INTRODUCTION
Non-typhoidal salmonellosis continues to be a substantial burden on public health in the Western world. Lower Saxony reports about 2000 notifications (25/100 000 inhabitants) each year, the majority being sporadic cases with an unknown source of infection [1]. Within Germany, the predominance of the two major serovars, Salmonella Typhimurium and S. Enteritidis, strongly differs between geographical regions [Reference Held2]. Since 2009, S. Typhimurium has been the most frequently reported serovar in Lower Saxony accounting for 41% of all notified cases in 2012 [3]. In contrast to S. Enteritidis infections, which are mainly associated with egg and poultry consumption [Reference Banatvala4–Reference Mølbak and Neimann9], risk factors and infection routes for S. Typhimurium serovars are less clear [Reference Delarocque-Astagneau10]. Associations with exposures that differ from the classical foodborne transmission [Reference Doorduyn6, Reference Dore11] substantiate that the agent seems to be widespread in the environment rather than in food only. Besides shifts in serovar prevalence, potential changes in food production industry or consumers' behaviour such as increasing popularity of exotic foods and frequent foreign travel justify re-assessment of risk factors in the same population, in order to tackle prevailing challenges of foodborne hazards [Reference Käferstein, Motarjemi and Bettcher12]. To extend our knowledge about recently described risk factors [Reference Ziehm13, Reference Ziehm14] and to assess the impact of hygiene in the domestic environment, we performed a case-control study on sporadic salmonellosis in Lower Saxony. We matched cases and controls for county district, age and sex. Besides an overall analysis, S. Enteritidis and S. Typhimurium infections were also considered separately.
METHODS
Study design
Sampling of cases and controls was performed between January 2012 and November 2013. Out of the 46 county districts of Lower Saxony, 14 were selected to participate in the study. We aimed at obtaining a representative sample with respect to population density and location within the Federal state. In order to take regional eating preferences or other behaviours into account, we included county districts that had reported a high proportion of infections with either S. Typhimurium or S. Enteritidis serovars in previous years. Cases were recruited from Salmonella reports of the German infectious disease notification system. Local health departments (LHDs) delivered the questionnaires to persons with laboratory-confirmed Salmonella infection and clinical symptoms (diarrhoea, abdominal pain, vomiting, fever). To ensure that only sporadic cases were included, patients were excluded from enrolment in the study if they were part of an outbreak identified by the LHDs' on-site investigations or if they affirmed a question about recent diarrhoea occurring in their contact persons. Controls were selected from a random sampling list provided by the registration offices in the respective areas and received the questionnaire by mail. For each individual case, four controls were selected, matched for age (±3 months), sex and county district. Observation periods were dated 3 days backwards from the day on which the first symptoms of disease were observed in the respective cases. Observation periods of cases and controls were matched for days of the week in order to account for possible differences between weekends and workdays.
Questionnaire
Cases and controls were invited to complete a self-administered questionnaire. Besides basic demographic information (e.g. level of professional education, number of household members, migration background), questions covered leisure activities, contact with animals and various food exposures during the 3 days preceding the illness, in particular consumption of different types of meat, cooking condition (raw, medium, fully cooked) and whether the food was consumed at home or while eating out. Medication over the last 4 weeks and comorbidities were also assessed. Detailed information was collected about consumption of food items that were obtained from either delivery services or takeaways. Other questions covered job-related contact with animals (e.g. farmer, veterinarian), small children (e.g. nurse) or raw meat (e.g. butcher) of the respective person and other household members. Questions about kitchen hygiene as well as practices of meat preparation and consumption were not specifically addressed during the observation period, but inquired about in general routine habits in the respective household. Questionnaires were only used if the attached consent form was signed by the index person or by the parent/guardian of an under-age index person. In some cases, oral consent was obtained via phone. Information about serovars was collected from the reports of the LHDs.
Statistical analysis
Microsoft Access2003 (Microsoft Corp., USA) was used as the database system and Stata statistical software, release 12 (StataCorp., USA) was used to conduct the statistical analyses. Before data entry, questionnaires were controlled for completeness and plausibility. We performed a conditional logistic regression analysis to consider the 1:n pair-matched design. Therefore, two versions of odds ratios (OR) are reported: The conditional OR taking into account the matching for sex, age and region and an adjusted OR accounting for multivariable risk factors. Based on the outcome of the univariable analysis, variables with P < 0·2 and more than four exposed subjects in each group were selected for the initial multivariable logistic regression analysis. The cut-off value for step-wise backward elimination of non-significant variables from the model was P ⩾ 0·2. In all analyses the level of significance α was set to 0·05 (two-sided). To avoid covariation in variables tested in the multivariable logistic regression analysis, certain exposures were combined to construct composite variables. For example ‘consumption of meat outside the home’ consisted of consumption of poultry, pork, beef (or veal) or lamb outside the home. Associations (Cramer's V) between variables selected for multivariable testing were <0·3. Due to the explorative nature of this study, multiplicity correction to control the number of false-positive results, as well as modelling of interactions were omitted. Sensitivity analyses were performed to validate the final model. The results of the multiple logistic regression analysis were quite stable. Backward and forward selection procedures as well as omitting certain variables did not lead to considerable changes in the calculated OR for the remaining factors.
