Department of Epidemiology, School of Public Health and Community Medicine, University of Washington, Seattle, WA. 3

Transcription

1 American Journal of Epidemiology Copyright 2004 by the Johns Hopkins Bloomberg School of Public Health All rights reserved Vol. 160, No. 4 Printed in U.S.A. DOI: /aje/kwh220 Influence of Study Population on the Identification of Risk Factors for Sexually Transmitted Diseases using a Case-Control Design: The Example of Gonorrhea Lisa E. Manhart 1,2, Sevgi O. Aral 3, King K. Holmes 1,2, Cathy W. Critchlow 2, James P. Hughes 4, William L. H. Whittington 1, and Betsy Foxman 5 1 Department of Medicine, School of Medicine, University of Washington, Seattle, WA. 2 Department of Epidemiology, School of Public Health and Community Medicine, University of Washington, Seattle, WA. 3 National Center for HIV, STD and TB Prevention, Centers for Disease Control and Prevention, Atlanta, GA. 4 Department of Biostatistics, School of Public Health and Community Medicine, University of Washington, Seattle, WA. 5 Department of Epidemiology, School of Public Health, University of Michigan, Ann Arbor, MI. Received for publication August 7, 2003; accepted for publication March 2, The population prevalence of many sexually transmitted diseases (STDs) is low. Thus, most epidemiologic studies of STDs are conducted among STD clinic populations to maximize efficiency. However, STD clinic patients have unique sociobehavioral characteristics. To examine the potential effect of study population on identification of risk factors, the authors compared 1) STD clinic patients with a random digit dialing telephone sample, 2) general population cases with random digit dialing controls, and 3) STD clinic cases with STD clinic controls (Seattle, Washington, ). Risk factors for gonorrhea identified among STD clinic patients formed a subset of those identified in the general population. In both populations, risk decreased with age (odds ratio for the general population (OR GP ) = 0.4, 95% confidence interval (CI): 0.22, 0.59; odds ratio for the clinic population (OR clinic ) = 0.5, 95% CI: 0.30, 0.81) and was increased among Blacks (OR GP = 15.5, 95% CI: 4.93, 49.0; OR clinic = 10.5, 95% CI: 4.51, 24.68) and persons whose partner had been jailed (OR GP = 5.4, 95% CI: 2.07, 13.9; OR clinic = 3.1, 95% CI: 1.32, 7.30). Additional factors associated with gonorrhea in the general population included secondary education (OR = 0.3, 95% CI: 0.11, 0.70), anal intercourse (OR = 10.5, 95% CI: 2.01, 54.7, STD history (OR = 5.9, 95% CI: 1.76, 19.5), meeting partners in structured settings (OR = 0.2, 95% CI: 0.09, 0.50), no condom use (OR = 3.2, 95% CI: 1.30, 7.89), and divorce (OR = 3.6, 95% CI: 1.07, 11.9). Risk factors identified in STD clinics will probably be confirmed in a general population sample, despite overcontrolling for shared behaviors; however, factors associated with both disease and STD clinic attendance may be missed. case-control studies; gonorrhea; risk factors; sexually transmitted diseases Abbreviations: RDD, random digit dialing; STD(s), sexually transmitted disease(s). In 2000, the prevalences of Chlamydia trachomatis and Neisseria gonorrhoeae in the United States were cases per 100,000 population and 132 cases per 100,000 population, respectively (1). Given these relatively low prevalences and the invasive clinical examinations often required for case ascertainment, most studies of the epidemiology of sexually transmitted diseases (STDs) draw cases from STD clinics to facilitate the rapid, efficient accumulation of adequate numbers of infected persons. Nondiseased controls are often drawn from the same clinics, both for reasons of convenience and to assure that controls are comparable to cases in terms of potential STD exposure. However, persons visiting STD clinics represent a subset of the general population and are characterized by different demographic and sociobehavioral factors (2); thus, risk factors identified in STD clinic populations may differ from those identified in the general population. Risk factors for infectious diseases provide clues to the underlying transmission dynamics and often constitute the basis for development of prevention interventions. Gonococcal infection presents an ideal example of the challenges present in descriptive studies of STD. Risk factors Correspondence to Dr. Lisa Manhart, Department of Epidemiology, School of Public Health and Community Medicine, University of Washington, 325 9th Avenue, Box , Seattle, WA ( lmanhart@u.washington.edu). 393

2 394 Manhart et al. commonly demonstrated for gonorrhea include young age, African-American race, low socioeconomic status, early onset of sexual activity, single marital status, a history of prior gonococcal infection, and same-sex activity among men (1, 3), whereas condom use has been found to be protective (4, 5). In response to these findings, STD prevention programs often target these segments of society and promote condom use. However, most of these risk factors have been identified among persons visiting STD clinics, raising the question of their generalizability. Despite the preponderance of STD clinic studies, population-level data do exist. The National Health and Nutrition Examination Survey (6), the National Health and Social Life Survey (7), the National Survey of Family Growth (8), the National Longitudinal Study of Adolescent Health (9), and the National Survey of Adolescent Males (10) in the United States, the National Survey of Sexual Attitudes