深圳流感发病的气象诱因及预测建模研究

The association between influenza onset and meteorological inducers in Shenzhen and construction of predictive models

  • 摘要: 深圳市在人口结构和经济社会发展方面很特殊,而对于其热带向副热带过渡的气候特征对流感发病的影响仍缺乏深入研究。本研究收集长序列(2003—2019年)的深圳市流感样病例(Influenza Like Illness,ILI)监测数据,采用分布滞后非线性模型(DLNM)系统分析了ILI与多种气象因子的关联,并分别使用Prophet时间序列和多元逐步回归模型对流感风险进行预报。近17年来深圳ILI发病在2003—2009年增加、2010—2014年平稳、2015—2019年下降,年周期特征凸显;多数年份发病率呈夏季单峰型,与高温、高湿的气候背景高度相符;个别年份在年末出现次高峰,常与大规模暴发疫情有关。DLNM揭示,高温对ILI风险的即时性影响较强,气温达到29.9℃,相对危险度值(RR)可达1.237(95%置信区间(95%CI):1.203—1.272);而低温效应在滞后2—3周起主导作用。70%—75%的湿度范围对应ILI高风险段, 70%相对湿度的RR为1.089(95%CI:1.046—1.135)。偏高的湿度与高温共存可诱使ILI最高风险点出现,即二者有协同增强效应,在其长夏短冬气候下尤其需要注意。ILI危险度在气温日较差为4—6℃或>9℃时均有显著增加,即日内温差对流感的活跃程度亦有显著影响;由于深圳的风速整体较小,其影响整体较弱。Prophet时间序列模型和逐步回归模型的回报准确率相近(>86%),而同时考虑了气象因子和前期发病人数的回归模型预测准确率更高(>80%)。简言之,深圳市ILI风险与温、湿度的非线性协同影响关系最为密切,其发病率很大程度上是可预测的。

     

    Abstract: Shenzhen is very unique in population structure as well as social and economic developments, but the relationship between its transitional monsoon climate from the tropics to subtropics and the influenza morbidity has rarely been evaluated. In order to reveal the epidemiologic features of influenza in Shenzhen, long-term surveillance data of influenza like illness (ILI) for the recent 17 years (2003—2019) are collected. The distributed-lag nonlinear model (DLNM) is adopted to investigate the associations between ILI morbidity and climatic variables. Further, the Prophet time-series approach and a stepwise-regression model are used to predict influenza risks, respectively. The results reveal that the number of ILI outpatients increases at the early stage of the study period (2003—2009), then stabilizes between 2010 and 2014, and decreases slightly from 2015 to 2019. Besides, a relatively clear annual cycle of ILI morbidity is discovered, which shows a unimodal pattern in most of the years. The number of ILI cases usually peaks during the summer months, which is highly consistent with the hot and humid climatic background. A second peak that is usually related to large-scaled influenza epidemics occurs occasionally at the end of the year. The DLNM shows that high air temperature has substantial and immediate impact on the ILI risk (T=29.9℃, RR=1.237, 95%CI: 1.203—1.272), while the effect of low temperature is relatively weak and occurs with a significant lag (2—3 weeks). Relative humidity of 70%—75% can induce the highest risk of ILI (e.g., RH=70%, RR=1.089, 95%CI: 1.046—1.135), and it also has synergistic cross-effect with high temperature. Besides, the number of ILI outpatients increases prominently when daily temperature range (DTR) is within in 4—6℃ or >9℃, indicating a nonlinear relation between DTR and the spread of influenza virus. Due to the overall low wind speed in Shenzhen, its effect on ILI is insignificant. The correct rates of prediction by the Prophet time-series approach and multiple-linear regression model are similar to each other (>86%). Nevertheless, the prediction accuracy of the regression model is relatively high (>80%) when considering meteorological factors and previous ILI cases simultaneously. In summary, the ILI morbidity in Shenzhen is significantly and nonlinearly associated with the synergistic effect of air temperature and humidity. Thus, the influenza risk is to some extent predictable.

     

/

返回文章
返回