Combined Lagged Effects Of Pollution And Meteorological Factors On The Incidence Of Epidemic Hemorrhagic Fever In Shenyang Analysis Of Joint Lagged Effects And Construction Of Their Prediction Models

Objective:To grasp the three-interval distribution characteristics of epidemic hemorrhagic fever in Shenyang and explore the predictive effect of time on hemorrhagic fever in the time series analysis model.By constructing distribution lag nonlinear model and generalized summation model,we analyzed the lag and interaction effect relationship of meteorology and pollutants on the onset of epidemic hemorrhagic fever at different stratification levels.The predictive effects of environmental factors on epidemic hemorrhagic fever were compared by multivariate machine learning regression methods.Methods:1.By descriptive epidemiological method,Excel 2019 and IBM SPSS 19.0software were used in order to analyze the trend and triple distribution of epidemic hemorrhagic fever in Shenyang from 2005-2019.2.By time series analysis method,R4.1.3 software was used to explore the prediction effect and fluctuation rate of epidemic hemorrhagic fever in Shenyang from 2005-2019.3.The lagged and interactive effects of exposure to atmospheric pollutants and meteorological factors on epidemic hemorrhagic fever were analyzed by distributional lagged nonlinear model（DLNM）and generalized summation model（GAM）from 2014-2019 in Shenyang city,and their distribution characteristics in different gender,age and diagnostic delay group dimensions were elucidated to further clarify the susceptible population.4.By gradient boosted regression tree（GBRT）,random forest（RF）,support vector machine（SVM）and Gaussian process regression（GPR）were used to evaluate the predictive effect of each model in different subgroups.Results:1.During 2005-2019,epidemic hemorrhagic fever in Shenyang showed an overall decreasing trend,showing the bimodal distribution characteristics of spring and summer pandemic（February-May）and winter mini-pandemic（October-December）.Epidemic hemorrhagic fever has a high prevalence among male young adults in the population distribution,with farmers as the main occupational workers,and more regional epidemics,but mostly concentrated in central locations.2.Traditional Holt-Winters and SARIMA models are better for monthly forecasting than quarterly,SARIMA model is better with model（1,1,0）（2,1,0）[12],GARCH model is better for epidemic trend and fluctuation magnitude.3.Temperature,barometric pressure,air humidity and wind speed were associated with epidemic onset,and seasonal distribution trends existed for both meteorology and disease according to the time series plot.In the meteorological dose-response relationship,air temperature and wind speed showed an"inverted U"shape with optimal lag effect,while air humidity and air pressure had a"parabolic"shape with a decreasing trend.In the preliminary study of extreme weather lags,it showed that low temperature/high pressure had a dangerous effect on epidemic hemorrhagic fever,while high temperature/high humidity/low pressure showed a protective effect.In the cumulative lag study,all effects were good except for wind speed,and barometric pressure showed lagging effects first after 2 days of lagging,followed by temperature,and finally humidity showed lagging after 11 days of accumulation.The meteorological interaction effect showed that the natural conditions with the simultaneous presence of very low temperature-low humidity had the greatest effect on the onset of epidemic hemorrhagic fever.4.PM_2.5,PM₁₀and SO₂were associated with epidemic onset.In the single-pollutant model,we found that the first maximum threshold effect of moderate level（200-250μg/m³）SO₂occurred at a lag of 5-10 days,followed by a second maximum effect after 15 days;while high level（600-700μg/m³）PM₁₀first showed a maximum lag effect at a lag of 0-2 days,and high level（above 850μg/m³）PM_2.5showed the maximum lag effect at both lag 0-5 days and 10-15 days.In the two-pollutant lag model,PM_2.5enhanced its protective effect against hemorrhagic fever onset after inclusion of PM₁₀in a population with a diagnostic delay of 10 days or more.In the multi-pollutant model,overall low levels of PM_2.5and high levels of PM₁₀had a protective effect in patients with a delay in diagnosis of more than 10 days,with RR values of 0.002（95%CI:0,0.897）and 0（95%CI:0,0.009）,respectively,while high levels of PM_2.5and low levels of PM₁₀also had a risk effect in female patients,with RR values of 3e+19（95%CI:1e+3,7e+35）and 1e+13（95%CI:5e+3,2e+22）,respectively.On the lag effect of SO₂on hemorrhagic fever,the lag effect was not significant when combined with the effect of particulate matter;however,it showed a dangerous effect at lag 0-2 days in the early period when acting alone,and a protective trend at high levels of lag days in the later period.In the pollutant-meteorological interaction model,the effect of high levels of PM_2.5and low levels of SO₂on epidemic hemorrhagic fever was comparable to that of high humidity and low levels of SO₂,which together triggered changes in the hemorrhagic fever epidemic.And the confounding variable in the interaction model was most significant with temperature,which played an important role in the lagged effect of hemorrhagic fever by pollutants.5.In terms of different models,radial basis and Sigmoid kernel function support vector machine regression predicted the best results,followed by random forest regression.In the complex Gaussian process regression,the Laplace operator kernel function with relatively simple parameters works better than the traditional polynomial kernel function with complex parameters and the optimized Bessel kernel function.From the different stratified analysis,men and patients with more than 10days delay in diagnosis had better prediction results,and patients in the age group above 50 years also confirmed better prediction results on support vector machine and random forest regression compared to other models.Conclusions:1.Epidemic hemorrhagic fever in Shenyang city showed an overall decreasing trend with a bimodal distribution characterized by a spring-summer pandemic（February-May）and a winter mini-pandemic（October-December）.2.For single-factor prediction,the traditional Holt-Winters and SARIMA models are better for monthly prediction,and the GARCH model is better for the epidemic trend and fluctuation magnitude.Support vector machines and random forests are better for multi-factor forecasting,and different groups of people have different forecasting effects.3.Among meteorological and pollution factors have non-linear lagged effects on epidemic hemorrhagic fever,and there are interactive effects.