Pathogen seasonality and links with weather in England and Wales: a big data time series analysis

Cherrie, Mark P. C.; Nichols, Gordon; Iacono, Gianni Lo; Sarran, Christophe; Hajat, Shakoor; Fleming, Lora E.

The limitations of the big-data approach in this analysis meant that it was not possible to undertake analysis on causative weather factors on pathogen incidence. Behavioural determinants that correlate with season and weather may explain the correlations found. For example, school closures for holidays can reduce transmission and therefore cases of influenza [21] , outdoor eating, when the temperature is higher increases risk of Salmonella, undercooking, raw meat contamination and recreational activities on water, are more likely to occur in summer, are associated with Campylobacter [22] . In separate work we are looking at methods to separate out the weather parameters from seasonality (and the associated behavioural determinants) using local weather data linkage, as described in 'recommendations for future research' [23] . The study was limited by the temporal and spatial aggregation of the data, and therefore we were unable to investigate the effect of day-to-day weather in regions of England and Wales. The results of the analysis were also dependent on the time-period used. For example, C. difficile have been reported to have a strong seasonal pattern previously using hospital episode statistics from England from 1995 to 2006 [24] ; however we did not find a strong seasonal component in our study period. In our analyses, C. difficile displayed a peak in 2006 and then reduced in prevalence and seasonality. Therefore, the results are presented with a caveat that the correlation coefficients with weather were sensitive to the time-period under analysis and would be expected to differ in a pathogen-dependent manner. The surveillance methods for collecting data changed over the years, with many pathogens having separate expert surveillance datasets that are independent of this data and some periods of enhanced surveillance or poor surveillance. There have also been periods where an intervention (e.g. vaccination) had been introduced, as well as those where the surveillance had improved (e.g. fungal infections; hospital infections), although we were unable to systematically account for these changes in the current analysis. Furthermore, the data were lab-confirmed and therefore do not represent milder unreported or undiagnosed cases which may display a different pattern of seasonality. Finally, we could not ascertain concomitant pathogens as they were not readily extractable from the database. The analysis was limited as it only considered a 1 month lag effect and did not consider time-varying confounders. Lag effects can vary for different environmental exposures. For example sunshine will induce 25-hydroxy-vitamin D production (the major circulating form of vitamin D) in human skin; 25-hydroxy-vitamin D will lag sunshine exposure by up to 2 months due to metabolism within the body [25] . Also, the life-cycle of the pathogen or vector varies between organisms producing a lag between weather exposure and clinical manifestations of pathogen and subsequent laboratory diagnosis [26] , but this has not been addressed in the current study. Lag effects may be more pronounced for organisms that are indirectly rather than directly associated with weather [27] , for example weather conditions that precede mosquito larvae growth do not immediately result in malaria transmission, due to development of both mosquito and pathogen being highly complex [28] . However, given that the analysis was undertaken at a monthly resolution some short-term lagged