A Systematic Review of the Aetiology of Salmonellosis in New Zealand

Transcription

1 A Systematic Review of the Aetiology of Salmonellosis in New Zealand Dr Nick Wilson, Public Health Physician, Wellington, New Zealand* Associate Professor Michael Baker, Public Health Physician, Wellington * nwilson@actrix.gen.nz June 2009 Contents Executive Summary...3 Acknowledgements...7 1) Introduction...8 2) Methodology...9 Limitations of this review ) Review of National Case-Control Studies...12 Background ) Prospective national case-control study of sporadic disease ) National case-control study of emergent S. Brandenburg ) Case-control study of an outbreak of S. Typhimurium DT Summary ) Review and Analysis of Notification Data and Related Studies...23 Background...23 Studies identified ) ESR Report (focusing on 2007 data) ) Case-case analysis (2006 data) ) Case-case analysis (2004 data) ) In-depth 2003 study ) Review by Thornley et al (2002)...32 Summary...34

2 Aetiology of Salmonellosis in NZ 2 5) Outbreak Surveillance Data and Published Outbreaks...35 Background ) Study of salmonellosis outbreaks by King & Lake ) Published studies and investigations of salmonellosis outbreaks ) Annual Outbreak Surveillance Report by ESR (2007 data)...42 Summary ) Review of Relevant Environmental, Laboratory and Risk Analysis Studies ) Source attribution study (a Modified Hald Model published in 2009) ) NZFSA Report Analysis of Foodborne and Other Pathways for the Exposure of New Zealanders to Salmonella (2005) ) Laboratory data on serotypes (various sources) ) ESR Laboratory data that reports associations between human disease and possible sources ) Risk profile on Salmonella and poultry product ) Survey for the NZFSA (retail eggs) ) Review for the NZFSA (raw milk and raw milk cheeses) ) Laboratory test results for other foods (other than poultry, eggs and milk) ) Studies of water contamination ) Selected other New Zealand studies of potential relevance...67 Summary ) Overall Assessment...70 Evidence for the aetiology of salmonellosis in New Zealand...70 Evidence for particular food vehicles...73 Identified data gaps that could be addressed to refine the knowledge base on aetiology...78 References...81

3 Aetiology of Salmonellosis in NZ 3 Executive Summary Aims To systematically review the available scientific evidence around the aetiology of human (non-typhoidal) salmonellosis in the New Zealand setting. In particular, to examine the possible role of foodborne transmission and to identify the most relevant food vehicles (if any). Method Searches for published and unpublished studies and reports were undertaken using: Medline, Google Scholar, key websites and hand-searches of selected New Zealand publications. Additional analyses were undertaken with regard to outbreak data, surveillance data (notifications) and national laboratory data on human and nonhuman isolates of Salmonella. Assessments were made on the totality of the evidence, but with particular regard to strength of associations and consistency of associations. Findings Review of case-control studies: This review identified three national level casecontrol studies, but only one of these was of sporadic disease. All three studies provided evidence for foodborne transmission, but the study of sporadic disease identified this as the source with the largest population attributable risk percentage (PAR%). The next largest PAR% in this study of sporadic disease was for contact with bird faeces. All three studies identified direct animal contact as a risk factor and likewise for person-to-person transmission. None identified contaminated domestic water or recreational water exposure as risk factors. Only the study of sporadic disease identified recent international travel as a risk factor. The results from the study of sporadic disease are clearly the most important when considering the national patterns of transmission at the current time. The findings of the two case-control studies into outbreaks provide some information on transmission mechanisms in New Zealand but it is difficult to interpret the relevance of these to the overall disease burden. Nevertheless, they do show that for specific serotypes involved in widespread outbreaks, there can be multiple transmission mechanisms. That is even if direct zoonotic transmission dominates for a specific serotype (eg, S. Brandenburg), other factors (eg, kitchen hygiene and person-to-person transmission) may also be relevant. Review of notification data and related studies: There are numerous limitations with notification data but it still provides some information of relevance to aetiological considerations. The case-case comparisons using notification data on risk factors are also only a crude guide to possible aetiology of salmonellosis. Nevertheless, the results are consistent with foodborne transmission being relatively important (compared to some other enteric diseases, but possibly less so than campylobacteriosis). These data also provide weak evidence for person-to-person transmission and a role of direct animal contact.

