Forecasting respiratory tract infection episodes from prescription data for healthcare service planning

Changing weather patterns affect the incidence of respiratory tract infections, which causes huge economic burden for healthcare services. Early warning for the infection may help healthcare service providers to prepare for an epidemic on time. The purpose of the current research is to explore the relationship between respiratory tract infection episodes and climatic factors and to predict the number of daily episodes in different weather zones of active weather stations in Bangladesh. Prescription data collected from clinics are integrated with climatic factors of the nearest weather stations, and the integrated dataset is used to predict the daily respiratory tract infection episodes. We apply panel generalized linear models and show that the number of episodes increases to a greater extent for increasing magnitude of rolling standard deviation of relative humidity and rolling mean of wind speed. A 7-day-ahead forecast of number of episodes based on rolling window models of regression tree, random forest, support vector regression, and deep neural network is estimated to know the severity of epidemic for healthcare planning. A further 1-day-ahead confirmation forecast is produced to assess the necessity of healthcare service plan adopted based on a 7-day-ahead forecast. Root mean squared forecast errors computed for both 7-day-ahead and 1-day-ahead forecasts from these models provide qualitatively similar results, except for three weather stations where an unusually high number of episodes are observed because of extreme climate and high level of air pollution.