and Lifestyles (11) in the United Kingdom, and demographic and health surveys carried out in many other countries have gathered information on sexual behavior to a greater or lesser extent from nationally representative general population samples and have reported prevalences of gonorrhea of percent. However, despite the availability of such data, comparisons of STD clinic populations and general populations remain limited to characteristics such as number of reported sex partners, number of concurrent partners, and patterns of condom use (12, 13), without concomitant laboratory diagnoses of STD. The case-control design is ideally suited to the investigation of multiple risk factors for rare conditions, but these studies are often criticized because of the opportunity for bias to arise in the selection of cases, controls, or both (14). Despite tacit acknowledgment that this opportunity for selection bias may affect the results of case-control studies conducted in STD clinics, to our knowledge, no empirical investigations that evaluate such bias exist. Therefore, to examine the effect of study population selection on sociobehavioral risk factors identified for gonorrhea, we first compared characteristics of an STD clinic population with those of a general population sample obtained from a random digit dialing (RDD) telephone survey. We then performed two separate case-control analyses to identify and compare risk factors for prevalent infection: 1) a general population sample of gonorrhea cases identified in a reference laboratory and an STD clinic compared with general population controls (RDD respondents) and 2) gonorrhea cases identified in an STD clinic compared with controls without gonorrhea from the same STD clinic. MATERIALS AND METHODS Study populations Cross-sectional STD clinic population. From 1992 to 1995, persons aged years seeking care at the Public Health Seattle-King County, Washington, STD clinic (Harborview Medical Center) were recruited as previously described (12). In brief, a research clinician reviewed patient charts during randomly assigned time blocks, and the first patient meeting the enrollment criteria was invited to participate. If the person refused (refusal rate = 37 percent), the next eligible person was approached. Consenting participants were interviewed regarding demographic and behavioral characteristics, sexual behavior, and characteristics of the four most recent sex partners within the past 90 days. Persons who had had same-sex partners during the past 90 days and those who could not understand spoken English were excluded. RDD population. In February 1995, a telephone survey was conducted in a sample of 544 Seattle residents. Extensive details on the sampling and interviewing process have been published previously (15). Eligibility criteria included current residence in Seattle, age years, and fluency in English. Potential respondents were told that the purpose of the study was to learn how often people engage in behaviors that might put them at risk of acquiring an STD. The survey instrument was based on that used in the Public Health Seattle-King County STD clinic and included questions on sexual history, partner and partnership characteristics (the two most recent partners and the first lifetime sexual partner), STD history, and demographic characteristics. It was administered using computer-assisted telephone interviewing software, with a cooperation ratio of 67 percent (number of completed interviews to the total number of completed and partially completed interviews and refusals). Gonorrhea case population. The total case group consisted of 103 cases of gonococcal infection; 61 were recruited from the STD clinic and 42 from the University of Washington Center for AIDS and STD Neisseria Reference Laboratory. From the STD clinic, in addition to 18 persons with gonococcal infection identified in the cross-sectional sample described above, 15 infected persons were recruited by research interviewers among patients diagnosed during their visit for gonococcal infection; 16 persons were recalled for therapy because of a positive test; and 12 were recruited from among contacts of gonorrhea cases. For persons diagnosed by clinical providers outside the STD clinic, we selected a random sample (n = 42) of persons whose N. gonorrhoeae isolates were supplied to the Neisseria Reference Laboratory, which received isolates from approximately 80 percent of all reported cases in Seattle-King County (16) during the study period. Cases were subject to the same exclusion criteria as the STD clinic cross-sectional population and were interviewed using the same questionnaire. For all groups, these analyses were restricted to sexually active heterosexuals aged years, and only the most recent heterosexual partnership within the past 90 days was included in the assessment of partnership and mixing characteristics. This yielded 373 persons in the STD clinic crosssectional population, 376 persons in the RDD population, 103 cases of gonococcal infection, and 332 STD clinic controls (the cross-sectional population excluding cases of gonorrhea and persons missing a gonorrhea test result). Because the data collection instrument for the RDD survey was based on that used in the STD clinic survey, most measures were identical. Measures for which comparable variables for the three groups could not be created were excluded. The Human Subjects Division of the University of Washington approved the study procedures for all three study groups.

