every 12, 6, or 3 months, respectively.

117 105 Table 4-3. Discounted health and economic outcomes associated with different syphilis screening strategies. Strategy a Discounted Cost b (CDN $) Incremental Cost (CDN $) Discounted Life Expectancy (y) b Discounted Effectiveness b (QALY) Incremental Effectiveness (QALY) ICER ($/QALY) Higher coverage, 6 months Higher coverage, annual Dominated Usual, 6 months Dominated Usual, 3 months Dominated Usual care Dominated Higher coverage, 3 months , Abbreviations: QALY, quality-adjusted life year; ICER, incremental cost-effectiveness ratio a Higher coverage, 100% coverage; Usual, 57% coverage b Discounted at 5%

118 106 Figure 4-4. Strategy acceptability for different willingness-to-pay thresholds. The frequency with which each strategy was optimal at willingness-to-pay thresholds of 0, $50,000, or $150,000 per QALY is shown for 1000 probabilistic trials with 1000 individuals assigned to each strategy within each trial. Usual care, usual 6 months, and usual 3 months refer to screening 57% of the population every 12, 6, or 3 months, respectively. Higher coverage annual, 6, months and 3 months refer to screening 100% of the population every 12, 6, or 3 months, respectively.

119 107 Sensitivity Analysis As uncertainty in model parameters was incorporated into second order Monte Carlo simulations, we performed additional sensitivity analyses using alternate assumptions about disease epidemiology, natural history and programmatic features (Table 4-4). As we reduced syphilis risk, we found that usual care became a viable strategy when risk of infection declined to < 20% of baseline risk, and even in this circumstance, higher coverage 6-monthly screening remained a highly cost-effective strategy ($39,096 per QALY). We identified no other circumstance under which usual care was a non-dominated strategy. Similarly, usual screening coverage strategies with increased frequency of screening seldom emerged as viable strategies. We found that usual coverage screening at a frequency of every six months emerged as a potentially viable strategy only when the probability of treatment of false-positive tests became equivalent to the probability of treating true-positive syphilis (ICER for usual coverage screening every 6 months $13,807 per QALY relative to higher coverage annual screening). In other sensitivity analyses, higher coverage strategies dominated usual coverage strategies (i.e., cost less and provided greater health benefit) but the optimal screening interval was sensitive to input parameters. In our base case we assumed that treatment with combination ART would begin when CD4 count reached 500 cells/ml; when we assumed ART at a CD4 count of 350 cells/ml, higher coverage every 3 months became costly relative to every 6 months (ICER $130,834 per QALY) but might still be considered cost-effective in high income countries [238]. Higher coverage every 6 months was the preferred strategy when duration of infection risk was varied from 10 to 30 years (assuming a WTP of $50,000/QALY), regardless of whether screening was halted once syphilis acquisition risk stopped or continued beyond the infection risk period.

120 108 Table 4-4. Preferred syphilis screening strategies under alternate model assumptions and for different willingness-to-pay thresholds. Preferred Strategy a by Willingness-to-Pay Threshold ($/QALY) Variable Details 0 50, ,000 Base case b HC-6 HC-6 HC-3 Probability of treating false positive cases among previously infected men Probability of treating false and true positive cases among previously infected men Syphilis incidence 0 HC-6 HC annual HC HC annual HC annual HC annual 0.95 HC annual Usual 6 HC HC-3 HC-3 HC HC-6 HC-3 HC HC annual HC-6 HC-6 2-fold increase HC-6 HC-6 HC-6 5-fold increase HC-3 HC-3 HC-3 2-fold decrease HC annual HC-6 HC-6

