时间序列模型在月平均住院日预测中的应用及评价

The Application and Evaluation of the Time Series Model in Prediction of Monthly Mean Hospital Stay

目的:对某综合医院的月平均住院日建立时间序列模型,利用指数平滑法和ARIMA模型对其进行模拟评价及短期预测。方法:在某综合医院统计报表中提取2014年1月—2017年12月的月平均住院日,其中2014年1月—2017年6月的数据用于创建时间序列,利用SPSS20.0进行统计分析,分别采用指数平滑法和ARIMA对创建的时间序列拟合模型,评价模型效果,并对2017年7月—2017年12月的平均住院日进行预测,比较实际值与预测值间的符合程度。结果:指数平滑法模型:平稳的R方为0.814,表明拟合程度较好。白噪声序列的Ljung-Box检验无统计学意义(Q18=18.730,P=0.226)。模型参数估计中平滑参数Alpha的估计值为0.200,且参数检验结果有统计学意义(T=2.106,P=0.042)。ARIMA模型:平稳的R方为0.361,Ljung-Box检验无统计学意义(Q18=15.215,P=0.580)。AR的参数检验有统计学意义(T=-4.652,P<0.001),为-0.654。两模型实际值与预测值间的相对误差绝对值均小于5%。结论:指数平滑模型比ARIMA拟合及预测效果更好,是某综合医院月平均住院日的首选预测模型,为医院管理决策提供科学依据。

Objective: To build the time series model of average length of stay in a month in a general hospital by exponential smoothing models and ARIMA models respectively, assess these models and provide short-term prediction. Methods: We exert the average length of stay in a month in a general hospital for 4 years from 2014 to 2017, which is split into two parts. One part containing the months from January 2014 to June 2017 is used to create time series model by the exponential smoothing method and ARIMA respectively, then access the effect of fitness. The other one covering the months from July 2017 to December 2017 is applied for prediction. The data analyze process goes on by SPSS 20.0. Results: The exponential smoothing model performs well with stationary R-squared 0.814. The Box-Ljung test on white noise series is not statistically significant(Q18 =18.730, P =0.226). The estimate of the smoothing parameter Alpha is 0.200, showing statistical significance(T =2.106, P =0.042). The ARIMA model: stationary R-squared is0.361, the Box-Ljung test is not statistically significant(Q18 =15.215, P =0.580), and the value of AR is-0.654, statistically significant(T =-4.652, P <0.001). The absolute relative errors between actual and predicted values are less than 5% in both methods. Conclusion:The exponential smoothing model performs better than ARIMA model in fitness and prediction by adopting the data, which can provide some useful information for the management and decision making.