4 Aetiology of Salmonellosis in NZ 4 Nevertheless, the analyses by rurality suggest that rural factors may not collectively be that important in salmonellosis transmission overall (though rural factors are undoubtedly important for some specific serotypes such as S. Brandenburg). When it was assumed that urban residents may experience between a fifth and one twentieth of the level of rural factors that rural people do, then the proportion of all Salmonella serotypes from rural factors nationally was only estimated to be between 4.4% and 7.1%. This suggests fairly minor roles for all of the following transmission mechanisms: (i) direct animal contact (at least for farm animals); (ii) contaminated drinking water; (iii) recreational water exposure; and (iv) to some extent contaminated environments. Data from outbreaks: The data from outbreaks in New Zealand provide evidence that foodborne transmission is the dominant transmission route for salmonellosis outbreaks. When considering all three levels of evidence (as described by: [King & Lake 2007]), the proportion is 68% (87/128) (95% confidence interval (CI) = 59.5% 75.6%). A higher figure of 71% (15/21) (95%CI = 49.8% 87.5%) is obtained when just considering the strongest evidence. This pattern of dominant foodborne transmission is also apparent when just considering published outbreaks and for outbreaks in 2007 (the most recent year with data). The next most important transmission route for salmonellosis outbreaks is person-toperson (at 20% in the review by King and Lake) and then waterborne transmission (6%). Environmental and zoonotic transmission were both under 5%. As detailed in a recent paper [Greig & Ravel 2009], there are limitations with studying outbreaks as part of determining aetiological pathways for a disease which mainly involves sporadic cases. Nevertheless, this approach has some strengths in it can demonstrate a direct link between a source and actual human cases. Also, for New Zealand the same serotypes are typically involved in both outbreaks and in comprising the majority of all cases. Environmental, laboratory and risk analysis studies: This Section of the Review dealt with a diverse range of work but the key issues relating to salmonellosis transmission can be summarised as follows: The work by Mullner et al (2009) is by far the highest quality analytic work conducted on food attribution and salmonellosis for New Zealand. It is also a significant contribution to the international literature in terms of methodological issues. While still having a range of limitations, it suggested that pork may be the major food source in this country albeit with a wide credible interval (of between a half and three quarters of food-related cases). Next in importance was poultry at around a fifth of cases, then beef and veal (at around a tenth). Analytic work and expert judgement in a NZFSA Report identified that the majority (63%) of salmonellosis transmission is likely to be foodborne. The next most important known category was considered to be direct zoonotic transmission (10%). The top five in decreasing order of discrete foods vehicles were considered to be: domestic poultry, eggs, beef / pork and sheep meat.

5 Aetiology of Salmonellosis in NZ 5 However, other categories such as ready-to-eat meat products may also be relatively important. For the 2007 data from ESR, human and non-human Salmonella isolates were compared. When considering the top five human serotypes, the similarity was greatest for isolates from food sources, followed by poultry (environmental sources), and then bovine ones. The most different pattern was seen for meal/bone meal, reptile sources and ovine sources. Although this analysis was relatively simplistic, it is compatible with food sources being the most important ones relative to those for direct zoonotic transmission and from environmental sources. ESR laboratory data regularly reports associations between human salmonellosis and possible sources. These data provide additional detail around regular foodborne disease outbreaks being associated with particular food products and particularly food premises. A report by Lake et al published in 2004 closely examined the risk profile from Salmonella in poultry product. It provided evidence that the prevalence of infection in poultry product has declined over the past decade (but this period has also seen a risk in per capita chicken consumption and so the overall trend in contribution is still unclear). The situation is likely to be dynamic however, with the poultry industry enhancing pathogen control measures in recent times. Additional evidence for poultry product contributing to salmonellosis in New Zealand came from outbreak data (18% of foodborne outbreaks) and also the close temporal patterns for Salmonella isolates in poultry and human cases (eg, for S. Typhimurium serotypes: DT1, DT12a, DT135, and DT160). Many foods have tested positive to Salmonella in the New Zealand setting. The list of domestic products includes uncooked retail meats, beef, sheep meat, pig meat, shellfish and apples. The list of imported foods includes: ready-to-eat meats, crustaceans, coconut, herbs and spices, and tahini. For most food products the prevalence of contamination is low but for others it exceeds 1% (especially some of the imported foods that have been tested to date). Since some of these foods may be eaten rare or raw, or may not be properly cooked, these test results are consistent with foodborne transmission being important in the New Zealand setting. Available New Zealand data are consistent with recreational water and untreated drinking water (eg, from roof-collection) being risk factors for salmonellosis in this country. Nevertheless, it is likely that contaminated water is a relatively minor transmission pathway given the relative infrequency of waterborne outbreaks and the lack of evidence from a range of other sources. Summary assessment of aetiology An abbreviated summary is contained in the table below. Of note is that for foodborne transmission this assessment is consistent with some other New Zealand evidence for

6 Aetiology of Salmonellosis in NZ 6 enteric disease transmission, with information around suboptimal food hygiene practices and with evidence from other developed countries. Table (i): Abbreviated summary of the assessment of the aetiology of salmonellosis in the domestic New Zealand setting Aetiology Contaminated food Person-to-person Direct animal contact Contaminated water (domestic) Recreational water exposure Contaminated environments (eg, from bird faeces) Overall assessment Very likely (>90% probability) to be the majority cause (ie, >50% of all cases). The data from the case-control study, the outbreak data and the comparisons of isolates provide the strongest evidence. Likely to be a moderate cause (ie, between 10-30% of all cases) Likely to be a minor cause (ie, <10% of cases) Very likely to be a minor cause (ie, <10% of cases). As above As above Summary of food vehicles An abbreviated summary is contained in the table below. Of note is that the patterns are generally consistent with the international literature for developed countries (except for a much lower role for eggs in the NZ setting). Table (ii): Abbreviated summary of the assessment of the role of food vehicles in the New Zealand setting Food vehicle Poultry Pig meat Meat in general (expoultry, ex-fish) Beef Eggs Sheep meat Dairy products Fresh produce Overall assessment Very likely (>90% probability) to be at least a moderate cause (ie, between 10-30% or higher of all cases) Likely (>66% probability) to be at least a moderate cause (ie, between 10-30% or higher of all cases). (But one sophisticated analysis puts this as the major cause at 60%, credible interval of: 47-74%). Likely (>66% probability) to be at least a moderate cause (ie, between 10-30% or higher of all cases). As above Likely to be a minor cause (ie, <10% of foodborne cases) As above As above As above