121 109 Preferred Strategy a by Willingness-to-Pay Threshold ($/QALY) Variable Details 0 50, ,000 5-fold decrease Usual care HC-6 HC-6 Linear decrease from current level to 0 over 20 years HC annual HC annual HC annual CD4 count at which initiate ART 350 cells/ml HC-6 HC-6 HC-3 Uptake of higher coverage (HC) screening 60-90% HC-6 HC-6 HC-6 Duration of infection risk/duration of screening program (yr) 10/10 HC-6 HC-6 HC-6 10/20 HC annual HC-6 HC-6 10/30 HC annual HC-6 HC-6 20/30 HC annual HC-6 HC-6 30/30 HC-6 HC-6 HC-6 a HC annual, -6, and -3 refer to higher coverage screening (i.e., 100%) of the population every 12, 6, or 3 months, respectively (except for the analysis where uptake was varied from 60-90%). Usual care and usual 6 refer to screening 57% of the population every 12 or 6 months, respectively. b In the base case, probability of treating a false positive case with prior history of syphilis infection was 0.2; probability of treating a false positive case with no prior history of syphilis infection and all true positive cases was 0.95; syphilis incidence was 4 per 100 person-years (py) in never infected men and 4.8 per 100 py in previously infected men; ART was initiated when CD4 count was <500 cells/ml; uptake of higher coverage screening was 100%; and duration of infection risk and duration of screening program were both 20 years.

122 110 When we assumed equal probability of treatment of true and false positive cases among men with a history of previous treated syphilis, higher coverage was the preferred strategy, with frequency of screening depending on the assumed level of treatment. Decreasing coverage in the higher coverage strategies to 60-90% resulted in higher coverage every 6 months dominating all of the other strategies. When costs and health outcomes were not discounted, higher coverage every 6-months dominated all other strategies except for 3-monthly higher coverage (ICER $239,539 per QALY) (Table 4-5).

123 111 Table 4-5. Undiscounted health and economic outcomes associated with different syphilis screening strategies. Strategy a Cost (CDN $) Incremental Cost (CDN $) Life Expectancy (y) Effectiveness (QALY) Incremental Effectiveness (QALY) ICER ($/QALY) Higher coverage, 6 months Higher coverage, annual Dominated Usual, 3 months Dominated Usual, 6 months Dominated Higher coverage, 3 months ,539 Usual care Dominated Abbreviations: QALY, quality-adjusted life year; ICER, incremental cost-effectiveness ratio a Higher coverage, 100% coverage; Usual, 57% coverage

124 Discussion We project that a universal syphilis screening program in HIV-infected MSM under medical care has the potential to improve health and save costs, relative to usual care. However, increased screening may result in unnecessary treatment of men with false positive tests and lead to excess adverse events associated with testing and treatment (e.g., adverse consequences of lumbar puncture). Although previous studies have examined the disease dynamic impacts of screening frequency, [224, 305] with projections that more frequent screening would reduce syphilis burden in MSM, to our knowledge, this model is the first to evaluate the cost-effectiveness of more frequent and higher coverage syphilis screening in a high-risk group. Perhaps most importantly, we project that both increases in test frequency and coverage for MSM with HIV infection would increase effectiveness relative to current standard of care, and that all more-intense screening regimens (with the exception of higher coverage, 3-monthly screening) would decrease, rather than increase, net healthcare costs due to aversion of downstream sequelae of untreated syphilis infections. Although the projected increase in effectiveness observed with enhanced (higher coverage and/or increased frequency) screening was small compared to usual care, this reflects the fact that the major health consequences of syphilis occur years, and often decades, after initial infection [4, 18], and consequently the health impact of screening is reduced by discounting. We restricted our analysis to HIV-positive MSM because surveillance data show this to be the population at greatest risk of syphilis in much of North America [66, 306, 307]. Men currently under medical care for HIV present an ideal target for this intervention, since syphilis screening can be included with existing blood-work with minimal inconvenience and expense [166]. By implementing an opt-out syphilis screening test, we would expect screening frequency in HIV-positive MSM to increase [166]. The major concern with implementing such a strategy is the management of false positive test results in individuals with a history of previously treated syphilis infection

