Association Analysis And Prediction Model Study Of The Relationship Between Ambient Temperature And Road Traffic Injuries Based On 15 Cities In China

Objective Heat stress is closely related to public health,and the effect of ambient temperature on road traffic injuries has become an increasingly serious public health problem.However,few studies have compared the differences between road traffic injury risks in different cities and analyzed the reasons.This study intends to investigate the association between temperature and road traffic injury risk using a case-crossover design and to explore the impact of different cities characteristics on temperature-related road traffic injuries.This study predicts road traffic injuries based on various machine learning models,trains and optimizes model parameters to improve the accuracy and reliability of predictions.The relationship between heat stress indices,such as humidex and apparent temperature,and road traffic injury risk will be analyzed to test the robustness of findings.Under the circumstances of diverse climatic conditions and frequent extreme temperature events,it is of great significance to investigate the effect of ambient temperature on road traffic injury in cities with different characteristics for future prevention and control work.Methods Using road traffic injury patients recorded by the National Injury Surveillance System as the study population,this study employs a time-stratified case-crossover design to adjust for potential individual-level exposure risk and confounding bias.The cities of the study included Hefei,Maanshan,Nanjing,Ningbo,Panzhihua,Qinhuangdao,Quzhou,Shanghai,Shenzhen,Shuangyashan,Suzhou,Yinchuan,Chongqing,Zhuhai and Zhuzhou.The road traffic injury risk was first modeled separately using the mean temperature as the city temperature and a cross-base of green exposure was included in the models.Significance of the effect of including meteorological variables was tested based on multivariate Wald analysis.The meta-analysis was performed after calculating the overall cumulative risk of ambient temperature and road traffic injury for all study cities separately.Heterogeneity among road traffic injury results for all cities was reported by Cochran Q test and I~2statistic.Subgroup analysis and meta-regression were used to process and analyze the heterogeneity.Prediction of road traffic injuries based on ambient temperature,meteorological factors and environmental factors.Bayesian generalized linear model,artificial neural network model,random forest model,support vector machine model and extreme gradient boosting model were used in the study.A 10-fold cross-validation of the study database was performed,and the road traffic injury population in the study city was randomly divided into training and validation groups according to 7:3.Model evaluation metrics included AUC,accuracy,sensitivity,specificity,positive predictive value and negative predictive value.The composite heat stress index,which incorporates multiple meteorological factors,is widely regarded as a more accurate indicator of the true heat stress environment associated with road traffic injuries.Humidex were calculated based on mean temperature and relative humidity,and apparent temperature was calculated from mean temperature,relative humidity and wind speed.Verify the relationship between different heat stress indices and road traffic injury risk models through different heat stress indices.Results The mean temperatures of the 15 study cities selected for the National Injury Surveillance System ranged from 3.38°C to 23.60°C,which can largely reflect the temperature conditions in most regions of China.A total of 514,062 road traffic injuries were included in the study during the study period of 2015-2019.Most cities showed a significant effect of high and low temperatures on road traffic injury risk.This study reveals that the effects of ambient temperature on road traffic injuries differ significantly across cities.Immediate effects of ambient temperature on road traffic injuries were observed in Chongqing,Maanshan,Shuangyashan,and Zhuhai,while other cities such as Shanghai,Hefei,Ningbo,Suzhou,and Zhuzhou showed a significant lag effect.The association pattern between ambient temperature and road traffic injury risk can be categorized into three types:in the first association pattern,the lowest risk of temperature occurs at the lowest temperature in the city.The effect on road traffic injuries gradually increases with temperature and begins to trend downward after reaching the highest effect value of risk.The second pattern also has the lowest risk at the lowest temperature,but the road traffic injury risk will not have a decreasing trend in the temperature range of the city.The highest risk effect value occurs at the highest temperature.The lowest risk temperature of temperature in the third association pattern is not the lowest temperature of the city.The injury risk first decreases as the temperature rises until it reaches the lowest risk,and then the risk begins to increase gradually.Meta-analysis calculates the pooled results of the cumulative effect of road traffic injuries in different cities.For high temperature,the pooled road traffic injury results from lag0 days(RR=1.120,95%CI:1.010,1.242)to lag5 days(RR=1.406,95%CI:1.204,1.642)consistently showed a significant risk.Road traffic injury results showed a significant lag effect in the cold temperature condition.The road traffic injury results were not significant(RR=1.054,95%CI:0.978,1.136)at lag0 days,while the pooled effect showed statistical significance at lag5 days(RR=1.094,95%CI:1.015,1.180).The study continued with subgroup analysis and multivariate meta-regression analysis according to various city-specific factors.The results showed significant correlations between green exposure,economic condition,motor vehicle ownership,road network density and heterogeneity of temperature-related road traffic injury risk outcomes.By training and validating a large amount of historical traffic injury data,machine learning models accurately predict road traffic injuries and obtain satisfactory model parameters.Different machine learning models differ in road traffic injury prediction performance and are influenced by city and lag days.Based on the results of model evaluation metrics in different cities,the study proposes the prediction advantages of Bayesian generalized linear model and extreme gradient boosting model in road traffic injury.The artificial neural network model and the random forest model also have more stable prediction ability.And the prediction ability of the support vector machine model is relatively weak.Most cities showed significant effects of both high and low heat stress indices on road traffic injuries.The effects of humidex and apparent temperature on road traffic injury risk showed closer risk trends.Meta-analysis was used to calculate the pooled results of the cumulative effects of road traffic injuries in different cities.For high humidex,the pooled road traffic injury results were significant at lag5 days(RR=1.405,95%CI:1.233,1.601).For low humidex,the pooled results were 1.089(95%CI:1.035,1.147)at lag5 days.For high apparent temperature,the pooled road traffic injury results were significant at lag5 days(RR=1.292,95%CI:1.210,1.380).For low apparent temperature,the pooled results were 1.053(95%CI:1.019,1.089)at lag5 days.Conclusions This study determined the association between ambient temperature and road traffic injuries through a case-crossover study of multi-city in China.It was found that temperature had a significant effect on road traffic injuries and that the relationship differed significantly across different cities in China.Green exposure,economic development level,motor vehicle ownership,and road network density have significant correlations with the heterogeneity of road traffic injury risk.The machine learning models can provide accurate and reliable predictive information for road traffic injury prediction and have been validated in different cities.The differences between predictive models contribute to personalized traffic injury predictions.Adjusting prediction models according to the characteristics of specific cities and environments can enhance the accuracy and practicality of predictions,providing practical value for traffic management and safety policy formulation.Both the humidex and apparent temperature exhibit nonlinear relationships with the risk of road traffic injuries.Using the humidex or apparent temperature as input features instead of temperature in predictive models also demonstrates reliable predictive performance.These findings provide important evidence for improving road traffic injury management and further suggest the importance of meteorological warnings of road traffic injury risk.These findings can help reduce the public burden of road traffic injuries,especially in the context of climate change.