Asthma attack prediction based on weather factors

Abstract

Asthma is a chronic disease which concern public health over all the world. Predicting overabundant need for healthcare services is helpful for healthcare suppliers, as it allows them to adequately plan and provide the resources that deliver services easier and suitably. Several researches have linked asthma triggers with weather and environmental changes. In this study, 'linear regression model (LRM)' and 'quantile regression model (QRM)' are used to predict overabundant need for healthcare services for asthma monthly admissions in Polk, using backdated data from 2010 to 2017. Five weather variables were examined: maximum temperature, minimum temperature, precipitation level, humidity and thunderstorms. LRM and QRM models of asthma monthly admissions are fitted in two ways: the first way is the base model which using all the weather variables, and the second way is the reduced model which using a subset of these variables according to the pseudo R2. Models were cross-validated using the mean absolute percentage error and the level of accuracy. The base QRM predictive model with the 45th percentile of the distribution was the best fit and it detect utmost number of monthly asthma admissions at accuracy of 87.55 %, followed by base LRM with an accuracy of 87.13 %. The reduced LRM and QRM models give lower results with an accuracy of 86% and 85.9% respectively. The finding suggest that the weather variables are giving better results for the predictive model even if they have a low correlation relationship. The combination between weather variables and asthma, taking in temperature, precipitation level, humidity and thunderstorms can be utilized in predicting future asthma admission patients.