125 113 and the resulting physician burden [308]. With more frequent RPR testing, clinicians will have a better history of an individual s titre, assisting with the interpretation of titre changes and helping to distinguish new infections from past cases (in particular serofast cases, whose treponemal titres remain unchanged despite treatment). Appropriate management of individuals with previous treated syphilis infection would be crucial for minimizing unnecessary treatment and the development of clear guidelines about how to manage test results would be critical. Our study is subject to several limitations. Our estimates of syphilis acquisition risk are based on HIV-positive MSM in Ontario [83], our target population, and are somewhat higher than reported estimates in MSM in other jurisdictions [50, 66, 309]. Nonetheless, we found our projections of cost-effectiveness to be robust in the face of variation in syphilis risk, and elevated syphilis risk in MSM is currently widespread in North America and Europe [68, 241]. It is important to note that syphilis epidemics tend to occur in waves separated by periods of low incidence [79], and during the latter periods the screening strategies we identify as cost-effective in this analysis may cease to be attractive. As discussed above, interpretation of titre changes is important for individual case management; we did not include treatment algorithms for serofast cases in the model, due to the associated complexity. Our model included simplifying assumptions and incorporates parameter values that are subject to uncertainty, but our findings were robust in the face of wide-ranging sensitivity analyses and alternate assumptions. Lastly, we used a traditional static risk Markov model to evaluate the healtheconomic attractiveness of syphilis screening in MSM. Such models are limited in their ability to capture the indirect effects of disease treatment and prevention that are characteristic of communicable disease control programs [180], including not only reductions in the future stream of syphilis infections but also the impact of untreated syphilis infections on the transmission of HIV [85]. However, our previously published work on screening effectiveness in a dynamic, agent-based model [305] similarly identified an increase in frequency of screening as a means of reducing syphilis incidence. Work remains for the quantification of the economic attractiveness of syphilis

126 114 screening strategies in a manner that includes both direct and indirect effects of screening. In summary, we have shown that when rates of syphilis acquisition are high, implementing routine syphilis screening in MSM currently under medical care for HIV is expected to be a highly cost-effective intervention. This strategy has been implemented in Australia and has been demonstrated to increase the detection of early asymptomatic syphilis [166], but the long-term effectiveness and cost-effectiveness of this intervention has not been evaluated. While a clinical trial would be an ideal means of testing our model projections, the policy-relevant time horizon would likely exceed the attainable duration of such a trial. Our model provides an estimation of the potential impact of an enhanced syphilis screening program among HIV-positive MSM that can help guide policy decisions.

127 Chapter 5: Go big or go home: impact of screening coverage on syphilis infection dynamics The work described in this chapter is published as: Tuite AR and Fisman DN. Go big or go home: impact of screening coverage on syphilis infection dynamics. Sexually Transmitted Infections doi: /sextrans

128 Abstract Background: Syphilis outbreaks in urban men who have sex with men (MSM) are an ongoing public health challenge in many high-income countries, despite intensification of efforts to screen and treat at-risk individuals. We sought to understand how population-level coverage of asymptomatic screening impacts the ability to control syphilis transmission. Methods: We developed a risk-structured deterministic compartmental mathematical model of syphilis transmission in a population of sexually active MSM. We assumed a baseline level of treatment of syphilis cases due to seeking medical care in all scenarios. We evaluated the impact of sustained annual population-wide screening coverage ranging from 0% to 90% on syphilis incidence over the short-term (20-years) and at endemic equilibrium. Results: The relationship between screening coverage and equilibrium syphilis incidence displayed an inverted U-shape relationship, with peak equilibrium incidence occurring with 20-30% annual screening coverage. Annual screening of 62% of the population was required for local elimination (incidence <1 case per 100,000 population). Results were qualitatively similar in the face of differing programmatic, behavioural, and natural history assumptions, although the screening thresholds for local elimination differed. With 6-monthly or 3-monthly screening, the population coverage required to achieve local elimination was reduced to 39% or 23%, respectively. Conclusions: Although screening has the potential to control syphilis outbreaks, suboptimal coverage may paradoxically lead to a higher equilibrium infection incidence than that observed in the absence of intervention. Suboptimal screening program design should be considered as a possible contributor to unsuccessful syphilis control programs in the context of the current epidemic.

