Wang Yongbin, Liang Yifang, Zhang Bingjie, Yi Shibei, Zhou Peiping, Lan Xianxiang, Xue Chenlu, Li Yanyan, Li Xinxiao, Xu Chunjie
Department of Epidemiology and Health Statistics, School of Public Health, Xinxiang Medical University, Xinxiang, China.
Beijing Key Laboratory of Antimicrobial Agents/Laboratory of Pharmacology, Institute of Medicinal Biotechnology, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China.
J Glob Health. 2025 Jul 25;15:04215. doi: 10.7189/jogh.15.04215.
Tuberculosis (TB) remains a significant public health challenge in Henan, China, requiring accurate forecasting to guide prevention and control efforts. While traditional models like autoregressive integrated moving average (ARIMA) are commonly used, they may not fully capture long-term dependencies in the data. This study evaluates the autoregressive fractionally integrated moving average (ARFIMA) model, which incorporates fractional differencing, to improve TB forecasting by better modelling long-range dependencies and seasonal patterns.
Monthly TB incidence data from January 2007 to May 2023 in Henan were collected. The data set was split into a training set (January 2007-May 2022) and a test set (June 2022-May 2023). Both ARIMA and ARFIMA models were developed using the training set, and their predictive accuracy was assessed on the test set using metrics such as mean absolute deviation, mean absolute percentage error, mean square error, and mean error rate. A sensitivity analysis was conducted to evaluate the robustness of the forecasts.
There were 1 074 081 TB incident cases in Henan during the study period. The TB incidence was reducing at an annual rate of 5.83%, with the seasonal factor >1 between March-July and seasonal factor <1 in other months. The ARIMA (2,0,1)(0,1,1) and ARFIMA (2,0,1)(0,0.38,1) models were identified as suitable for the data. The ARFIMA model consistently outperformed ARIMA model in the forecasting phase, with lower errors across all metrics (e.g. mean absolute deviation: 467 vs. 569.54; mean absolute percentage error: 0.19 vs. 0.21; mean square error: 620.48 vs. 690.11; mean error rate: 0.14 vs. 0.17). This indicated that the ARFIMA model better captures long-term dependencies and seasonal patterns, leading to more accurate forecasts.
Tuberculosis incidence in Henan shows a clear downward trend with distinct seasonal variation. The ARFIMA model provides more accurate TB incidence forecasts than ARIMA, particularly in capturing long-term trends and seasonality. Effective management of TB at the population level requires proper monitoring and understanding of disease patterns. Forecasting serves as a critical tool for detecting deviations from expected trends, which may signal changes in disease dynamics. Continuous use of the ARFIMA model is essential for guiding public health interventions and ensuring timely responses to emerging challenges in TB control.
在中国河南，结核病仍然是一项重大的公共卫生挑战，需要进行准确预测以指导防控工作。虽然自回归积分滑动平均（ARIMA）等传统模型被广泛使用，但它们可能无法完全捕捉数据中的长期依赖性。本研究评估了自回归分数积分滑动平均（ARFIMA）模型，该模型纳入了分数差分，旨在通过更好地模拟长期依赖性和季节性模式来改进结核病预测。
收集了河南2007年1月至2023年5月的月度结核病发病率数据。数据集被分为训练集（2007年1月至2022年5月）和测试集（2022年6月至2023年5月）。使用训练集开发了ARIMA和ARFIMA模型，并使用平均绝对偏差、平均绝对百分比误差、均方误差和平均误差率等指标在测试集上评估它们的预测准确性。进行了敏感性分析以评估预测的稳健性。
研究期间河南共有1074081例结核病发病病例。结核病发病率以每年5.83%的速度下降，3月至7月的季节性因素>1，其他月份的季节性因素<1。ARIMA（2,0,1）（0,1,1）和ARFIMA（2,0,1）（0,0.38,1）模型被确定为适合该数据。在预测阶段，ARFIMA模型始终优于ARIMA模型，所有指标的误差均较低（例如，平均绝对偏差：467对569.54；平均绝对百分比误差：0.19对0.21；均方误差：620.48对690.11；平均误差率：0.14对0.17）。这表明ARFIMA模型能更好地捕捉长期依赖性和季节性模式，从而实现更准确的预测。
河南的结核病发病率呈明显下降趋势，且季节性变化明显。ARFIMA模型比ARIMA模型能提供更准确的结核病发病率预测，尤其是在捕捉长期趋势和季节性方面。在人群层面有效管理结核病需要对疾病模式进行适当监测和了解。预测是检测与预期趋势偏差的关键工具，这可能预示着疾病动态的变化。持续使用ARFIMA模型对于指导公共卫生干预措施和确保及时应对结核病控制中出现的新挑战至关重要。