3 Study Population and Identification of STD Risk Factors 395 Statistical methods Categorical characteristics of the RDD and STD clinic samples were compared using Pearson s chi-squared tests; the Wilcoxon rank-sum test was used to assess differences in continuous variables. The same statistical methods were used in case-control comparisons of 1) all gonorrhea cases with RDD controls (comparison A) and 2) the subset of gonorrhea cases identified in the STD clinic with controls recruited from the STD clinic (comparison B). We performed multivariable logistic regression analyses to identify independent predictors of gonococcal infection in the two case-control comparisons (A and B). Because the proportion of gonorrhea cases recruited from the STD clinic in our general population proxy sample (59.2 percent) differed from the proportion of all reported gonorrhea cases that were identified in STD clinics during the enrollment period (34.6 percent), we performed weighted logistic regression analysis for the general population case-control comparison (comparison A). Interaction terms were tested with a likelihood ratio test, and the software package Stata (17) was used for all data analyses. RESULTS Comparison of the STD clinic and RDD populations The STD clinic sample consisted of more men (55 percent) than women, while the reverse was true for the RDD sample (54 percent women) (p = 0.02; table 1). Each of the demographic and sexual behavior characteristics examined differed significantly between the two samples, and, with the exception of race, these differences were similar for men and women. STD clinic patients were significantly younger than RDD survey respondents (despite being from the same age range), had lower median annual income and education levels, were more likely to be single or to have spent a night in jail, and were less likely to be employed. Among women, STD clinic patients were more often White than RDD survey participants, but among men, STD clinic patients were less often White. Persons visiting the STD clinic were younger at first sexual intercourse and reported a greater number of lifetime partners, a higher rate of partner change, more concurrent partnerships, and a more diverse repertoire of sexual activities. Over 40 percent of male and female STD clinic patients reported engaging in anal intercourse, as compared with only 5 8 percent of RDD respondents. Men and women who had visited the STD clinic were more likely than those in the RDD population to report a history of any STD, and among STD clinic patients, nine (4.6 percent) of 197 men and seven (4.2 percent) of 168 women were infected with N. gonorrhoeae. While all individual and partnership characteristics (describing events in a relationship) differed between the two populations, only some of the mixing characteristics (describing mixing between population subsets) differed. The percentages of sex partners who differed from participants with respect to age, education, and sexual activity class were similar for the STD clinic population and the RDD population (table 2). However, compared with RDD respondents, STD clinic patients were significantly less likely to have had a partner of a different race. They were more likely to have a partner who had spent a night in jail; to have met a partner in an unstructured environment (e.g., parties, bars/ nightclubs, on the street, on a vacation or business trip, or at a health club); to report a shorter time interval between first meeting and first sexual intercourse; to have relationships of shorter duration; and to believe that a partner had had other sexual partners. Although frequencies of condom use during the relationship were similar for the two groups, when data were stratified by marital status, unmarried STD clinic patients reported using condoms significantly less often than did unmarried RDD respondents (39 percent vs. 27 percent; p < 0.001). Comparison of gonorrhea case groups Of the 103 cases of laboratory-confirmed N. gonorrhoeae infection, 61 patients were recruited directly through the STD clinic and 42 through the Neisseria Reference Laboratory, which obtains isolates from 80 percent of all gonorrhea cases diagnosed in King County. The two different case groups reported similar individual and partnership characteristics, except for age at first sexual intercourse (the median age was 15 years for STD clinic cases vs. 13 years for laboratory cases; p = 0.05) and rate of partner change (median of 2.8 partners per year for STD clinic cases vs. 1.4 partners per year for laboratory cases; p = 0.05). In comparison with cases recruited from the reference laboratory, STD clinic cases also reported somewhat more lifetime partners (a median of 17.5 partners vs. 10 partners; p = 0.09), a history of herpes (12 percent vs. 2 percent; p = 0.09), and shorterterm relationships (median duration of 3 months vs. 7 months; p = 0.07) and were somewhat less likely to believe that their partner had had other partners (36 percent vs. 54 percent; p = 0.09); however, these differences were not statistically significant. Median income did not differ significantly between the two case groups, and all cases had an annual household income of less than $9,375 (the lowest quartile of the income categories established for the RDD group in previous analyses). Given the similarity of the two groups of persons with gonorrhea, they were combined in subsequent general population analyses. Comparison of case-control results In case-control comparison A, which served as a proxy for the general population, the case group comprised all 103 gonorrhea cases; the 376 RDD survey respondents comprised the control group. Case-control comparison B examined the subset of persons who sought care at STD clinics (61 cases and 332 controls). For case-control comparison A (general population proxy), cases differed significantly from controls with regard to virtually all individual characteristics (table 3). Similar to comparison A, in case-control comparison B (STD clinic patients), cases and controls also differed with regard to gender, age, education, race, having spent a night in jail, rate of partner change, and history of anal intercourse. However, in case-control comparison B, there was no significant difference between cases and controls with respect to

