Missing Data Processing Method And Its Application In Clinical Trials

Objective:To perform sensitivity analysis of efficacy evaluation based on an RCT-designed clinical trial through different missing data processing methods,which providing methodological guidance for common missing data analysis in clinical trials.Methods:A total of 452 patients with growth hormone deficiency were enrolled in the randomized controlled trial,who were randomly divided into low-dose group?0.12mg/kg/w?and high-dose group?0.20mg/kg/w?for a kind of recombinant human growth hormone treatment according to the ratio of 1:1.After three patients with error inclusion or non-medicated patients were excluded,there were 224 patients in the low-dose group and 225 patients in the high-dose group.Efficacy evaluating items were height standard deviation score based on actual age(Ht-SDS_CA),height standard deviation score based on bone age(Ht-SDS_BA),and height velocity?HV?.Monte Carlo simulation is used to evaluate the robustness of missing data processing results by multiple imputation,EM algorithm and random forest propensity score joint doubly robust inverse probability weighting method under the condition of mean absolute error?MAE?and mean-square error?MSE?in different sample size and missing data ratio.In addition,clinical trial cases are combined for application analysis.Results:?1?The simulation results show that when the sample size is fixed,MAE and MSE from three methods are increasing with the proportion of missing data,when the proportion of missing data is fixed,the MAE and MSE are shrinking with the sample size increasing.?2?The smallest MAE and MSE are from random forest propensity score joint doubly robust inverse probability weighting method,which indicate the best accuracy of its effect estimates,and outcomes from multiple imputation method are similar to the EM algorithm.?3?Based on the clinical trial data of RCT design,Ht-SDS_CA,Ht-SDS_BA and HV were analyzed.The results showed that for the items Ht-SDS_CA and HV,the effect of the high-dose group was higher than that of the low-dose group,and the difference was statistically significant while the difference between the groups of Ht-SDS_BA was not statistically significant.The multiple imputation method,EM algorithm and random forest propensity score joint doubly robust inverse probability weighting method were used to handle the missing data,with the obtained treatment effect values including point estimation and interval estimation,were basically consistent with the results of LOCF imputation.Conclusion:The multiple imputation method,EM algorithm and random forest propensity score joint doubly robust inverse probability weighting method are well applicable for handling missing data in RCT clinical trials,with results basically consistent with LOCF method.After comparison,random forest propensity score joint doubly robust inverse probability weighting has the most accurate estimates.