Ethical approval
The study was exempted from ethical approval by the ethics committee of the Hannover Medical School.
RESULTS
Between January 2012 and November 2013, over 3500 questionnaires were dispatched. Response rates were 60% (425/716 questionnaires) in cases and 28% (800/2864) in controls. Seventy-nine controls were excluded as they reported to have suffered from diarrhoea recently. The final dataset comprised 1127 participants (416 cases, 711 controls) resulting in 285 matched pairs (732 observations). Mean (±s.d.) time lag between cases' and controls' observation periods was 14 (±9) days. Serotyping was performed in 398 isolates. S. Typhimurium serovars were found in 180 (43%) cases, S. Enteritidis serovars in 80 (19%) cases and S. Infantis in 22 (5%) cases. An overview of the sociodemographic characteristics of cases and controls (non-matched dataset) is given in Table 1. Multivariable logistic regression analysis revealed significant associations for consumption of raw ground pork, intake of gastric acidity inhibitors and eating meat outside the home. Affirmative answers to the questions regarding kitchen hygiene were generally more frequent in cases than in controls, and in particular, a significant association between sporadic salmonellosis and the statement that the dishcloth in the respective household was changed daily or cleaned by using a minimum washing temperature of 60 °C on a daily basis was detected. Furthermore, significant associations with OR <1 were detected for animal contact and consumption of ice cream (see Table 2). Serovar-specific risk-factor analyses of 124 cases with sporadic S. Typhimurium infections and their respective controls (192 controls; 316 observations) revealed a significant association for raw ground pork consumption in the multivariable analysis (see Table 3a ). S. Enteritidis infections were significantly associated with having travelled abroad in the multivariable analysis of data from 57 cases and 96 controls (153 observations; Table 3b ).
aOR, Adjusted odds ratio; CI, confidence interval.
Variables that are significantly (P⩽0·05) associated with the disease appear in bold.
† Odds ratio accounting the matching for age, sex, region.
‡ Odds ratio accounting the multivariable risk factors.
* Variable removed (P > 0·2) from model; **variable not used in final model; ***variable integrated in composite variable.
aOR, Adjusted odds ratio; CI, confidence interval.
Variables that were significantly (P ⩽ 0·05) associated with the disease appear in bold.
† Odds ratio accounting the matching for age, sex, region.
‡ Odds ratio accounting the multivariable risk factors.
* Variable removed (P > 0·2) from model;
** variable not used in final model.
DISCUSSION
The majority of reported salmonelloses in Germany are sporadic [1] and the source of infection often remains unidentified. In order to develop, implement and update successful control and prevention strategies, a proper understanding of the current epidemiology and potential risk factors of a disease is necessary. To this aim, we investigated notified cases with sporadic salmonellosis in Lower Saxony. Controls were matched on age, sex and region, to exclude these potential confounders. In the present study, 60% of the cases and 28% of the controls returned the questionnaire. Although we approached four controls for each case, still 131 cases remained without a matched control. A similarly designed study on yersiniosis that also used self-administered questionnaires reported response rates of 42% for cases and 36% for controls [Reference Rosner15]. In a previous study on sporadic salmonellosis, telephone interviews could be performed successfully with 51% of the recruited Salmonella cases whereas only 30% of controls consented to an interview [Reference Ziehm13]. These data indicate that in controls, response rates are low for both oral and written questionnaires. The serovar distribution found in our study closely reflects the general distribution of serovars in all notified cases in Lower Saxony during the study period (2012: 43% S. Typhimurium; 23% S. Enteritidis; 3% S. Infantis; 2013: 44% S. Typhimurium; 16% S. Enteritidis; 7% S. Infantis; [3]), emphasizing the significance of S. Typhimurium serovars for sporadic human infections in this region. Eating ‘mett’ (raw pork finely minced or ground and used as a spread on bread) is very popular in the northern and eastern regions of Germany, and the current risk-factor analysis revealed the highest OR for this food exposure. In line with previous findings [Reference Ziehm14], a significant association with S. Typhimurium infections for consumption of raw ground pork was also detected. Prevalence in German slaughter pigs as well as transmission of the pathogen along the food chain have been documented [Reference Käsbohrer16, Reference Hauser17], emphasizing the relevance of pork as a source for Salmonella infections in the German population [Reference Ziehm13, Reference Ziehm14, Reference Ziehm18].
