Analyze the epidemic, lagging effect and prediction in time series models, applied to monthly weather and pollution related HFRS

Background: Previous studies have typically explored daily lagged relationships among hemorrhagic fever with renal syndrome (HFRS) and meteorology, with little seasonal exploration of monthly lagged relationships, interactions and multiple predictions among hemorrhagic fever and pollutants. Methods Our researchers collected HFRS cases data from 2005–2018 as well as meteorological and contaminative factors from 2015–2018 for the Northeast region. Firstly, we reported the application of the moving epidemic method (MEM) to estimate epidemic threshold and intensity level. Then we developed a distributed lag non-linear model (DLNM) and generalized additive model (GAM) with a maximum lag of 6 months to evaluate the lagged and interaction effects of meteorological and pollution factors on HFRS cases. Multiple machine learning was then performed after applying Spearman analysis to screen environmental factors in the Northeast. Results There has been a yearly downward trend in the incidence of HFRS in the northeastern regions. High prevalence threshold years were in 2005–2007 and 2012–2014, the epidemic months were mainly concentrated in November. During the low prevalence threshold period, the main lag factor was low wind direction. And the meteorological lag effect was high during the high prevalence threshold period, where the main lag factors were cold air and hot dew point. Low levels of AQI, PM 10 and high levels of PM 2.5 showed a dangerous lag effect on the onset of HFRS, but a protective effect at extreme high levels of PM 2.5 . And high levels of AQI, PM 10 and low levels of PM 2.5 showed a protective lag effect. The model of PM 2.5 and AQI interaction pollution is better. The SVM-Radial algorithm outperformed other algorithms, where the predictive variables of pollutants performed well. Conclusions This is the first mathematically based study of seasonal threshold of HFRS in Northeast China, which allows accurate estimation of epidemic level. Our findings support that long-term exposure to air pollution is the risk factor for HFRS. We should focus on pollutants monitoring in cold condition and HFRS prediction modeling.