4 396 Manhart et al. TABLE 1. Comparison of characteristics of sexually active heterosexuals visiting a sexually transmitted disease clinic in Seattle, Washington, in with characteristics of Seattle residents participating in a random digit dialing survey in 1995 Characteristic STD* clinic (n = 204) Males RDD* (n = 173) p value STD clinic (n = 169) Females RDD (n = 203) p value No. % No. % No. % No. % Demographic characteristics Median age (years) 26 (18 39) 30 (18 39) < (18 39) 30 (18 39) <0.01 Median annual income (dollars) 7,200 (0 120,000) 18,000 (0 500,000) <0.01 5,400 (0 45,600) 17,250 ( ,000) Education of high school or more < Race White Black Other < <0.001 Single marital status < <0.001 Employed < <0.01 Ever spent a night in jail < Sexual behavior characteristics Median age (years) at first sexual intercourse 15 (4 25) Median lifetime no. of sexual partners 20 (1 900) Median rate of partner change# 2.0 ( ) 17 (8 31) 9 (1 250) 0.8 ( ) < (9 31) < (1 600) < ( ) 17 (10 33) 6 (1 2,000) 0.7 ( ) Concurrency <0.001 Douching <0.001 History of anal intercourse < <0.001 History of any STD** < <0.001 Gonorrhea (culture) 9/ / Chlamydia (culture) 14/ / * STD, sexually transmitted disease; RDD, random digit dialing. p value for Pearson s chi-squared test unless otherwise specified. Numbers in parentheses, range. p value for Wilcoxon rank-sum test. Adjusted for family size. # Rate of partner change was defined as the average number of new partners acquired per year (lifetime number of partners/number of years sexually active). ** Any STD included nongonococcal urethritis, pelvic inflammatory disease, gonorrhea, chlamydia, syphilis, genital warts, and herpes. Number positive/number of cultures. <0.01 <0.01 <0.01 <0.01 income, marital or employment status, age at first sexual intercourse, lifetime number of partners, concurrency (overlapping sex partners), douching, or history of STD. Mixing and partnership characteristics for both case-control comparisons differed less than individual characteristics. For comparison A, these differences included a history of the partner having spent a night in jail, the sexual activity class difference, the place in which the partner was met, the interval between first meeting the partner and sexual intercourse, the duration of the partnership, and a belief that the partner had had other partners (general population proxy). Among the subset of persons visiting the STD clinic (comparison B), even fewer differences in partnership characteristics between cases and controls were observed. Race difference, the partner having spent a night in jail, and the belief that the partner had had other partners were the only factors that differed. Multivariate analysis In multivariate logistic regression models adjusting for gender and all other factors included in the model, the risk

5 Study Population and Identification of STD Risk Factors 397 TABLE 2. Mixing and partnership characteristics of the most recent partnership among participants in a case-control study who were recruited through sexually transmitted disease clinics and random digit dialing, Seattle, Washington, Characteristic Sexually transmitted disease clinic (n = 373) Random digit dialing (n = 376) p value* No. % No. % Mixing characteristics Race difference Age difference Male partner >5 years older Male partner 2 years younger to 5 years older Male partner 3 years younger Education difference Same level of education Male more educated than female Female more educated than male Partner had spent a night in jail <0.01 Sexual activity class difference Same class Respondent in higher class than partner Partner in higher class than respondent Partnership characteristics Met partner in a structured environment < week between meeting and first sexual intercourse <0.001 Median duration of relationship (months) 4 (0 166) 35 (0 266) <0.01 Believed that partner had had other partners Frequency of condom use Never Sometimes Frequently Always * p value for Pearson s chi-squared test unless otherwise specified. Four sexual activity classes were defined on the basis of quartiles of the lifetime number of partners: very low (1 2 partners), low (3 5 partners), moderate (6 19 partners), and high ( 20 partners). A structured environment included work, school, church, a private residence or private party, friends or family, or a social or volunteer service organization, as compared with a party, a bar/nightclub, the street, a vacation/ business trip, or a health club. Numbers in parentheses, range. p value for Wilcoxon rank-sum test. factors associated with gonococcal infection identified in case-control comparison B (STD clinic patients) formed a subset of the risk factors identified in the larger general population case-control analysis (comparison A) (table 4). In both case-control group comparisons, risk of gonorrhea decreased with age and was elevated among Blacks and persons whose most recent partner had spent a night in jail. In the general population case-control analysis (comparison A), additional risk factors for gonococcal infection identified included a lower educational level, anal intercourse, a history of STD, lack of condom use, and being divorced, while risk was reduced for persons whose annual income was greater than $9,375 and persons who had met their partner in a structured setting. None of these additional factors were significantly associated with gonorrhea in the case-control comparison in the subset of STD clinic patients. DISCUSSION Although the characteristics of gonorrhea cases from the STD clinic and cases from the Neisseria Reference Laboratory did not differ in this study, the characteristics of the RDD and STD clinic populations differed substantially. The effect of socioeconomic status was most notable, since

