INTRODUCTION
Among bacterial causes of gastroenteritis, Campylobacter jejuni and C. coli are important because they are such common pathogens [Reference Ruiz-Palacios1, Reference Kubota2] with a considerable health burden due to acute enteric disease [Reference Havelaar3]. In the European Union (EU) notification rates vary from 0·5 to 70/100 000 (1/year) [4]. In addition, Campylobacter is associated with potentially severe sequelae. Infection with C. jejuni is a strong risk factor for Guillain–Barré syndrome: an immune-mediated disease of the peripheral nerves [Reference Endtz5–Reference Tam7]. Studies of the occurrence of Campylobacter in foods (poultry, pork, raw milk) and the environment (untreated water) indicate that human exposure may be a frequent event [Reference de Wit8, Reference Amar9], perhaps much more frequent than reported cases of Campylobacter enteritis would suggest [Reference Madden, Moran and Scates10–Reference Nauta13]. An important cause for the gap between risk estimates of Campylobacter infection and epidemiological estimates of the incidence of campylobacteriosis is the difference in modality of these two measures. Notification rates are based on reported cases of illness, occasionally corrected for under-ascertainment. Risk estimates can be obtained by translating exposure estimates, based on microbiological surveillance of foods and water into estimates of the probability of infection, including asymptomatic cases, using a dose–response model for infection [Reference Teunis14]. Estimation of the fraction of infections that are symptomatic is difficult, because it may depend on covert properties of the infected host or the infecting microorganism [Reference Teunis14].
Here we present a new method based on simple assumptions that allows analysis of measured antibody levels (as optical densities in an immunosorbent assay) quantitatively, instead of first categorizing them (into ‘positives’ and ‘negatives’). The distribution of antibody titres at the time of sampling is related to the infection rate: the higher the infection pressure the more frequently seroconversion occurs and, as a consequence, antibody titres are shifted towards higher values. When the serum antibody response to infection is known, the incidence of infection (seroconversion) can thus be estimated from the distribution of antibody titres in a cross-sectional serum sample.
MATERIALS AND METHODS
Serological data
To describe the serum antibody response to infection, peak levels and decay rates were estimated from a published longitudinal study of anti-Campylobacter IgA, IgM, or IgG antibodies in symptomatic cases. Longitudinal sera were obtained from a Danish study (1996–1997) of 210 culture-confirmed cases of campylobacteriosis (C. jejuni or C. coli) with blood samples taken about 3 weeks, 3 months, 6 months and 2 years after infection. Ages of patients ranged from 10 to 76 years (median age 33·5 years). Details of the study design can be found in Strid et al. [Reference Strid15]. Serological testing was performed as reported previously [Reference Ang16]: IgA, IgM and IgG antibodies were measured as a ratio against a reference sample [Reference Ang17]. The ELISA measured antibodies against C. jejuni and C. coli but not against non-thermophilic Campylobacter spp.
Cross-sectional samples of sera were obtained from existing serum banks in Denmark, Finland, France, Italy, Poland, Sweden and The Netherlands. Sera from Finland, Sweden and The Netherlands were subsamples of national studies representative of the general population. Three sets of samples from The Netherlands were included, for successive periods (1995–1996, 1998–2002, 2006–2007). Sera from Denmark were also obtained from an existing collection, sampled in Copenhagen and its peri-urban region. Sera from Poland and Italy were obtained by sampling persons consulting health services for reasons unrelated to gastrointestinal problems. In Romania blood samples were collected prospectively, from people attending district medical services with non-gastrointestinal problems, in the course of the present study (September 2007). As a consequence, the sera from Italy, Romania and Poland may not be representative samples of the general population, because they selected for health problems, and excluded gastroenteric illnesses. Further details on the representativeness of samples have been published [Reference Falkenhorst18]. The sampling period and sample sizes are summarized in Table 2.
These sera, sampled from the adult population (aged 18–60 years) in the contributing countries, were analysed with the same ELISA as used for the longitudinal sample, to ensure identical units [Reference Ang17].
In Denmark, Finland, France, Poland, Sweden, and The Netherlands serum banks had been approved by the appropriate ethics committee. Sera from Italy and Romania were left over from blood samples taken for diagnostics, where patients had consented to their use for research purposes. For this reason formal ethics committee approval was deemed unnecessary by the responsible public health institutes. All serum samples were anonymized.