7 Aetiology of Salmonellosis in NZ 7 Identified data gaps that could refine the knowledge base The following areas were identified as being ones where additional data could assist in further refining knowledge of salmonellosis aetiology in the current New Zealand setting: 1) Additional laboratory-based data attribution studies (expanding on the work by Mullner et al in 2009 and on other work on campylobacteriosis in New Zealand). There is a particular data gap around testing of pig meat samples. 2) Additional data analyses of Salmonella serotypes by rurality (an approach that has already yielded useful insights). 3) Additional analyses of salmonellosis outbreak data. 4) Use of notification data in case-case analyses (eg, comparing risk factors for different major serotypes such as S. Typhimurium vs S. Enteritidis). 5) Improved risk factor data collection (on salmonellosis) as part of a possible sentinel surveillance system approach. 6) Data collection and analyses of the possible co-benefit impacts on salmonellosis from improved Campylobacter control in poultry. 7) Data collection around new targeted interventions (eg, in one part of the country with the rest of the country as a control area). Acknowledgements This project was managed by Dr Donald Campbell and Andrew Pavitt of the New Zealand Food Safety Authority, with peer review kindly provided by Dr Campbell. Thanks are also owed to those New Zealand researchers and health workers who have collected data and reported it, so as to better inform salmonellosis surveillance and control. New Zealand citizens have also contributed valuable information as either cases or controls in outbreak investigations and through surveillance systems.

8 Aetiology of Salmonellosis in NZ 8 1) Introduction The concern with improving control of enteric diseases is widespread in developed countries and this encompasses salmonellosis. It is certainly in New Zealand s interest to improve control of salmonellosis, as well as other enteric diseases. The four major reasons can be summarised as follows: 1) Preventing harm to health (both acute and chronic morbidity and mortality) and the associated costs to the health sector and economy (via work absenteeism). 2) Preventing the contamination of food exports (and therefore preventing any associated downstream harm to export industries and associated economic and employment impacts). 3) Preventing damage to the quality reputation of cuisine in New Zealand that contributes to the success of the tourist industry, a major contributor of national revenue. 4) Preventing harm to livestock of economic importance (ie, the associated morbidity and mortality from salmonellosis especially in sheep and pigs). Salmonellosis is currently the third most frequently notified enteric disease (after campylobacteriosis and giardiasis) in New Zealand [ESR 2008b]. In 2007 there were 1274 notifications (a national rate of 30.1 per 100,000 population). All these statistics and the following ones refer to non-typhoidal salmonellosis (the focus of this review). Recent trend data are suggestive of lower rates of salmonellosis than occurred in the 1998 to 2001 period for notified cases and laboratory-reported cases [ESR 2008b]. However, there was no notable change in rates during the 2005 to 2007 period. In a comparison for 2006 data, New Zealand was found to have relatively mid-range values of notified salmonellosis compared to the other countries considered (Table 1.1). Table 1.1: International comparisons of annual incidence rates of notified salmonellosis Country Incidence rate of salmonellosis (per 100,000 population) For around * 2006 (ranked)** Australia 42.1 (2002) 40.1 [Begg et al. 2008] Norway 33.3 (2000) 39.1 Iceland 38.7 European Union (27 countries) 33.9 (overall) Range: 3.7 (Portugal) to (Czech Republic) New Zealand 50.0 (2002) 31.9 [ESR 2008b] Denmark# 38.6 (2002) 30.6 England & Wales# 28.0 (2000) 23.4 (UK) USA (selected sites) 16.1 (2000) 14.8 [CDC 2007] Ireland# 11 (2001) 10.0

9 Aetiology of Salmonellosis in NZ 9 Notes: * Data from [Sneyd & Baker 2003] and [Lake et al. 2004b]. ** All data from the following reference unless otherwise indicated [European Centre for Disease Prevention and Control 2008]. # These countries are also included in the European Union grouping but are shown as there is comparison data between the time periods. In the latest year for which there are data (2007), New Zealand had eight reported outbreaks of salmonellosis (with 141 cases and with 9 cases hospitalised) [ESR 2008b]. Also in 2007 there were 110 hospitalisations attributable to salmonellosis but many of the notifications (34.6%) did not have hospitalisation status data recorded [ESR 2008b]. Endovascular infection is one of the most serious forms of extraintestinal infection and has been described in the New Zealand setting [Drinkovic et al. 2004]. In a 10-year period, salmonellosis was reported as a cause of 16 deaths (for EpiSurv data: ) [ESR 2008b]. One estimate as part of modelling work for the New Zealand situation [Lake et al. 2000], assumed that for every reported case of salmonellosis, there were 3.2 community cases in which a GP was not visited (based on an English study [Wheeler et al. 1999]). It estimated over 24,000 annual productive days lost due to salmonellosis per year. A related study estimated annual costs from salmonellosis that totalled over $4 million ($245,000 direct costs, $2.93 million indirect costs and $1.29 million intangible costs ) [Scott et al. 2000]. This made salmonellosis the second most costly foodborne infectious disease in the country after campylobacteriosis. Of note was that this study used overseas estimates to assume that 92% of salmonellosis was foodborne. In New Zealand there have been case-control studies, outbreak investigations, analyses of notification data and laboratory-based studies into salmonellosis. A review of some of these data were published in 2002 [Thornley et al. 2002]. Others have reviewed the presence of Salmonella within animal populations [Clark et al. 2002]. Nevertheless, a detailed review of the available New Zealand-specific evidence around aetiology of human salmonellosis has not been undertaken. To address this issue more fully, the New Zealand Food Safety Authority (NZFSA) commissioned this particular review. 2) Methodology Definitions: This document uses an expanded definition of foodborne that is inclusive of both infection from contaminated food from an food production source as well as from an infected food handler. The latter issue was articulated in an expert workshop coordinated by the NZFSA: Definition of foodborne. Some transmission of pathogens occurs from an infected food handler to the food, followed by consumption. Although the