129 Introduction Despite the dramatic decline in incidence following the introduction of penicillin in the 1940s, syphilis continues to cause outbreaks and epidemics. These have tended to occur in distinct populations over time [147, 310, 311]. In North America and Northern and Western Europe, an epidemic resurgence in syphilis, focused in men who have sex with men (MSM) has been ongoing since approximately 2001 [312, 313]. Elevated incidence continues to be seen in MSM in the face of significant public health responses, including public awareness campaigns, and intensified screening and partner notification efforts [248]. The reasons for syphilis reemergence and persistence are unclear. Among the possible epidemic drivers are: changes in sexual behaviour, including increased rates of new partner acquisition and concurrent partnerships [22], high rates of population turnover [149], and changes in the intensity of syphilis control programs [22, 149]. Treatment-induced interruption of the development of protective immunity to reinfection has also been proposed as a contributor to syphilis persistence, by replenishing the supply of susceptible individuals in a population [149]. This could suggest that the current MSM-focused syphilis epidemic is persisting because of, rather than despite, enhanced public health control measures aimed at curbing spread by treating infectious syphilis cases. Cases with untreated syphilis appear to be protected from re-infection, while those previously treated for syphilis can be re-infected [19]. However, despite clinical observations of repeat syphilis infections in treated cases [81, 142, 143], there is some evidence to suggest a degree of transient acquired immunity in individuals treated for latent syphilis [4, 140]. Both the development of immunity to syphilis re-infection and the durability of such a response, remain areas of uncertainty, with the development of an immune response appearing to depend on duration of infection prior to treatment [4, 140].

130 118 Previous modeling work [224, 305] has demonstrated that frequent and sustained screening of at-risk populations is required to control syphilis outbreaks. We hypothesized that when screening coverage is suboptimal, this may itself contribute to observed syphilis dynamics in outbreak situations. We used a mathematical model to evaluate the relationship between population-level coverage of syphilis screening and the ability to control syphilis transmission. We also explored how alternate assumptions about epidemic drivers might influence our model projections.

131 Methods Model Overview We developed a dynamic compartmental mathematical model of syphilis transmission in a population of sexually active men who have sex with men. An overview of how individuals transition through the model is presented in Figure 5-1. Model parameters (Table 5-1) were literature-derived where available, or estimated via model calibration. Additional model details are provided in Appendix B. Disease Natural History The population was divided into compartments representing different disease states: susceptible (S), exposed (E; infected but not infectious), primary syphilis (I1, I1_2), secondary syphilis (I2, I2_2), early latent syphilis (L1, L1_2), and late latent syphilis (L2). We did not model recurrent syphilis and assumed that only the primary and secondary stages were infectious. The model included two susceptible compartments one for infection-naïve individuals (S) and one for those with previously treated infection (Sr) to calculate seroprevalence, which was used for model calibration. Transmission of infection occurred through sexual contact between susceptible and infectious individuals. In the absence of treatment, infected individuals progressed through the various stages of disease and remained in the late latent stage until model exit. We assumed that untreated individuals were not susceptible to re-infection with different syphilis strains (i.e., no superinfection). As we did not model the health consequences of syphilis, we did not include tertiary syphilis.

132 120 Figure 5-1. Outline of model structure, showing population flows between compartments. The following health states are included: susceptible (S), susceptible with history of treated infection (Sr), incubating (E), primary syphilis (I 1, I1_2), secondary syphilis (I 2, I2_2), early latent syphilis (L 1, L1_2), late latent syphilis (L 2 ), and treated (T 1 - T 3 ). Transitions rates between health states are defined in Table 5-1. The early syphilis stages are divided into two compartments to allow for symptomatic individuals to seek treatment for infection at the midpoint of the infection stage (indicated by the grey dashed arrows). Individuals whose infection is treated following screening or inadvertent receipt of antibiotics can enter the treated state at any point during their infection (grey arrows). Individuals with incubating syphilis (E) who receive treatment are assumed to return to the susceptible (S) compartment, while all other treated individuals eventually enter the susceptible with history of treated infection (Sr) compartment. Individuals enter the model via the susceptible state and can exit by all states. The model is stratified by sexual activity group.