Statistical methods
The serum antibody response to infection consists of a short-term increase (seroconversion) to peak levels, followed by a slow decrease towards baseline [Reference Versteegh19, Reference Simonsen20]. Antibody responses may vary substantially between individual subjects [Reference Teunis21, Reference Teunis22], therefore peak levels and decay rates are specified as distributions, representing the variation between individuals in the population. For use in the marginal cross-sectional model, peak titre and decay rate were characterized by parametric distributions: heterogeneity in peak titre as a gamma distribution, heterogeneity in decay rate as an inverse gamma distribution. Correlation between peak level (A) and halftime (T ½) is weak (0·10, 0·34, −0·003 for IgG, IgM, IgA, respectively), therefore we conveniently assumed A and T ½ (and A and k) to be independent.
Seroconversions (infections) were assumed to occur at random with a given rate, the incidence of infection. Given the response to infection, the distribution of antibody levels in a cross-sectional (snapshot) sample can be expressed as a function of the incidence: when infection is infrequent, low titres are likely; when infection occurs frequently the probability of finding high antibody levels increases. A parametric model assuming that incident infections occur as a homogeneous time Poisson process, with gamma-distributed antibody peak levels and inverse gamma-distributed antibody decay rates was used to obtain likelihood-based estimates of the seroconversion rate [Reference Teunis22].
RESULTS
Longitudinal serum antibody responses to Campylobacter infection are illustrated in Figure 1. Important characteristics of the serological response to infection are summarized in Table 1.
Fig. 1. [colour online]. Longitudinal model output: serum antibody responses of (a) IgA, (b) IgM and (c) IgG to Campylobacter infection. Observed data (circles) and model responses, for each individual person in the study.
Table 1. Longitudinal characteristics of response to Campylobacter infection
A, peak titre; T ½, decay halftime.
Q 0·05 and Q 0·95 indicate 0·05 and 0·95 quantiles, respectively.
Given the serum antibody response, cross-sectional (1-point) serum samples from the surveyed countries were translated into seroconversion rates, shown in Table 2.
Table 2. Estimated yearly seroconversion rates of Campylobacter [maximum likelihood value 
$\hat{\gamma }$ and likelihood-based 95% confidence interval (CI)] from joint (IgG, IgM, IgA) antibodies (also shown: notification rates as reported by the European Food Safety Agency and ratio of seroconversion and notification rates)
Incidence rates (maximum likelihood estimates $\hat{\gamma }$) are given in Table 2 as estimated numbers of seroconversions per person per year. Seroconversion rates estimated from IgA, IgM and IgG data were consistent (judged by the likelihood ratio test): estimates in Table 2 are therefore based on joint IgA, IgM and IgG data. Table 2 also gives notification rates reported to the EU in 2007 (no figure reported for Romania).
Figure 2 shows seroconversion rates for all cross-sectional samples. Error bars indicate (95%) predictive intervals, predominantly reflecting differences in sample size between countries. Differences between countries were small with error bars overlapping.
Fig. 2. Seroconversion rates based on combined antibodies (joint IgG, IgM, IgA).
Age profiles of Campylobacter seroconversion rates were flat (not shown), showing no marked differences across the range tested (18–60 years). Gender differences could not be detected at any age (likelihood ratio test for incidence estimates in males and females).
Comparing the illness rates in Swedish travellers to the seroconversion rates for the same countries did not show strong correlation (Fig. 3). Countries with high illness rates (Poland, Romania) also had high seroconversion rates, but countries with low illness rates did not have low seroconversion rates. This is especially true for Italy and Finland, with high estimated seroconversion rates and very low reported numbers of campylobacteriosis in returning Swedish travellers.
Fig. 3. Comparing seroconversion rates with incidences estimated in returning Swedish travellers (1997–2003) for Finland (FI), Denmark (DK), The Netherlands (NL 1996, 1998, 2006), Italy (IT), France (FR), Poland (PL) and Romania (RM) using data from Ekdahl & Giesecke [Reference Ekdahl and Giesecke27]. Ellipsoids indicate 95% confidence intervals for either estimate.
DISCUSSION
A recent estimate of the incidence of culture-confirmed cases of Campylobacter infections in The Netherlands (2000–2004) was 36/100 000 (1/year) [Reference Doorduyn23]. A report by the European Food Safety Agency (EFSA) on foodborne illnesses in the EU [4] provides estimated rates of notified cases of campylobacteriosis for several EU countries (see Table 2). These estimates of symptomatic illness rates range from 40 to 70/100 000 (1/year) down to 0·5–5/100 000 (1/year) with the lowest reported numbers in Poland. It should be noted that the reported illness rates for France are based on a sentinel surveillance for Campylobacter, where only a small proportion of laboratories submit data and strains, leading to low reported rates of notified illnesses.