6 398 Manhart et al. TABLE 3. Characteristics associated with Neisseria gonorrhoeae infection in two case-control analyses,* Seattle, Washington, Characteristic All gonorrhea cases (n = 103) General population (comparison A) STD clinic patients (comparison B) RDD controls (n = 376) p value Gonorrhea clinic cases (n = 61) Clinic controls (n = 332) p value No. % No. % No. % No. % Demographic characteristics Female gender Median age (years) 22 (18 39) 30 (18 39) < (18 38) 25 (18 39) <0.01 Median annual income# (dollars) 4,800 (0 60,000) 17,500 (0 500,000) <0.01 5,664 (0 24,000) 500 (0 10,000) Education of high school or more < <0.001 Race White Black Other < <0.001 Single marital status < Employed < Ever spent a night in jail < Sexual behavior characteristics Median age (years) at first sexual intercourse 15 (3 25) 17 (8 33) < (5 25) 16 (4 31) 0.09 Median lifetime no. of sexual partners 12 (2 504) 8 (1 2,000) < (3 504) 15 (1 900) 0.75 Median rate of partner change** 2 ( ) 0.7 ( ) < ( ) 2.0 ( ) Concurrency Douching < History of anal intercourse < History of any STD < Mixing characteristics Race difference Age difference Male partner >5 years older Male partner 2 years younger to 5 years older Male partner 3 years younger Table continues persons in the general population had higher levels of income, education, and employment than did STD clinic patients. RDD respondents also more often reported partnerships characterized by factors associated with stable relationships: marriage, a longer duration of the relationship, and meeting in a structured environment. STD clinic patients were younger, were more likely to have spent a night in jail, reported a younger age at first sexual intercourse and a more diverse sexual repertoire, and more often selected partners of a similar race. Among the correlates of gonorrhea risk identified in the general population case-control analysis, only a subset (young age, Black race, having a partner who had spent a night in jail) were identified in the STD clinic. However, all of the factors identified in the STD clinic were also identified as risk factors for gonorrhea within the general population sample. Our cross-sectional results agree with those of previous studies showing that STD clinic patients more frequently report high-risk behaviors than do persons in the general population (2, 11, 12). However, the strongest and most consistent factors predicting gonococcal infection in the case-control analyses were sociodemographic rather than behavioral. The three effects that emerged consistently as risk factors for gonorrhea, irrespective of study population (race, age, partner having spent a night in jail), represent strong and far-reaching factors that cut across populations.

7 Study Population and Identification of STD Risk Factors 399 TABLE 3. Continued Characteristic All gonorrhea cases (n = 103) General population (comparison A) STD clinic patients (comparison B) RDD controls (n = 376) p value Gonorrhea clinic cases (n = 61) Clinic controls (n = 332) p value No. % No. % No. % No. % Education difference Same level of education Male more educated than female Female more educated than male Partner had spent a night in jail < <0.001 Sexual activity class difference Same class Respondent in higher class than partner Partner in higher class than respondent Partnership characteristics Met partner in a structured environment < week between meeting and first sexual intercourse Median duration of relationship (months) 4 (0 117) 35 (0 266) < (0 104) 4 (0 166) Believed that partner had had other partners < Frequency of condom use Never Sometimes Frequently Always * Comparison A: all gonorrhea cases versus random digit dialing controls; comparison B: sexually transmitted disease clinic gonorrhea cases versus clinic controls. STD, sexually transmitted disease; RDD, random digit dialing. p value for Pearson s chi-squared test unless otherwise specified. Numbers in parentheses, range. p value for Wilcoxon rank-sum test. # Adjusted for family size. ** Rate of partner change was defined as the average number of new partners acquired per year (lifetime number of partners/number of years sexually active). Any STD included nongonococcal urethritis, pelvic inflammatory disease, gonorrhea, chlamydia, syphilis, genital warts, and herpes. Four sexual activity classes were defined on the basis of quartiles of the lifetime number of partners: very low (1 2 partners), low (3 5 partners), moderate (6 19 partners), and high ( 20 partners). A structured environment included work, school, church, a private residence or private party, friends or family, or a social or volunteer service organization, as compared with a party, a bar/nightclub, the street, a vacation/business trip, or a health club These characteristics can be described as risk factors for exposure rather than factors that enhance transmission once a susceptible person has been exposed to an infected partner (18). For example, the long-standing association between Black race and gonococcal infection may be due to patterns of partner choice, which regulate exposure. In analyses of data from the National Health and Social Life Survey, African Americans with a low level of sexual activity (few lifetime partners) were more likely to select a partner from a higher sexual activity class (representing disassortative mixing) than were corresponding groups of Whites (19). Furthermore, African Americans in that study were more likely to select same-race partners (representing assortative mixing). These network characteristics and the fact that over 75 percent of reported cases of gonorrhea in 2000 were diagnosed in Black persons (1) make the risk of exposure to gonorrhea relatively high for Blacks and may explain the consistency of the association between Black race and gonococcal infection across case-control comparisons. Although age may be classified as a factor for transmission (physiologic changes such as decreasing zones of ectopy in the female cervix with increasing age may reduce the likelihood of transmission), if younger persons have sex with greater frequency than older persons, young age would also increase the risk of exposure. Having a partner who has spent a night in jail probably also represents increased risk of