133 121 Table 5-1. Model parameter values. Parameter Symbol Value Range Source Average time in model (years) Transmission probability per sexual partnership Average duration of syphilis stage (months) 1/ μ 20 Assumption β 0.6 Garnett et al. [4] Garnett et al. [4] Incubating 1/δ 0.9 Primary 1/γ1 1.5 Secondary 1/γ2 3.6 Early latent 1/γ3 6.9 Background antibiotic treatment rate (per year) Time in treated (and immune to re-infection) state Primary and secondary pabx Assumption 1/λ1 1 week 1 week 1 year Garnett et al. [4]; assumption Early latent 1/λ2 1 week 1 week 5 years Late latent 1/λ3 5 years 1 week 5 years Proportion of population in sexual activity group Ni Remis et al. [64] Low 0.4 Moderate 0.4 High 0.2

134 122 Parameter Symbol Value Range Source Relative rate of partner acquisition per year (by activity group) rpi Remis et al. [64] Low 1 Moderate 5 High 20 Mean rate of partner acquisition (per year) Mixing between sexual activity groups Proportion of population screened for syphilis Screening interval (years) Probability of actively seeking medical care for infection cmean Calibration ε Calibration coverage Tuite et al. [305]; Assumption interval Assumption Calibration Primary ptrt Secondary ptrt Early latent ptrt3 0.25

135 123 Sexual Mixing The model included three levels of sexual activity: high, moderate, and low. The different activity levels were characterized by their relative rates of partner acquisition. The proportion of the population in each group was approximated based on reported partner numbers in a survey of urban MSM [64]. We assumed that individuals remained in a given activity group for the duration of their sexual lifespan. Mixing between risk groups was described by the term ε, which could range from 0 (proportionate or random mixing between groups) to 1 (assortative mixing, where individuals partner exclusively with individuals of the same risk class). The total population size was assumed to remain constant. Screening and Treatment Individuals with syphilis infections could receive treatment by actively seeking medical care for their infection, by participating in opportunistic screening programs, or through receipt of antibiotic treatment for another medical indication. Individuals were assumed to seek medical care because they were symptomatic or had an infected partner (i.e., partner notification). For individuals actively seeking treatment, treatment of infection was assumed to occur at the midpoint of the infection stage. We assumed a constant hazard of screening (α), with probability of screening converted to a rate, assuming an exponential distribution [314]: ln(1 coverage) interval where interval describes the average time between screening events (1 year in our base case), in years, and coverage represents the proportion of the population screened in each interval [315]. For simplicity, we assumed that all individuals had their infection status correctly identified, were successfully treated if infected, and screening and treatment were applied equally across all sexual activity groups. We assumed that there was a background rate of antibiotic use leading to inadvertent treatment of syphilis in some

136 124 individuals, who were moved to the appropriate treatment compartment but were not counted as diagnosed cases. Immunity from Re-Infection To examine the impact of different assumptions about how treatment influences the development of immunity, we included transiently immune states following treatment for primary and secondary (T1), early latent (T2), and late latent (T3) syphilis. Treated individuals passed through one of these immune states before returning to the susceptible state. In our base case we assumed that individuals were transiently immune only after treatment for late latent infection. Those treated for early infection (primary, secondary, or early latent infection) were not considered biologically immune to re-infection but dwelled in the immune state for one week after treatment as a means of representing abstinence from unprotected sexual activity after medical treatment. Model Calibration and Outcomes To capture the impact of the introduction of antibiotic treatment on syphilis dynamics, we allowed prevalent syphilis to reach endemic equilibrium following model initiation, followed by a 20-year period of treatment (including background antibiotic treatment). After this period, we assumed that opportunistic screening was increased rapidly once syphilis incidence increased above 1 case per 1000 person-years in the population. This marked the beginning of the 5-year calibration period. In addition to individuals seeking medical care, annual screening was assumed to reach 30% of the population during this period, based on estimates for the city of Toronto [305]. Mean rate of partner acquisition and the mixing parameter (ε) were estimated by calibration. We used data from the Canadian city of Toronto on early syphilis diagnoses in MSM, proportion of cases by stage (primary, secondary, or early latent) [241, 305] and syphilis seroprevalence [64] for model calibration, minimizing the sum-of-squares difference between model projections and data. Additional details about model calibration are provided in Appendix B.

