A meteorological-based forecasting model for predicting minimal infection rates in complex using Québec's West Nile virus integrated surveillance system.
Background: The (Québec's health authority) has expressed an interest in the development of an early warning tool to identify seasonal human outbreaks of West Nile virus infection in order to modulate public health interventions. The objective of this study was to determine if a user-friendly meteorological-based forecasting tool could be used to predict minimal infection rates for the complex-a proxy of human risk-ahead of mosquito season. Methods: Annual minimal infection rate (number of positive pools/number of mosquitoes) was calculated for 856 mosquito traps set from 2003 to 2006 and 2013 to 2018 throughout the south of Québec's. Coefficient of determination (R) were estimated using the validation dataset (one third of the database by random selection) with generalized estimation equations, which were prior fitted backwards with polynomial terms using the training dataset (two thirds of the database), in order to minimize the Bayesian information criteria. Mean temperatures and precipitation were grouped at five temporal scales (by month, by season and by 4, 6 and 10-months groupings). Results: Mean temperatures and cumulative precipitation from the previous months of March (R=0.37), May (R=0.36), December (R=0.35) and the autumn season (R=0.38) accounted for ~40% of annual minimal infection rates variations. Including the "year of sampling" variable in all regression models increased the predictive abilities (R between 0.42 and 0.57). Conclusion: All regression models explored have too weak predictive abilities to be useful as a public health tool. Other factors implicated in the epidemiology of the West Nile virus need to be incorporated in a meteorological-based early warning model for it to be useful to the provincial health authorities.