8 400 Manhart et al. TABLE 4. Factors associated with gonococcal infection in a case-control study conducted in the general population (comparison A) and a case-control study conducted among persons who had visited a sexually transmitted disease clinic (comparison B), Seattle, Washington, Characteristic General population (comparison A) STD* clinic population (comparison B) OR*, 95% CI* OR 95% CI Age (per 5 years) , , 0.81 Black race , , Partner had spent a night in jail , , 7.30 Education of high school or more , , 2.16 Anal intercourse , , 1.80 History of any STD , , 3.65 Met partner in a structured environment , , 1.19 Never used condoms , , 3.23 Divorced marital status , , 8.50 Annual income $9,375 $17, , , 4.65 $17, , , 5.37 Not reported , , * STD, sexually transmitted disease; OR, odds ratio; CI, confidence interval. Adjusted for gender. Other variables that were tested in the model but did not appreciably change these estimates included employment status, ever spending a night in jail, age at first sexual intercourse, rate of partner change, concurrency, education difference, age difference, time between meeting and first sexual intercourse, belief that partner had had other partners, difference in grouped number of lifetime sexual partners, and duration of the partnership. A structured environment included work, school, church, a private residence or private party, friends or family, or a social or volunteer service organization, as compared with a party, a bar/nightclub, the street, a vacation/ business trip, or a health club. Reference category: <$9,375. exposure, since previously jailed persons are, by definition, prone to taking more risks and may be more likely to be infected. Low income characterized all cases of gonorrhea, irrespective of source, and was a consistent risk factor in both univariate case-control comparisons. Furthermore, in multivariate analyses, persons who chose not to report their income were at higher risk of gonococcal infection than those in higher annual income brackets ( $17,500). Although low income is highly associated with health risk behaviors in some studies (20), it is not in others (21, 22), making it difficult to determine whether income is directly related to the behaviors that result in acquiring an STD or whether a third factor (related to both income and a predisposition to engage in risky behaviors) is responsible for the observed associations. As with Black race, income may also be a marker for exposure, particularly if partners are chosen assortatively by income and STDs are concentrated in lowincome populations. However, this effect may be stronger for gonorrhea than for chlamydia, herpes, or human papillomavirus infections, which cut across income groups. The identification of many more risk factors in the general population case-control comparison than in the STD clinic subgroup may be entirely due to the high prevalence of characteristics such as low education, anal intercourse, history of STD, encountering partners in unstructured environments, lack of condom use, and divorce among STD clinic controls. With such small variation in behavior in the study population, these characteristics cannot differentiate between persons who are at risk for exposure to gonorrhea and those who are less so. On the other hand, these characteristics were strongly associated with gonorrhea in the general population comparison, where the baseline prevalence was low, and may be useful in directing more broad-based interventions. The choice of a clinic-based or general population-based study population will depend largely on the question of interest and the resources available for conducting the study. In this analysis, the cross-sectional STD clinic populations with and without gonorrhea had similar sociodemographic characteristics, probably related to utilization of public STD clinics. The STD clinic patients also had similar risk exposures or outcomes other than those given here that motivated them to seek care at an STD clinic, including symptoms of STD, referral for STD, suspected exposure to STD, and lack of private health insurance. If the objective of a study is to identify risk determinants, independent of sociodemographic correlates of public health clinic use, then selecting public health clinic patients will produce relatively unbiased estimates. However, if those risk determinants are also associated with seeking care at STD clinics, selecting both cases and controls from an STD clinic could represent overcontrolling. In this case, a more economical alternative to a general

