EVALUATION OF EFFECTS ON BP AND CHOLESTEROL CHANGES AFTER MULTIPLE IMPUTATION FOR MISSING VALUES IN OBSERVATIONAL STUDY DATA OF POLYPILL

초록

Purpose The purpose of this study is to prove the effect of polypill on BP and cholesterol reduction effects by analyzing observational study data including missing. Method Medical records were collected for a minimum of 8 weeks and a maximum of 24 weeks after the first prescription of the polypill for each subject for patients who changed their administration to the polypill, which was collected in a retrospective observational study at 71 centers in Korea. In the collected data, there were 1412 subjects, but at 8 weeks, 16 weeks, and 24 weeks, BP and cholesterol information were 30–50% or more missing. Therefore, the available data were analyzed with GLM and LMM, and complete case data with missing removed and the imputed data by MICE estimated with pmm algorithm at all time points were analyzed by LMM. Results As a result of analyzing the available data with GLM in the improvement effect of SBP, it decreased relatively linearly until the 24th week, and as a result of analyzing the complete case data with LMM, overestimation appeared. However, as a result of analyzing the available data by MICE and LMM, the effect appears until the 8th week, which can be judged as relatively normal data, and is maintained thereafter. Available data can also be analyzed by LMM, and in this case, when the missingness is less than 50%, the results are similar to the case with MICE and LMM analysis. The improvement effects of DBP and LDL-C were similar to SBP. According to the data of this observational study, the administration of the polypill reduced SBP by 3.7 mmHg, DBP by 2.8 mmHg, and LDL-C by 14.8 mg/dL within 8 weeks of comsuming the polypill compared to the previous treatment. The effect was well maintained by 24 weeks. Conclusion Interpretation of such RWD requires a lot of attention. Basically, errors in analyzing repeatedly measured data with GLM and errors in randomly deleting subjects with missing data should be taken care of. In addition, if there are less than 50% missing, it is convenient to LMM the available data, but if there are more than 50% missing, an unreasonable result has shown. In conclusion, it is expected that the change of administration to polypill can show additional improvement effects in the control of blood pressure and lipids in patients who are taking two or more of the existing antihypertensive and antidyslipidemic drugs at the same time.