Malaria prediction using weather-based time-series distributed lag nonlinear model

Context/Purpose

Malaria is a climate-sensitive vector-borne infectious disease. There have been massive efforts to develop an early warning system for malaria, but either successful nor sustainable system remained. Well-organized malaria surveillance and high quality of climate forecasts are required to persist an early warning system based upon an effectively performed malaria prediction model. We developed a weather-based malaria prediction model and simulated it using time-series data from 1998 to 2015 in Vhembe district, Limpopo province, South Africa.

Methods

We used a distributed lag non-linear model (DLNM) to simulate the weekly basis malaria prediction consisting of the three main parts: 1) model building to investigate the associations between malaria incidence and weather factors (temperature and precipitation), 2) the 1^st simulation of the malaria prediction with the observed weather (assuming perfect weather forecast), and 3) the 2^nd simulation of the prediction with hybrid datasets of the weather observations and predicted weather forecasts.

Results

A total of 53,689 malaria patients reported by the malaria surveillance system was included. The malaria transmission showed a clear seasonal pattern with the higher number of patients in warmer and rainy seasons. The malaria prediction model by DLNM showed high performance for short-term predictions (correlation coefficient, r > 0.8 for 1- and 2-week ahead). The accuracy has decreased as the lead-time increased retaining fairly good performance with r > 0.7 up to 16-week ahead prediction. By the 2^nd simulation, the 2-weeks-ahead prediction values coincided closely with the observed values, although the predictions tended to be higher during some years.

Discussion

We developed the highly performed malaria prediction model, considering the predicted weather forecasts that would be feasibly applicable together with the malaria surveillance data. Establishing an automated operating system based on real-time data inputs would be potentially beneficial for the malaria early warning system, which would be an instructive example for other malaria-endemic areas.