9 Study Population and Identification of STD Risk Factors 401 population-based sample might be to enroll study subjects from public health clinics other than STD clinics, if they serve patients with similar sociodemographic characteristics. Finally, since persons who perceive that they are at low risk for infection would be unlikely to visit an STD clinic in the absence of symptoms, a clinic setting is less than ideal for identifying and enrolling a representative sample of asymptomatic cases. A major strength of these analyses is the direct comparability of the questions posed to the three groups studied (STD cases, STD clinic controls, and RDD controls). In most cases, the questions posed to all persons were identical, irrespective of the source population. However, the risk factors identified may be specific to the Seattle metropolitan area; risk factors for gonococcal infection among both STD clinic patients and the general population in other areas may be different. Furthermore, we restricted our study population to persons aged years, whereas rates of STD are highest among persons aged years (1), who were partially excluded from our analysis. RDD is a generally accepted method of identifying population-based controls (23), yet it is not without limitations. Respondents to our RDD survey may have been of higher socioeconomic status (by virtue of having a telephone), and those who answered and agreed to participate may have had more liberal views on sexual behavior, which in turn may have been associated with higher levels of risky behavior than would be found in the population as a whole. Our general population case-control population may have been subject to misclassification bias if some participants not tested actually had gonorrhea and to participation bias if nonrespondents differed from respondents and the case group did not include asymptomatic or nonreported cases. However, Rogers et al. (24) noted that persons with gonorrhea and chlamydia infections detected by nucleic acid amplification test (likely to be asymptomatic prevalent cases) did not have a history of high-risk behavior, nor did they report more recent occurrences of those behaviors; this suggests that misclassification bias due to unidentified gonococcal infection among subjects in the RDD population may have been small. General population cases identified through the Neisseria Reference Laboratory were somewhat different from those identified through the STD clinic. Few of these differences were statistically significant, although it is possible that more differences would have been detected had our sample size been larger. During the study period, cases with and without isolates sent to the Neisseria Reference Laboratory did not differ by source of care, gender, or age, and recruited cases did not differ from the total population of non-std cases (our unpublished data). Thus, these two groups together probably capture the spectrum of gonorrhea cases. Although the epidemiology of gonorrhea in Seattle has changed somewhat in the years since these data were collected (e.g., it is increasingly concentrated among men who have sex with men (25)), and this may have changed some specific risk factors for the disease, these changes are not likely to have influenced the comparisons between study populations presented here. These data suggest that risk factors identified in an STD clinic-based case-control study, despite overcontrolling for care-seeking behavior, are likely to be confirmed in a general population survey. However, this overcontrolling may lead to failure to identify other risk factors that could be identified in a general population case-control study. In the case of a relatively unknown but suspected etiologic agent, choice of study population is particularly important. For example, most of the research showing an association between reproductive tract disease and Mycoplasma genitalium suggests that this may be a new sexually transmitted pathogen (26 28), but studies have been conducted in high-risk populations to maximize the opportunity to detect this rare organism. Complementary studies using general population samples are unlikely to negate these initial observations but may identify additional risk factors for infection and enrich our understanding of the transmission dynamics associated with this emerging organism. Thus, appropriately conducted studies using each of these two groups, STD clinic populations and general population samples, can yield valid information on risk factors for STD. ACKNOWLEDGMENTS This research was supported by an Association of Schools of Public Health/Centers for Disease Control and Prevention/Agency for Toxic Substances and Disease Registry Cooperative Agreement (Project S105-14/14); by a grant from the Bristol-Myers Squibb Company (New York, New York); and by the University of Washington STD Cooperative Research Center (grant AI from the National Institute of Allergy and Infectious Diseases). Dr. Lisa Manhart was supported by a training grant (grant A107140) from the National Institute of Allergy and Infectious Diseases. The random digit dialing telephone survey was conducted by the Social and Economic Sciences Research Center at Washington State University (Pullman, Washington). Rosie Pavlov was the study director. REFERENCES 1. National Center for HIV, STD and TB Prevention, Centers for Disease Control and Prevention. Sexually transmitted disease surveillance, Atlanta, GA: Centers for Disease Control and Prevention, (World Wide Web URL: Howards PP, Thomas JC, Earp JA. Do clinic-based STD data reflect community patterns? Int J STD AIDS 2002;13: Hook EW, Handsfield HH. Gonococcal infections in the adult. In: Holmes KK, Sparling PF, Mardh P-A, et al, eds. Sexually transmitted diseases. New York, NY: McGraw-Hill, Inc, 1999: Cates WC Jr, Holmes KK. Re: Condom efficacy against gonorrhea and nongonococcal urethritis. (Letter). Am J Epidemiol 1996;143: Sanchez J, Campos PE, Courtois B, et al. Prevention of sexually transmitted diseases (STDs) in female sex workers: prospective evaluation of condom promotion and strengthened STD services. Sex Transm Dis 2003;30: National Center for Health Statistics. National Health and Nutrition Examination Survey: survey content.

