(276) Addressing bias due to socioeconomic status in the comparative effectiveness of sodium-glucose co-transporter 2 inhibitors vs. sulfonylureas in patients with type 2 diabetes
Research Fellow Division of Pharmacoepidemiology, Dept. of Medicine, Brigham & Women’s Hospital, Harvard Medical School Boston, United States
Background: Few studies have addressed the potential for bias occurring when comparing initiators of newer diabetes medication classes relative to those of older and cheaper diabetes agents, such as sulfonylureas (SUs), especially in the US, where out-of-pocket costs of newer medications can be cumbersome.
Objectives: To address potential bias due to differential socioeconomic status (SES) in comparing the cardiovascular effectiveness of sodium-glucose cotransporter 2 inhibitors (SGLT2i) relative to SUs.
Methods: Using Medicare Data [2013-19], we identified patients >65 years with type 2 diabetes (T2D) initiating SGLT2i vs. SU. Primary outcomes were modified major adverse cardiovascular outcomes (MACE) [myocardial infarction, stroke, or all-cause mortality] or heart failure hospitalization (HHF) based on validated claims-based algorithms. We estimated HR (95% CI) after 1:1 propensity score (PS) matching, adjusting for 130 baseline covariates. We conducted as-treated analyses censoring for treatment changes. To account for confounding and differential dropout due to SES, we (i) adjusted for dual Medicaid eligibility, low-income subsidy amount, out-of-pocket cost, and SES score (a validated zip-code level geo-coded measure of income and other SES), (ii) performed PS matching separately within each quintile of the SES score (iii) stratified analyses by matched sets restricting to matched sets with similar follow-up times, and (iv) conducted censoring-weighted analyses. We compared linagliptin (a DPP-4 inhibitor) vs. glimepiride (a SU) as a negative control comparison with the CAROLINA trial estimates for the primary outcome (MACE) as a benchmark.
Results: We identified 58,055 matched pairs of SGLT2i and SU initiators with balanced baseline characteristics. Mean age was 72 years, 52% were male, and 44% had baseline CVD history. Overall, we observed 5.7 and 11.2 HHF events per 1000 PY in SGLT2i and SU, with an HR (95% CI) of 0.51 (0.45, 0.58). For modified MACE, the incidence rates per 1000 PY were 21.3 and 29.6, with an HR of 0.72 (0.67, 0.77). All analyses accounting for confounding and differential censoring due to SES provided similar findings. For linagliptin vs. glimepiride, the MACE estimates were similar to those from the CAROLINA trial: HRs: 0.94 (0.87, 1.01) vs. 0.98 (0.84, 1.14) for the CAROLINA trial.
Conclusions: Among older adults with T2D, SGLT2i were associated with reduced HHF and modified MACE outcomes vs. SU, a finding robust to analyses accounting for SES. We will repeat analyses in the UK Clinical Practice Research Datalink to compare estimates in a single-payer healthcare system.