10 Aetiology of Salmonellosis in NZ 10 source of contamination is more immediate than for most foods, the consultation decided that this did indeed represent foodborne transmission. [Cressey & Lake 2005] This document also focuses on non-typhoidal salmonellosis and uses the term salmonellosis to refer to this grouping. Nevertheless, in some places the data sources used in certain publications have included S. Typhi and/or S. Paratyphi and in these cases this is reported in the text. Medline searches: Numerous Medline searches were conducted to identify New Zealand specific articles and also to identify international scientific literature from which to put the New Zealand work into context. The search period covered up to the end of March The various searches used the following search terms: Zealand and salmonell* (which identified 199 articles in Medline). Searches for publications by the New Zealand-based authors who had contributed two or more publications to the Medline-indexed literature relating to salmonellosis. More specific searches around attribution and salmonell* were undertaken. Other New Zealand specific case-control studies relating to enteric pathogens ( Zealand and case-control and: campy* / cryptosp* / giard* / crypto* / listeri* ). Searches for non-medline indexed literature: The Internet search engine most relevant to the academic literature was used (ie, Google Scholar Search terms focused on salmonella/salmonellosis, food, and New Zealand. The following publications were hand-searched for articles relating to salmonellosis in New Zealand: Communicable Disease New Zealand: All the A5 format issues for 1991 to New Zealand Public Health Report: From the first issue (volume 1 in June 1994) to the last issue (volume 9 in October-December 2002). New Zealand Public Health Surveillance Report: For all issues up to the most recent available (ie, from the first issue in 2003 to including the March 2009 issue). These publications are all available online at the ESR website. New Zealand Journal of Environmental Health (covering four issues per year for the period January 1990 to 2008) though more recently this journal has had a name change to Environment and Public Health. These publications are not online. For all the articles obtained, the bibliographies were searched for additional published and unpublished work of potential relevance to the New Zealand setting. Examination of key websites for reports: To identify other publications, the websites of ESR and the NZFSA were examined. Overall assessment of the quality of the evidence and role of a transmission route: The use of standard approaches for assessing causal relationships (eg, [Institute of Medicine 2000]) was considered but were not entirely appropriate for this review

11 Aetiology of Salmonellosis in NZ 11 given that there is no scientific doubt that salmonellosis in humans can be caused by contaminated food, water and from direct zoonotic contact. So the focus was more on the totally of the evidence around the strength of associations and the consistency of the associations in the New Zealand setting. Furthermore, the following issues around the scientific quality of the evidence obtained were given particular consideration: The strengths and limitations of case-control studies [Rothman & Greenland 1998; Giesecke 2002]. The strengths and limitations of outbreak investigations [Goodman et al. 1990], including there use for considering attribution [Greig & Ravel 2009]. The strengths and limitations of surveillance data [Giesecke 2002] and particularly New Zealand notification data [Simmons et al. 2002; ESR 2003; Sneyd & Baker 2003]. The numerous limitations inherent in risk factor epidemiology [Taubes 1995]. With regard to terminology the approach taken by a United Nations Science Organisation in Table 2.1 was taken, along with another scale devised for the purposes of this review (Table 2.2). Table 2.1 Likelihood scale used in this report (based on: [Manning 2006]) Terminology Virtually certain Very likely Likely About as likely as not Unlikely Very unlikely Exceptionally unlikely Degree of confidence in being correct > 99% probability of occurrence > 90% probability > 66% probability 33% to 66% probability < 33% probability < 10% probability < 1% probability Table 2.2 Terminology for scale of size of cause in this report Terminology The majority cause A moderate cause A minor cause A very minor cause Quantification > 50% of cases Between 10-30% of all cases <10% of cases <1% of cases Limitations of this review The scope of this review was limited in that it did not go into exhaustive detail into studies of peripheral relevance to human salmonellosis (eg, New Zealand studies identifying the presence of Salmonella in species and environments with which humans usually have little direct contact). The search for studies on other enteric diseases in New Zealand and of overseas studies of potential relevance to New Zealand was also limited. Similarly, there may still be studies of relevance in the grey