For sporadic S. Enteritidis infections, foreign travel was identified as a risk factor, whereas in contrast to earlier studies [Reference Banatvala4, Reference Delarocque-Astagneau5, Reference Ziehm14] consumption of food items containing raw eggs was not significantly associated with the disease. Across Germany, overall numbers of notified S. Enteritidis infections in humans have been decreasing during the past decade, reflecting the decreasing prevalence of this serovar in laying hen flocks and hence eggs [Reference Frank19]. This decline is interpreted as a consequence of the recently implemented national Salmonella control and monitoring programmes in egg production sectors in accordance with regulation (EC) No. 2160/2003 [20]. Our current findings demonstrate the first epidemiological evidence at the population level for the success of these measures in Germany. Similarly, human S. Enteritidis infections are mostly travel-related in other European countries with low prevalence of S. Enteritidis in laying hen flocks [Reference Mølbak and Neimann9, Reference Emberland, Nygård and Aavitsland21].
In line with our current findings, taking antacids has been identified previously as a risk factor for sporadic salmonellosis [Reference Banatvala4, Reference Doorduyn6, Reference Ziehm13, Reference Jansen22]. Intake of antibiotics seems to favour human S. Typhimurium [Reference Delarocque-Astagneau10, Reference Dore11] but not S. Enteritidis infections [Reference Mølbak and Neimann9], probably due to different antimicrobial resistance profiles in serovars [Reference Doorduyn6]. In our dataset, treatment with antibiotics was significantly associated with S. Typhimurium infections in the single-factor analyses, but the association failed significance in the multivariable analysis.
In the past, ice cream has been identified as a vehicle of S. Enteritidis infection, mostly due to cross-contamination with raw eggs [Reference Mølbak and Neimann9, Reference Dodhia, Kearney and Warburton23]. In contrast to these observations, our current analysis revealed a negative association between ice cream consumption and sporadic Salmonella infections, which is difficult to interpret. Although contaminated eggs are rarely found nowadays [20], which minimizes infection risk, a truly protective effect of ice cream consumption does not seem biologically plausible and lacks biomedical evidence. Whether ice cream consumption is associated with another, still unidentified protective exposure or whether subjects who consume ice cream reduce their exposure to other risk factors requires further exploration. It should be mentioned that negative associations, in particular for food exposures, are occasionally found in case-control studies and are usually considered as artefacts or chance associations [Reference Doorduyn6, Reference Kimura7, Reference Dore11, Reference Bassal24–Reference Sethi26].
A variety of domestic and wildlife animal species can host and shed non-typhoidal Salmonella spp., thereby contributing to human infections [Reference Hoelzer, Moreno Switt and Wiedmann27]. However, our current data, indicate a lower risk of illness for people that had experienced animal contact, which could signify a protective effect. It is conceivable that continuous exposure to foodborne pathogens reduces clinical symptoms due to acquired immunity. The incidence of acute gastrointestinal illness caused by Escherichia coli-polluted drinking water decreased with years of residence in the respective area [Reference Strauss28]. The risk of E. coli and Salmonella infection via animal contact has also been found to decrease with age [Reference Ziehm18, Reference Werber29]. During childhood, clinical or subclinical Salmonella infections caused by repeated animal contact have the potential to modulate the immune system [Reference Pelosi30], thereby reducing the risk for allergic sensitization [Reference Riedler31]. On this evidence, a certain protective effect of animal contact against human salmonellosis appears biologically plausible. On the other hand, pet owners have been found to visit their physician less frequently [Reference Headey32], which could have led to underreporting in the current study. The fact that previous studies identified animal contact as risk factor only when notified Salmonella cases and notified rotavirus cases were compared, but not against healthy population controls [Reference Ziehm13, Reference Ziehm18] also indicates a certain pre-selection of notified cases regarding this exposure. Although negative associations between animal contact and other exposures were not detected in the present study, it is also conceivable that pet owners reduce their exposure to other risk factors, for example by travelling less or eating meat outside the home less frequently.
