Seasonality and hydrometeorological predictors of rotavirus infection in an eight-site birth cohort study: Implications for modeling and predicting pathogen-specific enteric disease burden
MetadataShow full item record
Improving understanding of the pathogen-specific seasonality of enteric infections is critical to informing policy on the timing of preventive measures and to forecasting trends in the burden of diarrhoeal disease. Longitudinal and time-series analyses are needed to characterize the associations between hydrometeorological parameters and pathogen-specific enteric infectious disease (EID) outcomes. Data obtained from active surveillance of cohorts can capture the underlying infection status as transmission occurs in the community. However, there is a need for an approach that can be systematically applied to evaluate the combined impact of multiple meteorological exposures at a level of spatiotemporal disaggregation sufficient to characterize potential lag effects, interactions and non-linearity. Earth Observation (EO) climate data products derived from satellites and global model-based reanalysis have the potential to be used as surrogates in situations and locations where weather-station based observations are inadequate or incomplete. However, these products often lack direct evaluation at specific sites of epidemiological interest. The first aim of the research presented here was to characterize rotavirus seasonality in eight different locations (the MAL-ED study sites) to demonstrate the feasibility of applying an adapted Serfling model approach to data on EID from a multi-site cohort study. The second was to select climate data products and assess their performance both in characterizing meteorological conditions at those specific eight locations and as predictors of a known climate-sensitive outcome – namely rotavirus infection episodes. The third aim was to characterize the associations between a suite of nine EO-derived hydrometeorological variables and rotavirus infection status ascertained through community-based surveillance in a way that can be used to predict future trends in disease burden. In all seven of the eight study sites where seasonality in rotavirus infection was identified, the primary annual peak occurred outside of the rainy season. In all except two of these, a smaller, secondary annual peak was identified occurring during the rainier part of the year. The patterns predicted by this approach are broadly congruent with several emerging hypotheses about rotavirus transmission and are consistent for both symptomatic and asymptomatic rotavirus episodes. These findings have practical implications for programme design, but caution should be exercised in deriving inferences about the underlying pathways driving these trends, particularly when extending the approach to other pathogens. The availability and completeness of weather station-based meteorological data varied depending on the variable and study site. The performance of the two gridded EO climate models varied considerably within the same location and for the same variable across locations, according to different evaluation criteria and for the peak-season compared to the full dataset in ways that showed no obvious pattern. They also differed in the statistical significance of their association with the rotavirus outcome. For some variables, the station-based records showed a strong association while the EO-derived estimates showed none, while for others, the opposite was true. Numerous hydrometeorological parameters – including several that are not commonly measured by weather stations – were found to exhibit complex, non-linear associations with rotavirus infection that differ by infection episode type and may be independently, and highly statistically significant over multiple consecutive or non-consecutive lags, including as short a period as two days. The results show evidence for the hypothesis that the effect of climate on rotavirus transmission is mediated by four independently-operating mechanisms: waterborne dispersion via rainfall and surface runoff, airborne dispersion in humidity-sensitive aerosols, virus survival on soil and fomites, and host factors. Researchers wishing to utilize publicly available climate data – whether EO-derived or station based - are advised to recognize their specific limitations both in the analysis and the interpretation of the results. Epidemiologists engaged in prospective research into environmentally driven diseases should install their own weather monitoring stations at their study sites whenever possible, in order to circumvent the constraints of choosing between distant or incomplete station data or unverified EO estimates. Since the EO datasets from which the predictors used in this analysis were extracted are available at global scale and sub-daily resolution and updated continually, as new studies using similar methods are carried out at different locations, they can be added to the MAL-ED data to derive more precise predictions for more diverse conditions. Furthermore, emerging tools for objective climate regionalization can be combined with the results of these models to divide extensive geographic zones into smaller regions that are homogenous with respect to important climate characteristics and to which predictions from these models can be applied.