12 Aetiology of Salmonellosis in NZ 12 literature that were not identified in the Google Scholar search, hand-searches or website searches. Other limitations of note are: Various limitations exist around the supplementary statistical analyses undertaken for this review (eg, especially the relationships between the serotypes of human and non-human isolates). These issues are discussed further in the relevant sections in this document. The authors of this 2009 review document were sometimes contributors of the work being reviewed. Nevertheless, the lead author of this document (NW) was only involved in one such study [Wilson et al. 2008], and he entirely wrote all those sections which reviewed work that involved the other author (MB). The latter author also has a role in giving independent scientific advice to the NZ Food Safety Authority for which he receives a small annual honorarium. But both authors have no competing financial interests and indeed their predominant research interests in the communicable disease area have been more focused on other non-enteric communicable diseases, such as influenza and close contact diseases. 3) Review of National Case-Control Studies Background Case-control studies are a useful epidemiological method for assessing risk factors for fairly uncommon conditions. This review identified three national-level case-control studies relating to salmonellosis: One study on sporadic salmonellosis [Baker et al. 2003]. A published study relating to a national outbreak of S. Brandenburg [Baker et al. 2007b], A published study relating to a national outbreak of S. Typhimurium DT160 [Thornley et al. 2003]. Other small case-control studies have been conducted in the context of local outbreak investigations and these are detailed in Section 6. In considering these studies, the limitations of case-control studies were considered (see the methodology section). Furthermore, the methodologies and results were considered in the context of published case-control studies concerning other enteric diseases in New Zealand: Campylobacteriosis [McMahon & Mahmood 1993; Ikram et al. 1994; Eberhart-Phillips et al. 1997], [Baker et al. 2005], and two unpublished casecontrol studies: [Neal & Bloomfield 1997; Bennett et al. 2003].

13 Aetiology of Salmonellosis in NZ 13 Giardiasis [Fraser & Cooke 1991; Mitchell et al. 1993; Hoque et al. 2001; Hoque et al. 2002; Hoque et al. 2003]. Yersiniosis [Satterthwaite et al. 1999]. 1) Prospective national case-control study of sporadic disease Background: This study [Baker et al. 2003], appears to be the first one in New Zealand to examine risk factors contributing to the overall burden of sporadic nontyphoidal salmonellosis. The study was in a report but a manuscript relating to it has been submitted to a journal but has not yet been published (as of June 2009). Methods: This was a prospective case-control study that was conducted with 139 cases and 139 controls, during the period February 2002 to April The methods used for the study have actually been published in the peer-reviewed literature previously [Baker et al. 2007b]. The Salmonella isolates were obtained for each case, and were serotyped. Main findings (risk factors and population attributable risk percentages PAR%): The study found statistically significant risk factors associated with various animal exposures, food and food-preparation related exposures, overseas travel and exposure to other sick people in the household or events where other people became sick (see Table 3.1). Table 3.1: Odds ratios for different risk factors from the univariate analysis in the case-control study by Baker et al [Baker et al. 2003] (for non-typhoidal salmonellosis in New Zealand, February 2002 April 2003, and excluding S. Brandenburg infections). (Some editing and organisational changes to the table have been made for the purposes of this 2009 review). Risk factor (all exposures in 3 days prior to illness)** Odds ratio 95%CI p-value Animal exposures Have puppy in house 9.00 (1.69, ) 0.01 Have touched puppy 8.00 (1.47, ) 0.01 Household occupational contact with animals or animal 2.22 (1.04, 5.13) 0.04 carcasses Household occupational contact with cattle or cattle carcasses 2.80 (1.07, 8.66) 0.04 Definite or possible contact with dog or puppy faeces 2.50 (1.02, 7.01) 0.05 Definite or possible contact with birds faeces 3.40 (1.35, 10.34) 0.01 Food and food preparation related exposures Ate pork steak 9.00 (1.69, ) 0.01 Ate hot dogs 2.80 (1.07, 8.66) 0.04 Ate food prepared at public bar 4.00 (1.00, 26.50) 0.05 Consumed food at event or occasion where others 7.50 (2.12, 47.58) became unwell Use of kitchen bench, sink, or table for chopping 5.67 (1.90, 24.27) Same chopping board used for meat and vegetables 1.67 (1.01, 2.80) 0.04 Other exposures Overseas travel 9.00 (1.69, ) 0.01

14 Aetiology of Salmonellosis in NZ 14 Risk factor (all exposures in 3 days prior to illness)** Odds ratio 95%CI p-value Other people in household with diarrhoea* 2.57 (1.12, 6.61) 0.03 Other people in household saw GP about diarrhoea* 6.00 (1.03, ) 0.05 Notes: * Secondary cases in the same household as an earlier case were excluded from the study. ** No associations were found for antacid consumption and prior antibiotic use. Statistically significant apparent protective effect was shown for consumption of: tomatoes, roast lamb or mutton, and salami. Statistically significant non-food protective effects were shown for contact with rabbits or guinea pigs. On multivariate analysis, exposures accounting for the greatest proportion of sporadic salmonellosis in the study population were the following (ordered by PAR%): Use of kitchen bench, table or sink for chopping food (PAR%=11.1%) Contact with bird faeces (10.7%) Consumption of food at an event where others became sick (9.6%) Recent international travel (5.9%) Contact with a pet puppy (5.2%). Strengths of the study: Good design: The study was national with matching by geographic area and rurality (using three strata: rural, small town and urban). It was also able to exclude cases with S. Brandenburg infection (these were reported on separately [Baker et al. 2007b]). The numbers of cases and controls were enough to allow for reasonable study power. The time period covered included all four seasons, which is relevant for a disease with marked seasonality. Appropriate analysis: Multivariate analysis was performed. Also population attributable risk (PAR) was estimated for factors associated with illness that were considered causal using an appropriate method (ie, of Bruzzi et al [Bruzzi et al. 1985]). Good contextualisation of the findings: The findings were very well placed in terms of what is known about the risk factors identified and with comparisons with other countries eg,: The proportion of NZ NTS [non-typhoidal salmonellosis] cases with a history of recent travel (16% among cases [Thornley et al. 2002]) is higher than that found in Canada [Doré et al. 2004] (10%) and the United States [Mermin et al. 2004] (10%), but lower than that in Norway [Kapperud et al. 1998a] (88% among cases ). This is consistent with other evidence that the international emergence of S. Enteritidis as a major foodborne pathogen, particularly phage types 8, 13a [Mishu et al. 1994] and 4 [Rodrigue et al. 1990] associated with shell eggs, has not extended to NZ. Good consideration of the laboratory evidence: For example: Only 6.5% of cases in this study were identified with S. Enteritidis infection, and none with phage types associated in the literature with transovarian transmission.

