Asymmetric Effects of Weather-Integrated Human Brucellosis Forecasting System Using a New Nonlinear Autoregressive Distributed Lag Model

Abstract

Human brucellosis (HB) remains a significant public health concern in China. This study aimed to investigate the long- and short-term asymmetric impacts of meteorological variables on HB and develop an early prediction system. Monthly data on HB incidence and meteorological variables were collected from 2005 to 2020. The study employed the autoregressive distributed lag (ARDL) and nonlinear ARDL (NARDL) to analyze the long- and short-term effects of climate variables on HB. Subsequently, the data were split into training (from January 2005 to December 2019) and testing parts (from January to December 2020) to develop and validate the forecasting accuracy of both models. During 2005–2020, there were 34,993 HB cases (2.03 per 100,000 persons) and there was an overall rising trend (average annual percentage change = 21.18%, 95%CI 18.36%–26.01%) in HB incidence, peaked in May and troughed in December per year. A 1 m/s increment and decrement in differenced (Δ) average wind velocity (AWV) contributed to 73.8% and 87.5% increases in ΔHB incidence, respectively (Wald long-run asymmetry test (WLR) = 1.17, P=0.25). A 1 hr increment and decrement in Δ(average relative humidity) contributed to both 3.1% increases in ΔHB incidence (Wald short-run asymmetry test = 3.01, P=0.003). Average temperature (AT) (P<0.001) and average air pressure (P=0.012) played a long-run linear impact on HB. Δ(aggregate precipitation) (WLR = 1.76, P=0.08) and Δ(aggregate sunshine hours) (WLR = 0.07, P=0.94) did not have a significant long-term asymmetric impact on Δlog(HB). ΔΔAT(+) and ΔΔAWV(−) at a 1-month lag had a meaningful short-run effect on Δlog(HB). In the forecasting aspect, the NARDL produced significantly smaller error rates compared to the ARDL. Weather variability played significant long- and short-run asymmetric roles in HB incidence. The NARDL by integrating climatic variables could accurately capture the dynamic structure of HB epidemic, meaning that meteorological variables should be integrated into the public health intervention plan for HB.