To our knowledge, this population-based sentinel surveillance system for campylobacter infection is unique because we have successfully linked detailed epidemiologic exposure information with detailed microbiologic strain characterization for a large sentinel population. Campylobacters are widely distributed in the environment, and this genus is adapted to a wide range of ecologic niches throughout the food chain (22)
. Microbiologic data show that the prevalence of different campylobacter species and subtypes varies between different potential sources of infection, including different animal species, foods, and water (23
). Although C. coli
infection accounts for a small proportion of laboratory-confirmed human campylobacter cases in England and Wales, the potential for prevention is substantial if the true population burden is much higher (3)
. Most case-control studies have so far sought to determine risk factors for sporadic infection with campylobacter and have not sought to differentiate between species (5
). This distinction is important if C. coli
and C. jejuni
differ in their etiology or if the contribution of similar risk factors differs between the two species. If exposures are aggregated for different pathogenic campylobacter species, the contribution of risk factors unique to or predominantly associated with C. coli
will be masked by the predominance of C. jejuni
(in the study population: C. jejuni
: C. coli
approximately 10:1). This source of bias can be overcome by comparing the exposure characteristics of cases with C. coli
infection with those of cases with C. jejuni
infection. The data for cases with C. jejuni
infection are then used to contrast with, rather than dilute, any observations for C. coli
infection. Therefore, in generating hypotheses for infection, we identified potential species differences by adopting case-case analysis.
Hypothesis: Bottled Water
Case-patients with C. coli
infection were more likely to report bottled water consumption than were those with C. jejuni
infection. This observation is biologically plausible. Raw water can be contaminated with C. coli
) and, while European legislation governing the marketing of natural mineral water makes it a condition that it be free from parasites and pathogenic organisms (30)
, testing for campylobacters is rarely undertaken (31)
. As the bottled water industry is large ($35 billion a year worldwide [3
2]) and expanding rapidly (consumption in the United States, which was 5 billion gallons in 2000, is predicted to increase to 7.3 billion gallons in 2005 [32
]), an accurate assessment of the risk associated with these products is required. Our hypothesis-generating questionnaire did not distinguish between types of bottled water (e.g., spring or mineral, carbonated, or still), but these issues merit further investigation by case-control study.
The finding that having eaten pâté was more likely to be reported by case-patients with C. coli
infection than those with C. jejuni
infection is also biologically plausible. Pork is often the main constituent of pâté, and C. coli
is found in pigs (33)
. In a recent study of the occurrence of campylobacters in 400 freshly eviscerated porcine liver samples, 6% were infected with Campylobacter
spp; most (67%) were C. coli
. Pâté is a perishable comminuted meat product containing nitrite, and possibly nitrate, ascorbate, or both (35)
. While the use of such preservatives might deter the growth of spoilage microorganisms (assuming adequate storage conditions are maintained), vegetative pathogens might not be destroyed; therefore, the ultimate critical control point during production is likely to be effective heat treatment.
Hypothesis: Meat Pies
The fact that retired people with C. coli infection were more likely to report having eaten meat pies is interesting. The types of meat in the pie fillings are not known, but the finding might point to the use of cheaper cuts of meat in these products.
Hypothesis: Foreign Travel
Persons from a South Asian ethnic background who had traveled abroad in the 2 weeks before onset of symptoms were more likely to have acquired a C. coli infection, but the reverse was true for men. This finding probably reflects the fact that travel abroad is simply a marker for activities or behavior while abroad, and a further study of the “travel cohort,” generated through the surveillance scheme, might provide a better indication of where the risks lie.
Campylobacter infection has marked seasonality, and case-patients infected with C. coli were less likely to be ill in the summer than those infected with C. jejuni. As data accumulate, generating season-specific hypotheses might be possible, which may have implications for the time period over which analytic studies are performed.
Sources of Bias
In interpreting the results from the sentinel surveillance system, likely sources of bias should be considered. Selection bias has been minimized by including all laboratory-confirmed cases of campylobacter infection identified by PHLS and National Health Service laboratories in the participating districts. Furthermore, both groups in the case-case comparison have been subjected to the same selection process, so selection bias should not influence our analysis.
The effect of time delays in reaching the patient, and hence recall bias for reported exposures, should be limited by close collaboration between the various participants in the scheme. While the time delay reported in this study introduces some recall bias, there is no reason to believe that recall is operating differently among patients infected with different species or among exposure groups, so that recall bias should not influence the case-case comparison.
Interpreting Case-Case Analyses
A detailed account of the pros and cons of case-case analysis is provided by McCarthy and Giesecke (16)
, but two important points influence the interpretation of this type of study. The first is that exposures that are a risk for infection for both comparison groups will not be identified or might be underestimated. By using patients with campylobacter infection, albeit with a different species, as “controls,” we may obscure an association with the infection of interest because the controls might share some of the risk exposures with the cases. Thus, exposures common to both infections are controlled for by the study design.
The second is that traditionally controls are selected to provide an estimate of the exposure prevalence that would be seen in the cases if there were no association between the exposure and disease. Since our controls have been differentially selected by factors that are related to certain exposures, they might not be representative of the exposure prevalence of the population group from which the cases originated. We cannot, therefore, use comparisons between our cases and controls to make statements about the magnitude or direction of population risk.