The estimated seroconversion rates are several orders of magnitude higher than the notification rates. This discrepancy reflects not only the detection deficit in the surveillance of foodborne diseases [Reference Scallan24], but also the often unnoticed fact that enteric infection may remain completely asymptomatic. For Campylobacter, this was known from a human challenge study [Reference Black25] where human volunteers were given different doses of a C. jejuni isolate; however, other epidemiological studies have also noted considerable asymptomatic carriage of various enteric bacterial pathogens [Reference Amar9, Reference Hellard26].
Recent prior exposure to specific Campylobacter strains may lead to protective immunity as illustrated by illness data collected from Swedish travellers who were exposed abroad to strains they were less likely to have been exposed to at home [Reference Ekdahl and Giesecke27]. Estimated seroconversion rates for Salmonella were strongly correlated with illness rates in Swedish travellers [Reference Falkenhorst18]. The same is not true for Campylobacter (Fig. 3). Campylobacter strains in Finland may not differ much from those in Sweden, so that most infections in the Swedish travellers visiting Finland remain asymptomatic, but it remains unclear why so few Swedish travellers return from Italy with campylobacteriosis. In contrast, illness incidence does seem to correlate with prevalence of Campylobacter in broiler flocks [Reference Havelaar28].
Asymptomatic infections may add to the public health burden, as complications and/or sequelae may also result from infections that do not result in acute enteric illness. Most asymptomatic infections would, however, not result in any health effects. Sero-incidence therefore is not a good indicator of health burden. Because seroconversion is closely related to exposure (more so than symptomatic illness) sero-incidence is a good indicator of infection pressure. The rate of ingesting a dose of Campylobacter high enough to cause infection informs about the presence of these pathogens in the food and other environmental compartments.
Whether a seroconverting subject moves on to symptomatic illness may depend on prior experience: an episode of campylobacteriosis may lead to transient protective immunity, possibly not preventing seroconversion, but protecting against symptomatic illness [Reference Teunis14]. In immunodeficient subjects this protective response may be impaired. Protection against illness in recently infected subjects may lead to an interesting relationship between infection pressure and illness incidence [Reference Swart29].
The cross-sectional serum samples collected in some of the surveyed countries may not be completely representative of the general population in those countries. Consideration might also be given to data collected in different time-frames (different years, or even different seasons) representing different exposures. The small differences between estimated incidence does not show any indication that such differences in exposure may be present.
A mismatch between longitudinal and cross-sectional units would cause a bias in the seroconversion rate estimates. If the cross-sectional data are measured in smaller units than the longitudinal data (so that the same number indicates a smaller ‘true’ antibody titre) the incidence is overestimated. If the cross-sectional data are measured in larger units the incidence is underestimated.
Longitudinal data have been collected for symptomatic cases, because only then is symptom onset known. In contrast the cross-sectional data were collected in the general population, from asymptomatic individuals. While we cannot know whether the serum antibody response to symptomatic infection is identical to the response to asymptomatic infection, it is unlikely that they are very different. If the magnitude of the longitudinal (symptomatic) response were higher than that of the cross-sectional (asymptomatic) data, such a difference would be interpreted as a lower incidence. This means that then the estimated incidence would be an underestimate, which is unlikely given the high incidence estimates. The converse: higher levels in the cross-sectional (asymptomatic) data than in the (symptomatic) longitudinal data would mean that the estimated incidence is biased upwards, but high titres in the cross-sectional sample would then exceed the highest peaks in the longitudinal response patterns (based on the longitudinal observations they would not be expected to ever occur) and the model would simply fail. Since that does not happen it is concluded that there is no reason for concern that asymptomatic responses are very different from the observed symptomatic responses.
ACKNOWLEDGEMENTS
The contribution of P.F.M.T. to this study was funded by ‘POLYMOD’, a European Commission project funded within the Sixth Framework Programme (contract number SSP22-CT-2004-502084). The contributions of Y.T.H.P.Vd., W.vP. and G.F. were funded by Network of Excellence ‘MedVetNet’ funded within the EU Sixth Framework Programme (contract no. 506122) WP32: Public Health surveillance for foodborne infections: design of epidemiological studies and applying seroepidemiology to validate the surveillance pyramid. The authors gratefully acknowledge the assistence of Dr Karin Krogfelt (Statens Serum Institute, Copenhagen).
DECLARATION OF INTEREST
None.