15 Aetiology of Salmonellosis in NZ 15 Infection with S. Typhimurium DT160 has previously been found to be associated with exposure to faecal material from passerine birds [Thornley et al. 2003], and in our study this serotype accounted for 29% of case-patients overall; Examination of limitations: There was thoughtful consideration of the limitations of the study (which are discussed further below) in the Discussion Section. Expertise: The study was performed by public health workers who have previously published other case-control studies and other analytic work in the field of enteric disease epidemiology in New Zealand (including in high quality international journals). Limitations with the study: The authors of this study acknowledged three specific limitations and their attempts to address these: 1. Selection bias: We were aware that selection bias may have become a potentially important problem for the study because our telephone recruitment methodology may have been more likely to identify controls who are more commonly at home. To mitigate this, we applied specific criteria to recruit controls, and called households multiple times to make contact with those controls meeting those criteria. However, controls may still have differed from cases, though such differences were not evident based on broad demographic characteristics. Added to this is the fact that 115 potential control subjects (out of 276 peoples meeting the matching criteria, ie, 42%) declined to participate. 2. Information bias: We were also aware of the possibility of information bias in recall of exposures, and attempted to mitigate this by truncating the interval for recruitment and by using aides memoire to help recall. However, any information bias would have tended to reduce the strength of the measured association between specific exposures and disease risk. The focus by the authors of this study on the 3-day time period before illness may have helped reduce the scope for information bias. But it also introduces another potential source of bias. That is, although the incubation period for Salmonella infection is usually considered to be hours, it is actually highly variable and dose-dependent [Abe et al. 2004]. This means that a 3-day exposure window would only capture foods eaten frequently or ones that tended to be heavily contaminated with Salmonella. Fortunately, however, this particular bias is in the direction of the null (ie, towards there being no measured association between specific exposures and disease risk). 3. Generalisability (regarding time periods): Finally, this study was carried out during and identified risk factors most important over that period. Findings may therefore not be completely generalisable to other time periods. Fortunately, the authors partly dealt with this problem by excluding cases infected with S. Brandenburg.

16 Aetiology of Salmonellosis in NZ 16 Other probable limitations that were not detailed in the manuscript include the following: Study power: Although reasonably sized, the study still appeared to lack adequate power to explore certain risk factors eg, consumption of antacids and use of antibiotics (see footnote to Table 3.1). Exclusion of secondary cases: The association of illness with contact with a household member with diarrhoea is likely to have been underestimated because secondary salmonellosis cases were excluded. Questionnaire design: In retrospect there are some areas in which further questioning would have been optimal. For example, there were no data on the handling of raw food for pets (around which there is some concern in overseas settings [Finley et al. 2006], [Anonymous 2000]). There were also no data collected on cleaning practices eg, use of fixed surfaces for chopping (which are more likely to be cleaned by wiping) compared to movable chopping boards. It also appears that there were no specific questions on hand hygiene practices and smoking behaviour. Finally there were no detailed data relating to potential exposure to bird faeces. Some of these possible transmission pathways include cleaning bird feeders, removing bird faeces, touching or handling dead or sick birds, and the eating (eg, by children) of snow, sand or soil, particularly in proximity to a bird feeder (as suggested in a Norwegian study [Kapperud et al. 1998b]). Of note however is that all questionnaires are constrained by time, and excessively long questionnaires can result in lower validity of responses. Social desirability bias: This bias may have played a role given that some questions could have been embarrassing for some respondents (eg, contact with animal faeces and other household members having diarrhoea). However, this would have biased the results towards the null. Overall assessment: This appears to be a relatively high quality study and the national focus is of particular value. Indeed, it is the only case-control study into sporadic salmonellosis conducted to date in the New Zealand setting. The general pattern of the findings is likely to be valid given: (i) biological plausibility for all the risk factors identified; (ii) general consistency with other New Zealand evidence concerning salmonellosis epidemiology; and (iii) general consistency with the findings from other studies of salmonellosis overseas (ie, all the identified risk factors have previously been identified elsewhere). Nevertheless, all case-control studies have limitations (especially around selection bias and information bias) and so the specific quantified findings (eg, ORs and PAR% estimates) should be treated with caution. A key finding was that food consumption and food preparation collectively comprised the largest PAR% (ie, 20.7% [11.1% + 9.6%]). But many other factors (especially contact with bird faeces) also appeared to be relatively important (ie, PAR%=10.7%). However, it is unclear if some of these results, such as contact with bird faeces apply to a particular time period such as when salmonellosis outbreaks in wild birds were of particular importance in New Zealand.