Univariable analysis revealed significant associations between salmonellosis and consumption of poultry, pork, beef (or veal) and lamb outside the home. Due to the high association within these variables, however, not all types of meat could be included in the multivariable model, which found a significant association between consumption of meat outside the home and salmonellosis. It is therefore not possible to discriminate between different types of meat, but all sorts of meat can be contaminated and act as infection vehicles for salmonellosis [20], especially if inadequate food handling behaviour is performed [Reference Redmond and Griffith33]. Similar to our data, eating out or eating chicken prepared outside the home have been identified as risk factors for salmonellosis previously, whereas consumption of food prepared at home has been found to be protective [Reference Kimura7, Reference Marcus8]. These reports are, however, in contrast to the assumption that the majority of foodborne infections occur in private settings [Reference Azevedo34]. Hygiene practices of consumers strongly affect microbial load in prepared meals [Reference Fischer35], highlighting the importance of adequate food-handling behaviour in the domestic environment [Reference Redmond and Griffith33]. We therefore chose to assess hygiene behaviour of cases and controls in our questionnaire and found that hygienic behaviour was reported more frequently by cases than controls. Changing the dishcloth on a daily basis or cleaning it daily by using a minimum washing temperature of 60 °C was significantly associated with sporadic salmonellosis in the multivariable analysis. One possible interpretation of this finding is that cases behaved more hygienically than controls. Salmonella spp. can develop resistance to disinfecting agents [Reference Braoudaki and Hilton36], but whether this also increases their virulence remains unclear [Reference Karatzas37]. It is conceivable that people who do not clean their dishcloth thoroughly are constantly exposed to small doses of bacteria, which could enhance their immunocompetence [Reference Strauss28, Reference Pelosi30]. A certain limitation of the study is that the respective question did not cover any details regarding the washing procedure, so it was not possible to assess the efficacy of the applied practice. On the other hand, cases tend to retrospectively overemphasize their hygiene precautions by describing their personal behaviour in a more favourable light [Reference Nederhof38]. Assessment of hygiene behaviour via questionnaire is difficult and results can be inconsistent with the actual microbiological findings [Reference Azevedo34]. Discrepancies between self-reported and observational data on hygiene behaviour are also common [Reference Redmond and Griffith33, Reference Fischer35]. In a previous investigation, data on hygienic behaviour collected via telephone interviews were biased by social desirability [Reference Ziehm13]. In the present study we therefore aimed at minimizing social desirability bias by using self-administered questionnaires. Thereby, results cannot be biased by interviewer presence and the perceived anonymity is higher compared to telephone interviews [Reference Nederhof38]. To avoid recall bias, our questions about kitchen hygiene as well as practices of meat preparation and consumption were phrased in a very general way and did not particularly address the observation period. Our current findings are in contrast to the widely held belief that poor hygiene is a risk factor for salmonellosis. A truly protective effect, however, should be biologically plausible, with underlying mechanisms clarified. To discover whether cases and controls really differ in their hygiene behaviour or whether they just answer differently is the aim of our future research.
CONCLUSION
In conclusion, our results confirm that different risk factors for different Salmonella serovars exist. In regions such as Lower Saxony where raw ground pork consumption is popular, these food items must be considered as the major vehicles of infection with S. Typhimurium serovars. Public education should aim at reducing exposure of vulnerable groups, such as the elderly, children, pregnant or immunocompromised persons to raw pork products. Control measures in farm-to-retail processing may include reducing Salmonella prevalence in pigs and minimizing the risk of contamination at all stages of the food chain [Reference Käsbohrer16]. Raw eggs now appear to be of decreasing relevance for infections with S. Enteritidis serovars in Germany, leaving foreign travel as the major risk factor. Advising travellers to avoid potentially contaminated dishes and beverages might at present be the most promising intervention. The fact that suppressing gastric acid production medically can increase the risk of foodborne infections might not be generally known and should be communicated by practitioners when antacids are prescribed. Our current results do not imply that animal contact or poor domestic hygiene increase the chance of acquiring sporadic Salmonella infections, but the potential bias of self-reported data on hygiene behaviour [Reference Redmond and Griffith33–Reference Fischer35] must be taken into account. Our findings also illustrate that a sound interpretation of associations found in case-control studies as risk factors or protective effects partly relies on accompanying microbiological and clinical investigations that elucidate the virulence mechanisms of an infectious agent as well as the pathogenesis of a disease.
ACKNOWLEDGEMENTS
We thank the participating health departments and all study participants. We are indebted to Nicola Jahn for her excellent technical assistance and thank Elke Mertens for stimulating discussions. This work was supported the German Federal Ministry of Education and Research (BMBF) through the German Aerospace Center (DLR), grant number 01KI1012D (http://fbi-zoo.net/).
DECLARATION OF INTEREST
None.