摘要
目的:选择最优时间序列模型进行重庆市痛风月发病趋势预测。方法:回顾性收集重庆市某三甲医院2011年1月至2014年12月痛风月发病数据,并采用指数平滑法、ARIMA(p,d,q)模型、GM(1,1)模型和Markov模型对其发病趋势进行拟合,根据平均百分绝对误差的大小判断模型优劣,选择最优时间序列模型进行痛风月发病趋势预测。结果:2012年9月后痛风月发病人数激增,并呈现一定的周期性。但就整个观察期而言,序列波动明显,周期性不明显。4种时间序列模型分析结果显示:实际发病数在各模型预测值95%的可信范围内波动,但不同模型预测结果均存在一定程度偏离。通过计算可知:ARIMA(1,1,0)、GM(1,1)和指数平滑法的平均百分绝对误差分别为24.72、109.21和24.95。因此GM(1,1)模型预测效果偏离最大,指数平滑法次之,ARIMA(1,1,0)模型拟合效果最贴近实际。结论:GM(1,1)模型拟合效果最差,ARIMA(1,1,0)模型拟合效果最好,建议用ARIMA(1,1,0)模型进行未来痛风发病预测。
Objective：To choose the optimal time series model for monthly gout incidence prediction. Methods：Monthly gout incidence data were collected retrospectively from an affiliated hospital of Chongqing Medical University from January 2011 to December 2014.And exponential smoothing method,ARIMA（p,d,q）,GM（1,1）and Markov model were adopted to fit the trend of monthly gout incidence,respectively. The efficiency comparison between different models was judged by mean absolute percentage error,and model with least mean absolute percentage error was used for future prediction. Results：Monthly gout incidence increased greatly,and cyclicality was observed after September 2012. But only sequence fluctuation was observed on the whole observation period. Results of trend prediction among four time series models were in consistent with each other,but their efficacy were different. The maximum and minimum mean absolute percentage error was in GM（1,1）model and ARIMA（1,1,0）model,respectively. Thus fitting results of ARIMA（1,1,0）model was mostly close to the actual situation. Conclusion：The ARIMA（1,1,0）model have the best imitative effect.Hence,ARIMA（1,1,0）model should be used for future monthly gout incidence prediction.
作者
胡敏
杜成凤
唐晓君
邓丹
Hu Min Du Chengfeng Tang Xiaojun Deng Dan(Teaching and Research Section of Health Statistics and Information Management,School of Public Health and Management,Chongqing Medical Universit)
出处
《重庆医科大学学报》
CSCD
北大核心
2017年第5期537-541,共5页
Journal of Chongqing Medical University
基金
国际合作资助项目(美国杜克大学全国健康研究中心
编号:x7383)
关键词
痛风
时间序列分析
趋势预测
模型选择
gout
time series analysis
trends prediction
model selection