10 402 Manhart et al. Hyattsville, MD: National Center for Health Statistics, (World Wide Web URL: nhanes/comp3.pdf). 7. Laumann EO, Gagnon JH, Michael RT, et al. The social organization of sexuality: sexual practices in the United States. Chicago, IL: University of Chicago Press, National Center for Health Statistics. National Survey of Family Growth: survey description. Hyattsville, MD: National Center for Health Statistics, (World Wide Web URL: Harris KM, Florey F, Tabor J, et al. The National Longitudinal Study of Adolescent Health: research design. Chapel Hill, NC: Carolina Population Center, University of North Carolina, (World Wide Web URL: projects/addhealth/design). 10. Ku L, St Louis M, Farshy C, et al. Risk behaviors, medical care, and chlamydial infection among young men in the United States. Am J Public Health 2002;92: Johnson AM, Mercer CH, Erens B, et al. Sexual behaviour in Britain: partnerships, practices, and HIV risk behaviours. Lancet 2001;358: Garnett GP, Hughes JP, Anderson RM, et al. Sexual mixing patterns of patients attending sexually transmitted diseases clinics. Sex Transm Dis 1996;23: Johnson AM, Fenton KA, Mercer C. Phase specific strategies for the prevention, control, and elimination of sexually transmitted diseases: background country profile, England and Wales. Sex Transm Infect 2002;78(suppl 1):i Rothman KJ, Greenland S. Case-control studies. In: Rothman KJ, Greenland S, eds. Modern epidemiology. 2nd ed. Philadelphia, PA: Lippincott-Raven Publishers, 1998: Foxman B, Aral SO, Holmes KK. Interrelationships among douching practices, risky sexual practices, and history of selfreported sexually transmitted diseases in an urban population. Sex Transm Dis 1998;25: Schwebke JR, Whittington W, Rice RJ, et al. Trends in susceptibility of Neisseria gonorrhoeae to ceftriaxone from 1985 through Antimicrob Agents Chemother 1995;39: Stata Corporation. Stata, version 6.0. College Station, TX: Stata Corporation, Aral SO, Peterman TA. A stratified approach to untangling the behavioral/biomedical outcomes conundrum. Sex Transm Dis 2002;29: Laumann EO, Youm Y. Racial/ethnic group differences in the prevalence of sexually transmitted diseases in the United States: a network explanation. Sex Transm Dis 1999;26: Lantz PM, House JS, Lepkowski JM, et al. Socioeconomic factors, health behaviors, and mortality: results from a nationally representative prospective study of US adults. JAMA 1998; 279: Blum RW, Beuhring T, Shew ML, et al. The effects of race/ethnicity, income, and family structure on adolescent risk behaviors. Am J Public Health 2000;90: Santelli JS, Lowry R, Brener ND, et al. The association of sexual behaviors with socioeconomic status, family structure, and race/ethnicity among US adolescents. Am J Public Health 2000;90: Olson SH, Kelsey JL, Pearson TA, et al. Evaluation of random digit dialing as a method of control selection in case-control studies. Am J Epidemiol 1992;135: Rogers SM, Miller HG, Miller WC, et al. NAAT-identified and self-reported gonorrhea and chlamydial infections: different atrisk population subgroups? Sex Transm Dis 2002;29: Xia M, Whittington WL, Shafer WM, et al. Gonorrhea among men who have sex with men: outbreak caused by a single genotype of erythromycin-resistant Neisseria gonorrhoeae with a single-base pair deletion in the mtrr promoter region. J Infect Dis 2000;181: Taylor-Robinson D, Horner PJ. The role of Mycoplasma genitalium in non-gonococcal urethritis. Sex Transm Infect 2001; 77: Cohen CR, Manhart LE, Bukusi EA, et al. Association between Mycoplasma genitalium and acute endometritis. Lancet 2002; 359: Manhart LE, Critchlow CW, Holmes KK, et al. Mucopurulent cervicitis and Mycoplasma genitalium. J Infect Dis 2003;187: