The study population included all persons living in the Cardiff area who consulted their general practitioner for gastrointestinal symptoms and subsequently submitted a diagnostic fecal sample for microbiologic testing from January 1 through December 31, 2001. Cardiff Public Health Laboratory is the sole laboratory providing a diagnostic microbiology service for the area. All specimens were cultured for Campylobacter spp., Salmonella spp, Shigella spp., and Escherichia coli O157 and examined for ova and parasites, by standard methods. Follow-up specimens from the same patient (<4 weeks after the previous specimen submission date); specimens received from hospital wards and other sites were excluded from study. The study was approved by the local research ethics committee.

Immediately upon receipt of the specimen at the laboratory (next working day), a questionnaire, together with an explanatory letter and a postage-paid envelope, was mailed to the patient. Patients who had not responded within 1 week were sent a reminder letter and provided with another questionnaire on request. The questionnaire asked about basic personal details, including age, sex, employment status, occupation, details of illness, and details of household contacts. It included sections on foreign travel, food and drink eaten, animal contact (pets and farm animals), outdoor leisure activities (gardening, walking, visits to parks or farms, fishing, swimming, and sports), and questions on specific medical conditions and medication (antacids, H2 antagonists, and antibiotics). The food history covered meat and fish, poultry and eggs, vegetables (raw vegetables, leaf vegetables [e.g., lettuce], salad vegetables [e.g., tomato], and prepared salads [e.g., coleslaw]), fruit, milk and dairy products, drinking water (tap water, bottled water, and other sources), and eating out (type of restaurant or takeaway). Participants were asked to respond yes or no and, to the question of exposure for tap water, to indicate the number of glasses drunk per day. All questions related to exposure in the 7 days before the onset of symptoms, except for those on antibiotics, which concerned the month before illness onset.

Case-patients were defined as any patient, not associated with an outbreak, who submitted a fecal sample that was positive for Campylobacter spp. on microbiologic culture. Case-patients were compared with patients whose samples were negative on culture and microscopic examination. Patients with an alternative microbiologic diagnosis were excluded (unless they had dual infection with campylobacter).

Initial univariate analysis was performed with Epi Info software (v. 6.04; Centers for Disease Control and Prevention) to calculate maximum likelihood estimates for Mantel-Haenzel odds ratios (OR) with exact 95% mid-p confidence limits. Continuous variables were analyzed using the t test or Mann-Whitney U test, as appropriate. All reported p values are two sided. Multiple logistic regression models were constructed with Stata software (v. 6, Stata Corporation, College Station, TX). Risk factors were selected a priori on biologic grounds and grouped into four exposure categories: food and drink consumption; animal contact; leisure activities, including foreign travel; and medical history. Logistic regression models were first constructed for risk factors within each exposure category (adjustment A). We then fitted a model that combined all the independent risk factors (for which the Wald test p value for the adjusted OR was <0.10) from the four exposure categories (adjustment B). Finally, to detect any residual confounding, we fitted all personal factors with a p value of <0.10 (age group, presence of a child <5 years of age in the household, and employment status). Of these, only age group interacted significantly with the other terms as tested by goodness of fit and was therefore included in the final model (adjustment C). The population-attributable fraction for each risk factor associated with campylobacter infection was calculated by using methods described by Greenland and Drescher (11).

Questionnaires were sent to 2,694 eligible patients; fecal samples from 346 (12.8%) were positive for Campylobacter spp. (including 4 dual infections: 3 with salmonella infection, 1 with giardiasis). No campylobacter outbreak occurred during the study period. Ninety-one patients (3.4%) were positive for other organisms (42 Salmonella spp., 20 Giardia lamblia, 12 Cryptosporidium sp., 7 Clostridium difficile, 2 Shigella sp., 2 E. coli O157, 1 amoebic dysentery, and 5 other parasites); these were excluded from further analysis. Responses were received from 213 (61.6%) of 346 persons with campylobacter infection and 1,144 (50.7%) of 2,257 persons with negative specimen results. Median delay in response (from date questionnaire sent to date questionnaire returned) was 6 days (range 2–73 days) for case-patients and 7 days (range 1–77 days) for non-case-patients (Kruskal-Wallis H 1.81, p=0.18).

Our study identified five risk factors for campylobacter infection that, if taken together, could account for most sporadic cases. Most important was eating chicken in the 7 days before onset of illness. Two other risk factors, not previously described, could also potentially account for a sizeable proportion of case-patients: eating salad vegetables such as tomatoes or cucumber and drinking bottled water.

The study used a retrospective cohort design that included all patients submitting fecal specimens through their general practitioner to a single laboratory. This design controls for patient characteristics associated with physician-consulting behavior and may also minimize recall bias associated with using healthy controls. Use of a laboratory study population does, however, have several disadvantages. Non-case-patients probably represent a group whose illnesses have disparate cause. Many may have had viral gastroenteritis since this is known to be common in the community and is not detectable by routine laboratory testing. This fact would explain why symptoms reported by non-case-patients were milder. Non-case-patients were also significantly more likely than case-patients to report a history of lower bowel problems, suggesting that some had pre-existing disease that might predispose to non-infectious diarrhea. Antibiotic use in the month before onset of illness was also more common in non-case-patients, and symptoms in these persons may therefore be a side effect of antibiotic treatment. Persons with pre-existing bowel problems may have atypical dietary habits, but neither a history of bowel problems nor of antibiotic use should affect the accuracy of food histories. The multivariate analysis controlled for both these variables.

