时间序列分析在手足口病流行预测中的应用研究

Numerous large outbreaks of HFMD have occurred in China since 2008, which have caused death and neurological sequelae, and have become a growing public health threat. Currently, neither vaccine nor effective drug against enterovirus is available for human use. Thus, epidemiological surveillance of HFMD and its pathogens is important to take the proper and timely public health interventions to prevent its outbreaks. Early warning of HFMD outbreaks could improve the efficiency of vector control campaigns and help to take prevention actions to delay the epidemic, thus reducing its impact on health system. HFMD morbidity and mortality would be minimized through earlier and efficient public health response. Many studies have indicated that meteorological conditions are the most important factors of HFMD outbreaks. Many mathematical models have been developed to predict the occurrence of outbreaks using a combined environmental approach. Seasonal Autoregressive Integrated Moving Average (SARIMA) models allow the integration of external factors, such as climatic variables, to increase their predictive power. In this study, we proposed to develop Seasonal Autoregressive Integrated Moving Average (SARIMA) models using time series analysis of the number of laboratory-confirmed HFMD hospitalizations. The goals of this study were to characterize whether climatic factors are associated with HFMD epidemics among children and whether inclusion of such factors is useful to predict epidemics with higher precision. This predicable model would be used to facilitate efficient HFMD control.

2008年以来，我国进入手足口病流行期，手足口病患者及重症、死亡病例逐年升高，防控形势十分严峻。对于手足口病的治疗，目前仍没有确切有效的抗病毒药物，且没有安全有效的疫苗应用于人群。建立高度敏感的监测系统，对疾病的爆发流行进行预测，是手足口病防治的关键。手足口病的流行有明显的季节差异，温度和湿度对手足口病发病有重要影响。时间序列法特别适用于时序规律不明显，或有明显季节性和周期性的情况，季节性自回归移动平均模型可引入自变量提高模型的预测精确度，其过程简便、经济、预测精度较高。本项目拟采用时间序列方法，分析气候因素对手足口病流行的影响，引入自变量（气候变量）建立手足口病季节性自回归移动平均模型，研究结果将为手足口病的预测和对制定预防和控制手足口病的长远或近期应对策略有重要意义。

2008年以来，我国进入手足口病（hand,foot and mouth disease, HFMD）流行期，HFMD重症、死亡病例逐年增多，防控形势十分严峻，对其流行进行预测是防控的关键。本项目利用河南省郑州市2008-2018年HFMD监测与住院资料，及同期该地区气候资料，构建HFMD季节性自回归移动平均（Seasonal Autoregressive Integrated Moving Average, SARIMA）模型，分析HFMD与气候参数的相互关系，比较引入参数前后模型的拟合度（R2）和预测精度（root mean square error, RMSE）。应用郑州市二七区2008-2014年HFMD监测资料构建模型SARIMA(1,1,0)(0,1,0)52 R2为0.807，RMSE为11.573；应用2008-2017年郑州市HFMD数据构建模型SARIMA(2,2,1) (2,2,1)12 R2为0.624，RMSE为14.251；应用郑州市2009-2016年HFMD发病数据构建模型SARIMA(1,0,1)(1,0,1)12 R2 为 0.734，RMSE为 10.497；经互相关分析，HFMD周发病人数与每周日平均气温滞后2周（rs=0.248，P<0.05）相关，引入气候参数前、后模型SARIMA(1,1,0)(0,1,0)52的R2分别为0.797、0.833，RMSE分别为11.573、10.611。应用机器学习筛选重症HFMD预测变量：年龄<2岁、发热≥3天、体温≥38.5℃、白细胞≥10.8×109/L、血糖≥8.3mmol/L、EV71阳性和呕吐等指标。应用危险因素评分、Logistics回归方程及列线图构建重症HFMD临床预测标准，其ROC曲线下面积分别为0.745（95%CI：0.699-0.803）、0.751（95%CI：0.693-0.797）和0.850（95%CI: 0.835-0.864），列线图预测正确率为85.2%。研究证实气候因素是HFMD季节性流行的主要驱动因素，引入气候参数构建的SARIMA模型，能够更好的拟合及预测HFMD的流行，对制定防控策略有重要意义。机器学习能充分利用临床数据，构建重症HFMD临床预测标准，帮助临床医生早期识别重症患者，为阻止其发生、降低病死率提供科学依据。