A time series analysis of the climate-related determinants of malaria in Buikwe District, Central Uganda : a data science approach

Abstract

Introduction: Climate change affects public health, with malaria being one of the most climate-sensitive diseases. Despite numerous control interventions, Buikwe District continues to experience high malaria morbidity, and the influence of local climatic factors on transmission remains insufficiently understood. This study aimed to analyze the temporal relationships between key environmental variables and malaria trends in Buikwe District from 2015 to 2024 and to forecast future malaria patterns using a time-series modelling approach. Methods: A retrospective longitudinal study design was used, drawing on DHIS2 malaria data and ERA5-Land climate data for temperature, precipitation, and humidity, from which heat index was derived. All monthly data were included using a census approach. Analysis in R 4.4.3 involved descriptive statistics, LOESS-based seasonal decomposition, Spearman’s correlation, Augmented Dickey-Fuller tests, Granger causality, and ARIMAX modelling for forecasting. Results: Analysis revealed strong seasonal patterns in the heat index with consistent annual cycles and a rising long-term trend between 2022 and 2024, suggesting increasing heat exposure likely linked to climate change. Temperature showed a statistically significant moderate negative correlation with malaria incidence (ρ = -0.25, p = 0.006), while precipitation and relative humidity had non-significant associations. Granger causality analysis confirmed that all three environmental variables significantly predicted malaria cases. Temperature exhibited the strongest effect, with both lagged (F = 9.81, p < 0.001) and immediate (χ² = 27.92, p < 0.001) impacts. Precipitation (F = 3.94, p = 0.009) and relative humidity (F = 3.15, p = 0.025) also demonstrated predictive power. The ARIMAX(1,0,1) model identified temperature as a significant negative predictor (-142,677.10) and humidity as a significant positive predictor (53,270.62) of malaria incidence. The model achieved good performance, with a low Mean Absolute Scaled Error (MASE = 0.613) and no significant residual autocorrelation (Ljung-Box p = 0.318). Forecasts for 2025 projected fluctuating malaria incidence following seasonal climatic patterns, highlighting periods of potential increased transmission risk driven by temperature and humidity variations. Conclusion: This study demonstrates that climatic factors—especially temperature—significantly influence malaria incidence in Buikwe District. The ARIMAX model successfully captured temporal patterns and forecasted malaria trends, highlighting the value of integrating climate data into surveillance systems. Strengthening climate-informed early warning systems, entomological monitoring, and predictive modelling, alongside collaboration between health, meteorological, and research institutions, is essential for implementing targeted interventions and building effective climate-responsive malaria control strategies.