The most consistent finding in studies of campylobacter infection etiology has been an association with eating chicken. This finding has been described in three previous U.K. studies (6,7,9), and in studies from the United States (12–16), Scandinavia (17–19), the Netherlands (20), Switzerland (21), and New Zealand (22,23). However, the relationship with chicken is sometimes only seen for eating undercooked chicken (12,22,23) or eating chicken away from home (8,15,22,23). Recent microbiologic studies of raw poultry in the United Kingdom indicate continuing high levels of campylobacter contamination in chicken and the occurrence of identical subtypes in both chicken and human isolates (24). Our finding of an association between illness and eating chicken or eating from a fried chicken outlet highlights the fact that chicken remains a major risk factor for campylobacter in the United Kingdom and that measures are needed both in the food industry and at home to promote thorough cooking of chicken and to reduce the risk for cross-contamination.

Neither eating salad vegetables nor drinking bottled water has previously been recognized as a risk factor. In our study, both these associations made a significant contribution to the final logistic regression model and could explain a substantial number of campylobacter infections. Both are also biologically plausible. Salad vegetables could be contaminated with campylobacter either before or after the point of sale. Contamination at the source could occur through contaminated soil or contaminated water during harvesting. Salad vegetables are often imported from abroad, and changes in the sourcing of such items could introduce new vehicles of campylobacter infection into the U.K. food chain. For example, contaminated imported lettuce has been identified as a vehicle of infection in recent salmonella and shigella outbreaks in the United Kingdom (25). However, recent extensive sampling of organic fruit and vegetables and ready-to-eat prepackaged salads in the United Kingdom found no pathogens such as Campylobacter, Salmonella, or E. coli O157, suggesting that contamination of such items is either rare or intermittent (26). Two reports on campylobacter outbreaks associated with salad have been published. Both of these involved contamination in the kitchen. The first was a 3-month long outbreak from cucumber served at a salad bar; the outbreak resolved after changes were made in food preparation and storage procedures (27). The second involved salad prepared by a foodhandler who exhibited symptoms of campylobacter infection (28). In a recent review of outbreaks in England and Wales (including five from campylobacter) linked with salad vegetables and fruit, cross-contamination was also the most frequently identified contributory factor (25). The association we observed was specific to items such as tomatoes and cucumber that require extensive handling during preparation and often the use of a chopping board, rather than with lettuce or with salads bought preprepared. This finding suggests that salad most likely gets cross-contaminated during food preparation.

Natural mineral water is usually obtained from springs and occasionally from borehole sources. In Europe, legislation requires that mineral water be free from parasites and pathogenic organisms but, unlike tap water, it may not be treated in any way that might alter its chemical composition (29). A variety of organisms, including coliforms, can be found in mineral water and will survive for a considerable length of time, particularly in uncarbonated water supplied in plastic bottles or bottled by hand (30). To our knowledge, campylobacter has not been identified in mineral water, but this may simply be because testing for campylobacter is rarely undertaken. Mineral water has, however, been identified in the past as a vehicle of transmission during a cholera epidemic (31) and as a potential source of typhoid fever in travelers (32). More recently, a study of diarrhea in HIV-infected persons found symptoms were significantly associated with drinking bottled or filtered water, whereas drinking tap water was protective (33). Drinking bottled water has also recently been identified as a possible risk factor for campylobacter infection acquired abroad (34), and for Campylobacter coli infection (35). These findings suggest that bottled water could, given the right circumstances, provide a vehicle of transmission for campylobacter.

A small proportion of cases were explained by contact of personswith cows or calves. This occurred exclusively within the context of a farm visit and probably reflects the urban context of our study. This association, though apparently uncommon, is entirely plausible. Occupational contact with animal feces (8), living on a farm (16,19), and contact with cattle (16,23) have all been previously described as risk factors for campylobacter infection. Healthy beef and dairy cattle both excrete campylobacter (36,37), and molecular evidence suggests a link between campylobacter in the farm environment with those causing disease in the community (38).

Our study confirms that eating chicken still plays an important role in the cause of campylobacter infection. It also identifies two potentially important new risk factors that merit further investigation: salad vegetables (and the associated risks of cross-contamination in the home) and bottled natural mineral water. Cross-contamination in the domestic kitchen is potentially preventable, but we need to know how it happens and what interventions are most effective at reducing the risk. Bottled water is a $35 billion worldwide industry (39). In the United States, 1.7 billion gallons of natural mineral water were consumed in 2000 (39). Consumption is also increasing dramatically in the United Kingdom (by approximately 10% each year), and approximately 300 million gallons of bottled water are now consumed annually (40). Consequently, increased illness from contamination of bottled water could be considerable. More studies of the microbiologic quality of natural mineral waters are required, and these should include testing for Campylobacter spp.