Comparative Analysis of Calendar Time-Specific and Conventional Propensity Score Analysis for Thiazolidinedione Use in Diabetes

OBJECTIVES: To evaluate whether there are differences in propensity score (PS) and treatment effects estimated using conventional and calendar time-specific PS (CTS-PS) approaches.

METHODS: A retrospective database analysis at a university-affiliated hospital in Thailand was used. Diabetic patients receiving glucose-lowering medications from July 2008 to June 2011 were included. Patients were categorized into those exposed and not exposed to thiazolidinediones (TZDs). PSs were estimated by using conventional PS and CTS-PS. In the CTS-PS, PS was separately estimated for three specific calendar time periods. Patients were matched 1:1 using caliper matching. The outcomes were cardiovascular and all-cause hospitalizations. The TZD and non-TZD groups were compared with Cox proportional hazard models.

RESULTS: A total of 2165 patients were included. The average conventional PS was 0.198 (95% confidence interval [CI] 0.195-0.202), while the average PS in the CTS-PS approach was 0.212 (0.206-0.218), 0.180 (0.173-0.188), and 0.205 (0.197-0.213) for July 2008 to June 2009, July 2009 to June 2010, and July 2010 to June 2011, respectively. The average difference in PS was 0.012 (P < 0.001), -0.009 (P ≤ 0.002), and 0.000 (P = 0.950) in the three calendar time periods. The adjusted hazard ratios of the conventional PS-matched cohort were 0.97 (95% CI 0.39-2.45) and 0.97 (95% CI 0.78-1.20) for CVD-related and all-cause hospitalizations, while the adjusted hazard ratios of the CTS-PS-matched cohort were 1.11 (95% CI 0.43-2.88) and 1.12 (95% CI 0.91-1.39), respectively.

CONCLUSION: CTS-PS is different from PS estimated by using the conventional approach. CTS-PS should be considered when a pattern of medication use has changed over the study period.