高级检索

基于广义相加模型的2023－2025年广东省深圳市发热门诊数影响因素分析与短期预测

Influencing factors and short-term prediction of fever clinic visits in Shenzhen, Guangdong, 2023−2025, a study based on generalized additive model

    摘要
  • 摘要:
    目的 从疾病和行为两个维度，分析社区常见传染病流行和发热百度指数对广东省深圳市发热门诊数的影响，从而对发热门诊数开展短期趋势预测。
    方法 从中国疾病预防控制信息系统和百度搜索引擎收集2023年4月至2025年6月深圳市二级以上医院和城市社区卫生服务中心的发热门诊数据、社区常见传染病发病数据及发热相关百度指数，筛选5种核心发热相关传染病，构建发热百度指数（FBSI），采用广义相加模型（GAM）分析发热门诊数的影响因素，再以2023年4月3日至2025年5月25日共112周的数据作为训练集，2025年5月26日至6月22日共4周的数据为预测集开展短期预测。
    结果 最终模型为GAM1.0，解释变量包括新型冠状病毒感染、流行性感冒和流行性腮腺炎，调整后的决定系数（R2-sq）=0.88，方差解释率91.80%，其中流行性感冒对发热门诊数贡献最大（F=30.062，P<0.001），其次为新型冠状病毒感染（F=10.549，P<0.001）；FBSI与发热门诊数相关系数为0.85（P<0.001），但在GAM1.0中未达到统计学显著水平（P=0.186），进一步分析显示，当FBSI作为唯一解释变量（GAM2.0）时，变量有统计学意义（F=32.810，P<0.001）；当与流行性感冒同时纳入模型（GAM3.0）时，FBSI的贡献（F=15.195，P<0.001）大于流行性感冒（F=4.943，P=0.028）；使用GAM1.0进行短期趋势预测，预测值与实际值的相关系数平方（R2）=0.98。
    结论 多类社区常见传染病与FBSI是深圳市2023－2025年发热门诊数的关键影响因素，GAM1.0短期趋势预测效果良好，可为发热门诊资源优化与防控提供可靠依据。

     

    Abstract:
    Objective To investigate the impacts of common community-acquired infectious diseases and fever-related Baidu search index on fever clinic visits in Shenzhen, Guangdong province, from both disease and behavioral dimensions, and develop a short-term trend prediction model for fever clinic visits.
    Methods The data of fever clinic visits, incidence of common community-acquired infectious diseases, and fever-related Baidu search index were collected from the China Disease Prevention and Control Information System and Baidu search engine. The data were from fever clinics in secondary and higher-level hospitals and urban community health service centers in Shenzhen from April 2023 to June 2025. Five core fever-related infectious diseases were selected to construct the Fever Baidu Search Index (FBSI). A generalized additive model (GAM) was applied to identify the influencing factors associated with fever clinic visits. The data from April 3, 2023 to May 25, 2025 (112 weeks) were used as the training set, and the data from May 26 to June 22, 2025 (4 weeks) were used as the prediction set for short-term trend prediction.
    Results The final selected model was GAM1.0, with explanatory variables including coronavirus disease 2019 (COVID-19), influenza, and mumps. The adjusted coefficient of determination (adjusted R2-sq) was 0.88, and the explained deviance was 91.80%. Among the included variables, influenza showed the greatest contribution to fever clinic visits (F=30.062, P<0.001), followed by COVID-19 (F=10.549, P<0.001). The correlation coefficient between FBSI and fever clinic visits was 0.85 (P<0.001). However, the effect of FBSI was not statistically significant in GAM1.0 (P=0.186). Further analysis revealed that when FBSI was included as the sole predictor (GAM2.0), its effect was significant (F=32.810, P<0.001). When FBSI and influenza were simultaneously included in the model (GAM3.0), the contribution of FBSI (F=15.195, P<0.001) was greater than that of influenza (F=4.943, P=0.028). The short-term trend prediction based on GAM1.0 demonstrated a strong consistency between predicted value and observed value, with a squared correlation coefficient (R2) of 0.98.
    Conclusion Multiple common community-acquired infectious diseases and fever-related Baidu search index were the important determinants of fever clinic visits in Shenzhen from 2023 to 2025. The GAM1.0 model showed excellent performance in short-term trend prediction and may provide reliable evidence to improve fever clinic resource allocation and strengthen infectious disease prevention and control.

     

    • HTML全文
    • 参考文献(21)
    • 相关文章
    • 施引文献
  • 资源附件(0)

/

返回文章
返回

在线交流