17 Aetiology of Salmonellosis in NZ 17 2) National case-control study of emergent S. Brandenburg Background: This study [Baker et al. 2007b] was conducted in response to the apparent emergence of one subtype of S. Brandenburg which began causing large numbers of human infections in New Zealand in There was also concern by clinicians that this new serotype was resulting in more invasive disease than other forms of salmonellosis [Clarke & Tomlinson 2004]. Methods: The study involved both a case-control study and analysis of combined notification and laboratory data on human and animal disease. The national casecontrol study was conducted from February 2002 to April 2003 and involved 43 cases and 43 controls. Main findings (risk/protective factors): In the univariate analysis (Table 3.2) the strongest effect sizes were seen for animal contact (occupational and household). But the risk of disease was also increased if there were other people in the household with diarrhoea and where the usual method of cleaning the chopping board or other surface was reported as Rinse with hot water only (this was in situations where the same chopping board or other surface used for cutting raw meat or poultry was also reported to be used for preparing other food such as salads and bread). An apparent protective effect was seen for the consumption of various food products (Table 3.2). In the multivariate analysis there were two exposures associated with a significant increase in disease risk: (i) occupational contact with live or dead sheep or lambs during the three days and (ii) having a household member who had occupational contact with sheep or lamb in the three days prior to illness or interview. Together these two exposures were able to explain 52.6% of the PAR% for this infection. Table 3.2. Univariate analysis of selected risk/protective factors for S. Brandenburg infection, New Zealand [Baker et al. 2007b] (Some editing and organisational changes to the table have been made for the purposes of this review). Risk/protective factor (all exposures in 3 days prior to illness) Animal exposures OR (95% CI) P value Occupation contact with live or dead sheep 9.00 ( ) 0.04 Occupational contact with animal carcasses 6.00 ( ) 0.10 Household contact with dog 5.50 ( ) 0.03 Household occupational contact with live animals or carcasses 4.33 ( ) 0.02 Household contact with live or dead sheep 4.00 ( ) 0.03 Dietary exposures Ate sheep or lamb 1.20 ( ) 0.67 Ate imported food 0.41 ( ) 0.05 Ate uncooked vegetables 0.30 ( ) 0.01 Ate unpeeled fruit 0.17 ( ) 0.00

18 Aetiology of Salmonellosis in NZ 18 Risk/protective factor (all exposures in 3 days prior to illness) OR (95% CI) P value Ate pies 0.35 ( ) 0.03 Ate whole chicken 0.40 ( ) 0.06 Ate bacon 0.33 ( ) 0.03 Ate small goods eg, luncheon sausage 0.30 ( ) 0.01 Ate egg 0.27 ( ) 0.02 Ate homemade food outside home 0.31 ( ) 0.02 Ate any dairy product 0.09 ( ) 0.02 Other exposures Other people in household with diarrhoea 8.00 ( ) 0.05 Kitchen chopping board cleaned between uses by rinsing with 3.33 ( ) 0.07 hot water Over the counter medicines 0.22 ( ) 0.01 Anti-inflammatory medicines 0.27 ( ) 0.02 Strengths of the study: Good design: The study was national with matching by geographic area and rurality (using three strata: rural, small town and urban). The time period covered included all four seasons, which is relevant for a disease with marked seasonality. Appropriate analysis: Multivariate analysis was performed. Also population attributable risk (PAR) was estimated for factors associated with illness that were considered causal using an appropriate method (ie, of Bruzzi et al [Bruzzi et al. 1985]). Multiple data sources and contextualisation of the findings: The study combined use of human notification data, laboratory data from humans and animals and a case-control study. It also appropriately considered the results in terms of international evidence relating to zoonotic transmission of salmonellosis. Expertise: The study was performed by public health workers who have previously published other case-control studies and other analytic work in the field of enteric disease epidemiology in New Zealand (including in high quality international journals). Limitations with the study: The authors of this study acknowledged two specific limitations: 1. Selection bias: Selection bias is potentially important because of the difficulty in fully matching controls by rurality. Added to this is the fact that 35 potential control subjects declined to participate. 2. Information bias: Information bias is likely because of the lengthy process of case recruitment. Recall of some exposures, particularly foods, is likely to have been far from complete. The use of a 3-day exposure window also introduces a potential source of bias.. Consequently, this study cannot

