Using Secondary Data to Evaluate Sex-based Heterogeneity of GLP-1 Agonists and SGLT2 Inhibitors on Cardiovascular-Kidney-Metabolic Health (CKMH) Outcomes in Real-world Settings (DASH-CKMH)

Overview

Complex pathophysiological interactions among obesity, metabolic risk factors, chronic kidney disease (CKD), and the cardiovascular system lead to poor cardiovascular-kidney-metabolic health (CKMH), which is a major determinant of premature morbidity and mortality. Poor CKMH may lead to cardiovascular-kidney-metabolic syndrome (CKMS) - the five-stage framework introduced by The American Heart Association (AHA) which accounts for the critical overlap between cardiorenal syndrome and cardiometabolic disease. Evidence from randomized controlled trials shows glucagon-like peptide-1 receptor agonists (GLP-1RAs) and sodium-glucose co-transporter-2 inhibitors (SGLT2is) may improve CKMH in individuals with Type 2 Diabetes (T2D) and/ or obesity. However, there is modest evidence suggesting differential effectiveness of GLP-1RA and SGLT2i drugs between males and females. The extent of these sex-based differences is currently unknown. In part, this may be due to underrepresentation of females in clinical trials. Exploring sex-based differences in GLP-1RA and SGLT2i treatment on CKMH outcomes is important to inform CKMS treatment and equity in CKMH. Robust secondary data sources present the opportunity to elucidate sex heterogeneity in GLP-1RA and SGLT2i treatment on CKMH outcomes. Using a target-trial emulation design, this study aims to observe differences in long-term CKMH outcomes between patients treated by GLP-1RA and SGLT2i medications versus those treated with active comparator medications, and whether there is an observed interaction between sex and treatment.

A target-trial design will be conducted in three sources of secondary data: 1)Merative Marketscan (claims-based data derived from commercial insurers), 2) All of Us (public database of Electronic Health Record \[EHR\] and survey data), and 3) LifeScale (EHR- derived from The Ohio State University Wexner Medical Center). To construct a clinically similar comparator group, we opted for patients treated with active comparator medications with similar indications to the GLP-1RA and SGLT2i intervention medications. The intervention group is defined as patients with any exposure to the intervention medications, and the comparator group is defined as exposure to the comparator medications with no exposure to the intervention medications in the 30 days following index. To balance baseline characteristics between intervention and comparator groups, for each cohort established in the three secondary data sources we will apply propensity score matching. Matching variables will include the following confounders: index age, U.S. region, race/ethnicity (where available), rurality, insurance type, Charlson comorbidity index score, index year, Medicaid expansion status in state of residence. Match quality will be assessed by examining standardized mean differences (SMDs) of matching variables by treatment and control, with SMD ≤ 10% indicating a well-balanced cohort after matching. The primary outcome of interest is 3-P MACE (three-point major adverse cardiovascular event), defined as any of the following: nonfatal myocardial infarction, nonfatal stroke, or cardiac-related death. Secondary outcomes include: all-cause mortality, advancing CKMS stage, stroke, myocardial infarction, incident coronary heart disease diagnosis, incident peripheral artery disease diagnosis, atrial fibrillation diagnosis, renal failure, kidney transplant, and kidney dialysis. Primary and secondary outcomes are time-to-event variables; thus, differences in risk of outcomes between intervention and comparator groups will be tested using survival analysis methods. Kaplan-Meier survival curves will be used to visualize the risk of the outcomes between intervention and comparator up to five years, and Cox modelling will be used to adjust for residual confounding and examine whether differences in risk are significant between intervention and comparator groups. The Cox models will include a variable for treatment (intervention versus comparator), sex, and a sex-treatment interaction term. The analyses will be conducted separately for each of the three secondary data sources. To evaluate our secondary hypothesis of whether the target trial emulation studies in the three data sources are aligned, we will compute 3 binary metrics for each outcome of interest: 1) full statistical significance agreement: treatment effect estimates and 95% confidence intervals (CIs) on the same side of the null, 2) estimate agreement: treatment effect estimates fell within the 95% CI of one another; and 3) standardized difference agreement: standardized differences between treatment effect estimates.