137 125 After the calibration period, coverage of asymptomatic syphilis screening was changed, with population coverage ranging from 0 to 90%. We evaluated the impact of screening on both short-term (over a 20-year time horizon) and long-term disease trends. In the short-term, we examined syphilis incidence and diagnoses of early syphilis cases (i.e., cases that would be observed via public health surveillance). For long-term effects, we determined syphilis incidence at endemic equilibrium and the necessary screening conditions for syphilis elimination (defined as the reduction of incidence to <1 case per 100,000 population per year). Alternate Assumptions We examined the impact of different potential epidemic drivers by changing base case assumptions. For each alternate assumption, we re-calibrated the model using the same procedure as in the base case. Parameter values for each of the scenarios are presented in Table 5-2. For these analyses, we evaluated screening every 6 months or every 3 months in addition to annual screening. (i) Development of Immunity In the base case, the development of protective immunity was assumed to occur only after an individual entered the late latent syphilis stage. Individuals with early syphilis were thus assumed to immediately re-acquire biological susceptibility to infection following receipt of treatment (though as noted above would abstain from unprotected sexual activity for one week after treatment). We evaluated the impact on results if immunity developed during infectious syphilis or during early latent syphilis. We also varied the duration of immune protection following treatment for individuals in each stage.

138 126 Table 5-2. Parameter values used for different model assumptions, derived by model calibration. Assumption Rate of partner acquisition (per year) Mixing (ε) Base case Stage at which immunity develops and duration of immunity following treatment* No immunity Late latent, 1 year Early latent, 1 year Infectious syphilis, 1 year Early latent, 5 years Background antibiotic use (% per year) Mixing between sexual activity groups Assortative Random *Cases receiving treatment prior to entering the indicated stage of infection were assumed to be protected from re-infection for one week prior to returning to the susceptible state.

139 127 (ii) Sexual Mixing We examined how mixing between risk groups impacted syphilis dynamics. We used a previously described approach [201] to vary mixing between groups from random (proportionate mixing with sexual partners chosen in relation to the number of potential sexual partnerships supplied by people in a given activity class [201]) to assortative (sexual partnerships with individuals of the same activity group only). Individuals were assigned to a single sexual activity group and remained at this level for the duration of their sexual life. Sexual risk was defined by rate of partner acquisition. (iii) Antibiotic Use In our base case, we assumed that there was a low background rate of antibiotic use leading to inadvertent treatment of syphilis. We evaluated the effect of higher proportions of the population receiving antibiotics sufficient to treat syphilis on thresholds required for syphilis elimination.

140 Results Model Calibration The model reproduced cumulative early syphilis diagnoses reported among MSM in Toronto, Canada. While average incidence was reproduced well, the model was less effective at capturing year-to-year variability observed over the five-year calibration period (Figure 5-2). Base Case Sustained annual population screening at coverage levels ranging from 50% to 90% resulted in an initial transient increase in diagnosed early syphilis cases, relative to continued screening at the 30% level used during the calibration period (Figure 5-3). Removing screening resulted in an initial decline and subsequent increase in diagnosed cases. By contrast, total (diagnosed and undiagnosed) incident cases were projected to immediately decrease with higher coverage levels and increase in the absence of screening. At the end of the 20-year period, both diagnosed and total incident cases were projected to be higher with intermediate levels of screening (30% and 50%) than those observed in the absence of screening. When screening reached a relatively large proportion of the population ( 70%), syphilis incidence and diagnoses declined over the 20-year period. Note that in all scenarios, including the no screening scenario, treatment of syphilis could still occur by cases actively seeking medical care or by receipt of antibiotics for another reason leading to inadvertent syphilis cure. We also evaluated syphilis incidence at endemic equilibrium assuming continued screening at different coverage levels. Consistent with the observed short-term dynamics, we observed an inverted U-shape relationship between screening coverage and equilibrium syphilis incidence (Figure 5-4). Peak equilibrium incidence was projected to occur when 20-30% of the population was screened annually. Greater than 60% of the population needed to be screened annually to achieve local elimination.

141 129 Figure 5-2. Model calibration. Model projected (A) annual and (B) cumulative incidence of diagnosed cases of primary, secondary, and early latent syphilis cases compared to estimates of diagnoses in men who have sex with men over a 5-year period in Toronto, Canada ( ).