SARS病例数的统计学与灰色模型预测效果评价  

Evaluation on statistical and grey models to predict daily SARS cases

作　　者：林昆[1] 朱碧柳[1] 曾旸[1] 张建军[1] 张庆英[1] 罗家逸[1] 

机构地区：[1]汕头大学医学院卫生学教研室,广东汕头515041

出　　处：《华南预防医学》2007年第2期13-17,共5页South China Journal of Preventive Medicine

基　　金：2006年汕头市重点科技计划项目(汕府科200682号)

摘　　要：目的探索香港SARS流行的规律,寻找预测SARS短期流行的方法。方法香港每日SARS病例数据(2003年3月11日至2003年6月12日)来源于香港卫生署。采用数理统计学与系统工程学的方法,包括曲线拟合、γ分布、时间序列以及灰色动态GM(1,1)模型等13种方法对每日病例数进行预测。使用预测误差、总体95%可信区间和实际病例数与预测病例数的配对t检验对13种方法的预测效果进行评价。结果对香港SARS日累计病例数预测误差最小的方法分别是:流行高峰期为时间序列、流行下降期为γ分布、流行终末期为灰色模型,其预测误差的中位数分别为0.29%、0.02%和0.03%,且3种方法的预测病例数平均数均不超过实际病例数平均数的95%可信区间。在流行全程预测方面,灰色模型对日累计病例数的预测误差中位数最小,为0.16%;γ曲线和时间序列对日发病数的预测误差中位数较小,分别为0.27%和3.09%。结论时间序列、γ分布和灰色动态GM(1,1)模型对香港SARS流行的短期预测效果较好。Objective To determine proper methods to predict SARS short- term epidemic according to patterns of SARS outbreak in Hang Kong. Methods Data on daily SARS cases was from Hong Kong Department of Health from March 11, 2003 to June 12, 2003. 13 statistical models including curve fitting, γ - distribution, time series and dynamic grey model（ 1,1 ） and so on to predict the daily SARS cases. Predicted error, 95% confidence interval of population and paired t test between objective and predicted value were performed to evaluate predictive efficacies of 13 different models. Results The minimal predicted errors were observed in time series model during peak period （ Median 0. 29% ）,γ - distribution at declined period（0.02% ） and dynamic grey model（ 1,1 ） （0.03%） at terminal stage of epidemic regarding daily accumulative SARS cases in Hang Kong. Moreover, the mean cases predicted from above 3 methods were within the 95 % confidence interval of actual mean cases. The dynamic grey model（ 1,1 ） had minimal predicted error（0. 16% ） at whole stage of epidemic as to daily accumulative SARS cases as well as γ - model （ 0. 27% ） and time series （ 3.09% ） as to daily SARS cases. Conclusion Time series, γ model and dynamic grey model（ 1,1 ） were perfect to predict short-term SARS epidemic in Hong Kong.

关 键 词：严重急性呼吸综合征 流行病学 预测 

分 类 号：R563.1[医药卫生—呼吸系统] R195.4[医药卫生—内科学]