基于季节性ARIMA模型的新疆肺结核发病预测分析

目的 探讨季节性时间序列模型(autoregressive integrated moving average,ARIMA)在新疆肺结核发病预测中的应用,并验证模型的可行性和适用性。 方法 采用季节性ARIMA(p, d, q )(P, D, Q)_s拟合2005年1月—2019年8月新疆地区肺结核月发病人数,建立多个季节时间序列模型并进行比较,选出最优模型对2019年9—12月肺结核发病人数进行预测。 结果 2005年1月—2019年8月新疆地区肺结核累积发病人数为627 869例,年平均发病人数为3 567例。 新疆地区肺结核月发病数具有季节性,1—5月平均发病数高于平均水平,6—12月平均发病数低于平均水平,发病高峰为1月和3月,发病低谷为9月。通过赤池信息量(Akaike Information Criterion,AIC)和贝叶斯信息量(Bayesian Information Criterion,BIC)最小原则得出,ARIMA(1, 1, 1 )(0, 1, 2)₁₂是最优模型,其残差序列为白噪声,参数的回归系数均具有统计学意义,拟合的平均绝对百分比误差MAPE为8.723%。预测的MAPE为18.674%,真实值均处于预测值的95%置信区间内。 结论 ARIMA(1, 1, 1 )(0, 1, 2)₁₂模型能够较好地拟合新疆肺结核发病数据,并进行短期预测,对新疆卫生防控措施的制定具有一定指导意义。

Prediction of pulmonary tuberculosis incidence in Xinjiang based on seasonal ARIMA model

Objective To explore the application of seasonal autoregressive integrated moving average (ARIMA) model to prediction of pulmonary tuberculosis incidence in Xinjiang, and to verify the feasibility and applicability of the model. Methods Seasonal ARIMA (p, d, q )(P, D, Q)_s was used to fit the monthly incidence of pulmonary tuberculosis in Xinjiang from January 2005 to August 2019. Multiple seasonal time series models were established and compared to select the optimal model to predict the incidence of pulmonary tuberculosis from September to December 2019. Results From January 2005 to August 2019, the cumulative incidence of pulmonary tuberculosis in Xinjiang was 627,869 cases, with an average annual incidence of 3,567 cases. The monthly incidence of pulmonary tuberculosis in Xinjiang showed a seasonal pattern. The average incidence from January to May was higher than the average level, while the average incidence from June to December was lower than the average level. The incidence peak was in January and March, whereas the incidence was found to be lower in September. According to the minimum principle of Akaike Information Criterion and Bayesian Information Criterion, ARIMA (1, 1, 1 ) (0, 1, 2)₁₂ was the optimal model, the residual sequence was white noise. The regression coefficients of parameters were statistically significant, and the average absolute percentage error of fitting was 8.723%. The predicted mean absolute percentage error was 18.674%, and the real values were within the 95% confidence interval of the predicted values. Conclusion ARIMA (1, 1, 1) (0, 1, 2)₁₂ model can better fit the incidence data of pulmonary tuberculosis in Xinjiang and make short-term prediction, which has a certain guiding significance for the formulation of health prevention and control measures in Xinjiang.