A Study Of Bias Correction When Different Missing Mechanisms Coexist In A Survey Of Medical Expenditure And Its Influence Factors For Urban Residents Enrolling In Urban Resident Basic Medical Insurance

Objective:In the study of measuring the medical expenditure and exploring its influence factors for urban residents enrolling in Urban Resident Basic Medical Insurance(URBMI) in Taiyuan city in 2013, Most of the previous studies using only the hospital outpatient and inpatient case information lead to insufficient sample representativeness; Non response bias and selection bias coexist in dependent variable of the survey data. Unlike previous literature on missing data analysis usually only for a missing mechanism, we put forward a two-stage method to deal with two missing mechanisms simultaneously, combining multiple imputation dealing with missing at random and sample selection model dealing with nonignorable missing. Methods:The objects in the data covers all kinds of urban residents owing URBMI, ranging from infants to the adults. The two-stage sampling design was taken in before the questionnaire survey. A simulation study applying the two-stage method was carried out to 5871 qualified cased after data cleaning with 3.68% nonignorable missing and 13.40% missing at random in dependent variable. In the first stage, making use of non-missing data(excluding the individual observations with non-random missing data) to impute the random non-response data which are missing at random(MAR) by multiple imputation, including Predictive Mean Matching(PMM) Method, Propensity Score(PS) Method, Markov chain Monte Carlo(MCMC) method and EMB algorithm. And then combined them with nonignorable missing data.; In the second stage, the two-step estimation of heckman selection model was used to the combined data to correct the selection bias. Based on 1000 times resampling, the best scheme of filling the random missing values is the Markov chain Monte Carlo(MCMC) method under the missing proportion. With this optimal scheme, we applied the two stage method to the actual survey data with variance estimation of the regression coefficients. Results:Finally, we find the influence factors of annual medical expenses of residents enrolling in URBMI in Taiyuan city include: respondents type, gender, whether or not having a other kind of insurance, self feeling of health, gross annual household income, chronic diseases, accepting hospital, outpatient clinics service, hospitalization, the recognition of medical insurance, take a self medication. Conclusions:The two-stage method combining multiple imputation and sample selection model can deal with non response bias and selection bias effectively in dependent variable of the survey data.