19 Aetiology of Salmonellosis in NZ 19 rule out foodborne transmission as making a contribution to S. Brandenburg transmission in New Zealand. Other limitations that were not detailed in the manuscript include the following: Study power: The study was reasonably small (at n=43 cases and controls) and so may have meant that some possible risk factors were not identified (in terms of statistically significant results). Exclusion of secondary cases: The association of illness with contact with a household member with diarrhoea is likely to have been underestimated because secondary salmonellosis cases were excluded. Social desirability bias: This bias may have played a role given that some questions could have been embarrassing for some respondents (eg, touching dead animals and other household members having diarrhoea). However, this would have biased the results towards the null. Overall assessment: This appears to be a reasonably high quality study that integrates multiple data sources and includes a case-control study. The findings appear to be consistent with other New Zealand and international work on transmission of certain serotypes of Salmonella. Indeed, when considering all the epidemiological and laboratory evidence, this study seems to provide strong evidence that S. Brandenburg had emerged at this time as a directly transmitted zoonotic infection in New Zealand. But the study is not particularly informative on other transmission mechanisms. The only food/food preparation exposure result that was statistically significant as a risk factor (in the univariate analysis) was kitchen chopping board cleaned between uses by rinsing with hot water. The food exposures that were protective factors are hard to interpret. Some foods may protect the gut from the infective process, some might reflect displacement effects (ie, eating fruit and vegetables induces satiety and reduces gastric volume for possible high risk foods not identified in this study), and eating these type of foods may be just highly correlated with certain healthy behaviours that are protective (such as regular hand washing). This study was conducted to investigate the increase of a specific serotype of salmonellosis and hence may have limited relevance to understanding salmonellosis in general. Indeed, S. Brandenburg only comprised 6.9% of cases of salmonellosis over the period [Baker et al. 2007b] and was concentrated in four health districts in the southern half of the South Island (ie, notably Southland, Otago, South Canterbury, and Canterbury). In contrast other Salmonella infections are more evenly distributed across New Zealand and they also show a difference in seasonal peaks (eg, late summer for salmonellosis vs spring for S. Brandenburg). The likely role of occupational transmission of this disease also probably explains the higher notification rates for males (compared to salmonellosis in general). The relatively high rates for infants may also reflect exposure to contaminated surfaces such as floors. Since data were collected for this study, S. Brandenburg infections in humans have become rarer, possibly due to agricultural control measures such as vaccination of

20 Aetiology of Salmonellosis in NZ 20 sheep. In 2007 this serotype comprised only 3.7% (47/1267) laboratory-confirmed human cases [ESR 2008b]. In terms of salmonellosis transmission in New Zealand more generally, the following appear to be implications from this study: That some serotypes of salmonellosis in the New Zealand setting can have transmission pathways that are dominated by direct zoonotic transmission. That even when direct zoonotic transmission dominates there is some evidence that other pathways may still relevant eg, relating to kitchen practices and to exposure to sick people. 3) Case-control study of an outbreak of S. Typhimurium DT160 Background: This study [Thornley et al. 2003] was conducted nationally in response to increasing rates of this disease in the community. The epidemic occurred in parallel with an epizootic leading to deaths in wild birds, mainly sparrows, due to septicaemia caused by S. Typhimurium DT160. Methods: Of 170 case-patients meeting the case definition, 119 (70%) agreed to participate and were enrolled in the study. Each case was matched with two controls (n=235 in total) who were found from randomly drawn telephone numbers (matching for neighbourhood and age). Cases and controls were interviewed by telephone. Main findings (risk/protective factors): The study found seven exposures which had statistically significant univariate associations with increased risk for illness: Four of these were for different levels of contact with other people with gastrointestinal illness (ie, within 28 days of illness onset; within 3 days of onset; within the household; or outside the household). Direct handling of dead wild birds. Consumption of fast food. Consumption of food at a large gathering, such as at a party or large barbecue. On the multivariate analysis, the following were independent statistically significant associations: Contact with a person with gastrointestinal illness in the 28 days before onset of illness in the case patient. Handling of dead wild birds. Consumption of fast food. Sampling and laboratory work also identified S. Typhimurium DT160 in roof collected water that was drunk by the cases (4 out of 8 sources). Other Salmonella serotypes were identified on the surfaces of two out of six different brands of eggs tested (out of those consumed by cases who reported eating raw eggs).

21 Aetiology of Salmonellosis in NZ 21 Strengths of the study: Good design: The study was national, reasonably large and with appropriate matching. Appropriate analysis: Multivariate analysis was performed but no population attributable risk percentages (PAR%) were estimated. Contextualisation of the findings: The study considered the associated epidemic in wild birds (mainly sparrows) and appropriately considered the findings in terms of international evidence relating to zoonotic transmission of salmonellosis. Expertise: The study was performed by public health workers who had at the time previously published other analytic work in the field of infectious disease epidemiology in New Zealand (including in high quality international journals). Limitations with the study: The authors of this study acknowledged the following three specific limitations: 1. Selection and information bias: Sampling and recall bias may have influenced the results of this study. Asymptomatic Salmonella carriers would not have been excluded from selection as controls, potentially reducing the magnitude of observed associations. Recall may have been influenced by delays between exposure and interview, although participants were asked to refer to a memory aid (personal diary or calendar). Recall of unusual exposures is less likely to have been affected. 2. Questionnaire design: In our study, information was not collected on exposure to environments contaminated by wild bird faeces, such as parks and play areas, a fact that may have underestimated the avian contribution to human illness. Similarly information on the setting of fast food consumption was not collected eg, Consumption of fast food may have occurred in environments contaminated by bird faeces, or the foods themselves may have been contaminated, either during production or by infected foodhandlers. 3. Exclusion of secondary cases: The association of illness with contact with another person with gastrointestinal illness is likely underestimated because secondary salmonellosis cases were excluded. In addition to the above, social desirability bias may have played a role given that some questions could have been embarrassing for some respondents (eg, touching dead animals and other household members having diarrhoea). However, this would have biased the results towards the null. Another limitation was that the published results were restricted to those showing statistically significant associations. The ideal would have been for more detailed results to have been published (eg, as online extra material). Overall assessment: This appears to be a high quality study that had the benefits of being large and national in scope. The findings appear to be consistent with other New Zealand and international work on zoonotic and foodborne transmission of certain serotypes of salmonellosis.