【摘要】 目的将零-K膨胀泊松回归模型(ZKIP)在公共卫生领域研究中加以实践,为实际工作中研究者所遇到的零-K膨胀数据拟合问题提供方法学支持。方法以美国2016年5～17岁学生因病缺课天数数据为例,拟合ZKIP模型,采用SAS 9.4软件进行参数估计,并与Poisson回归模型、零膨胀泊松回归模型(ZIP)进行比较,依据数据过离散和零膨胀检验结果及模型拟合优度指标等选择最优模型,同时探讨各可能影响因素与学生因病缺课天数间的联系。结果数据过离散和Vuong检验结果表明存在过离散和零膨胀现象(O=773.40>1.96、V=19.7330>1.96);依据因变量取值分布直方图取膨胀值K=2。ZKIP模型取得最优模型拟合度(AIC=40736,BIC=41274),参数回归结果表明高年龄组学生因病缺课行为发生风险更高;白人学生因病缺课行为发生风险性最低;手术史、身体不适症状、长期服药、急诊室次数、过敏、平衡不佳情况等变量对数据零膨胀参数和Poisson分布参数影响均有统计学意义;视力问题、长期服药、地区(西部地区学生较东北部地区学生)3个变量对K=2的膨胀参数η影响显著。结论 ZKIP模型不仅在零事件处理能力上和ZIP模型相当,还能提供更多的K膨胀值处信息,是零-K膨胀数据拟合方法学上的突破。更多还原

【Abstract】 Objective In this stndy,the Zero-K Inflated Poisson Regression Model(ZKIP)is put into practice in the field of public health research,and it provides methodological support for the zero-K inflated data fitting problems encountered by researchers in practical work.Methods Taking the data of the number of days of absence from school due to illness of students eged 5～17 in the United States in 2016 as an example,the ZKIP model was fitted,and the parameters were estimated using SAS 9.4 software,and compared with Poisson regression model and zero-inflated Poisson regression model(ZIP).The optimal model is selected based on the data overdispersion and zero-inflated test results,and the model′s goodness-of-fit index.At the same time,the relationship between the possible influencing factors and the number of days students absent from school due to illness is discussed.Results Data overdispersion and Vuong test results show that there are overdispersion and zero inflated phenomena(O=773.40>1.96、V=19.7330>1.96);According to the histogram of dependent variable value distribution,take the inflated value K=2.The ZKIP model obtained the optimal model fitting degree(AIC=40736,BIC=41274),and the parameter regression results of ZKIP model showed that the high age group students were more likely to be absent due to illness;White students have the lowest risk of absenteeism due to illness;Variables such as surgical history,physical symptoms,long-term medication,emergency room frequency,allergies,and imbalance had statistically significant effects on the zero-inflated part and Poisson distribution part of the data;Vision problems,long-term medication,and region(students in western region are more than students in northeastern region)have significant influence on the inflated parameter η of K=2.Conclusion ZKIP model is not only comparable to the ZIP model in terms of zero event processing capacity,but also can provide more information at the K inflated value,which is a breakthrough in the zero-k inflated data fitting methodology.更多还原