We compared monthly listeriosis data from England and Wales with temperature records from 1989 through 2007 to determine the influence of various potential predictors on the number of listeriosis cases. UK Health Protection Agency (HPA) data listing total monthly cases of human listeriosis in England and Wales during 1990–2007 (4,5
) are aggregate. All age categories and regions were included and were collated by the HPA Centre for Infections from voluntary reporting by microbiology laboratories and from referrals of cultures. These publicly available data were also validated by the HPA, and in our analysis we used revised figures based on that validation. Our analysis covered the period from 1990, when active surveillance of listeriosis began, through 2007. Pregnancy-associated cases (mother and neonate) were counted as 1 case. Undated cases that could not be assigned to a particular month were excluded from analysis. We used the UK Met Office mean monthly area temperature time series for 1989–2007 and 30-year means averaged during 1961–1990 (6
Exploratory linear regression analyses suggested a positive correlation between the number of listeriosis cases and the monthly mean UK ambient temperature, as well as suggesting a change in this relationship after 2000 (p = 0.001; , panel A). However, residual variability was not constant, and the monthly counts are likely to be overdispersed due to clustering of cases (3
). The data were fitted again by using a negative binomial generalized linear model with a logarithmic link function, a common model for time series of foodborne illness cases (7,8
). To separate seasonality of listeriosis rates from dependence on temperature, we considered the 30-year mean monthly temperatures from 1961–1990, as well as monthly temperature anomalies (observed mean temperature minus 30-year mean). To determine whether temperatures could have a delayed effect on listeriosis incidence, we also included mean and anomaly temperature variables lagged by 1 or more months. To allow for 2 break points at which the incidence of listeriosis may have suddenly changed, dummy variables were used to represent periods before, between, and after the months in which these increases might have occurred. A best-fit model was selected according to the corrected Akaike Information Criterion (AIC) (9
) by using stepwise regression at all combinations of 2 break point months from January 1996 through December 2007. To examine changing effects on incidence, interactions between break point indicators and other variables were considered, even if the main effects were not yet in the model. Main effects were subsequently added to the final model if any interaction terms were included.
Figure 1 Exploratory analyses of changing rates of listeriosis. A) Listeriosis cases compared with mean observed monthly UK temperatures, 1990–2000 (triangles) and 2001–2007 (circles). Shown are an increased overall incidence in 2001–2007 (more ...)
The best-fit model included 2 break points (, ; , panel B; ). The overall rate of cases increased by 40.83% in April 2001 (adjusted p = 0.001, 95% confidence interval [CI] 17.29%–68.93%); a further increase of 47.76% occurred in March 2003 (adjusted p<0.001, 95% CI 27.92%–71.13%). According to the corrected AIC, there is statistical evidence of changes in the incidence of listeriosis in a range of months around these best-fit values, and weak support for a change in early 2005, apparently due to low numbers of cases in the first months of that year and the following winter (, panel B). Most of the seasonality in the number of cases was accounted for by the 30-year mean monthly temperatures. Each extra degree Celsius of mean monthly temperature corresponded to a 2.42% increase in cases in the current month (adjusted p = 0.044, 95% CI, 0.55%–4.33%) and a 4.09% increase in the following month (adjusted p<0.001, 95% CI 2.19%–6.03%). The 1-month lag reflects delays between food production and consumption plus the known long incubation of listeria infections. Similar lags have been observed for other enteric pathogens (6,11
). A relationship also appears to exist between the rate of listeriosis and the 1-month lagged temperature anomaly starting in April 2001, which corresponded to an additional 5.71% of listeriosis cases per degree Celsius (95% CI –2.10%–14.15%) when compared with the main effect of temperature anomaly in 1990–2007. This increase was not significant (adjusted p = 0.312), although the power of this test is only ≈15% at an adjusted significance level of 0.05 (post hoc power analysis performed by computer simulation of the fitted model in the R statistical computing environment). The model also includes a slight negative trend in cases over time (–0.13% per month, 95% CI –0.26%–0.01%), but evidence for this trend is very weak (adjusted p = 0.188).
Coefficient names and descriptions for the best-fit negative binomial generalized linear model of listeriosis incidence
Estimated coefficients for terms in the best fit model for the monthly incidence of listeriosis*†
Figure 2 Monthly listeriosis cases and temperature observations, England and Wales, 1990–2007. The monthly number of listeriosis cases (circles, lower plot) is strongly seasonal, with a close relationship to the monthly mean temperature (solid line, upper (more ...)
The changes in the incidence of listeriosis appear to have been quite sudden (). Causative explanations based on gradual demographic or behavioral changes have previously been ruled out (3
), and those based on dynamic processes seem unlikely because no evidence exists for epidemiologic feedback between the source of infections and clinical cases. One possibility is that the increases are due to contamination of a small number of food products, as has been suggested as an explanation for an upsurge in listeriosis rates in the late 1980s. However, a restricted range of strains was responsible for most of the additional cases at that time (12
), and no evidence exists of such a pattern since 2001 (3
). An alternative possibility is that the phenomenon is a consequence of changes in government policy or business practices that have had more widespread effects on food processing, distribution, or preparation.
We found notable coincidences between the dates of the increases in listeriosis infection rates and the dates of events associated with the 2001 foot and mouth disease (FMD) outbreak in the United Kingdom. The April 2001 increase in listeriosis rates occurred shortly after the outbreak of FMD in February 2001, allowing for a delay similar to the incubation period of listeria infections. The March 2003 increase in listeriosis rates occurred in the same month as the relaxation of movement restrictions on livestock instituted after the 2001 FMD outbreak (13
). The diverse and widespread consequences of the 2001 FMD outbreak are well-documented (14,15
), and it seems plausible that such major disturbances to agricultural production could be the ultimate cause of the large increases in listeriosis rates that have been observed. The coincidence of these events raises the possibility that the change in listeriosis rates would be an unrecognized outcome of the 2001 UK FMD crisis, although we caution that our analysis of the of these cases does not demonstrate whether a causative link exists.