This was an observational cohort study using the Optum Clinformatics® Data Mart (CDM) database to compare cardiovascular outcomes in adults with T2D and established ASCVD who initiated a OW GLP-1 RA or a DPP-4i. The study period was from January 1, 2017, to September 30, 2021. ASCVD history was examined back to 2001. OW GLP-1 RAs studied included exenatide, dulaglutide, and semaglutide. DPP-4is included sitagliptin, saxagliptin, linagliptin, and alogliptin. Individuals who switched within the same drug class during the follow-up period were included, but those who switched between classes were excluded.

The index date, which was between January 1, 2018, and June 30, 2021, was defined as the prescription date of the index drug (OW GLP-1 RA or DPP-4i). The beginning of the index window was selected to fall immediately after the approval of the newest agent included in this study (once weekly semaglutide, approved for T2D in December 2017). One year prior to the index date constituted the baseline period. Individuals were followed for ≥ 3 months, until censoring due to the earliest of death; end of the study (September 30, 2021); new initiation of a sodium-glucose cotransporter-2 (SGLT-2) inhibitor, DPP-4i (among those in the GLP-1 RA group), or GLP-1 RA (among those in the DPP-4i group); lapse of continuous enrollment; or discontinuation of the index drug for > 60 days. The follow-up period, in which outcomes/end points were assessed, was the interval between the index date and end of follow-up.

The Optum CDM contains administrative claims for enrollees of commercial health care plans and Medicare Advantage across the US. The administrative claims include verified, adjudicated, adjusted, and de-identified medical and pharmacy claims. The CDM also includes outpatient laboratory test results from large national laboratory vendors.

Individuals included in the study had ≥ 2 separate diagnoses of T2D on different dates during the study period, identified using the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) code E11 in the primary or secondary diagnosis positions; ≥ 1 prescription for the index drug; and use of the index drug for ≥ 90 days (with ≤ 60-day gaps). Individuals were ≥ 18 years old on the index date and had a history of ASCVD between 2001 and the index date, identified using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) and ICD-10-CM codes in any position (see Additional file 1).

Exclusion criteria included baseline GLP-1 RA or DPP-4i use, missing demographic information (age or sex), or pregnancy or type 1 diabetes at any time during the baseline or follow-up periods.

Detailed baseline characteristics are listed in Table 1 and Additional file 2. As this study used only de-identified patient records and did not involve the collection, use or transmittal of individually identifiable data, institutional review board approval was not required.

Clinical effectiveness outcomes included ischemic stroke, MI, and their composite. Ischemic stroke events were identified as a primary diagnosis of inpatient claims using ICD-10-CM code I63 (except I63.1, I63.4, and I63.6; see Additional file 1). One hospitalization for ischemic stroke was considered one stroke event. MI was measured as a primary diagnosis of inpatient claims using ICD-10-CM code I21 or I22 (see Additional file 1). One hospitalization for MI was considered one MI event. An ischemic stroke or MI event was defined as the composite of ischemic stroke and MI [14]. Incidence rates and time to event occurrence were evaluated (events could be incident or recurrent).

HCRU outcomes included ASCVD-related and all-cause outpatient visits, hospitalizations, and emergency room (ER) visits. Cost outcomes included ASCVD-related and all-cause hospitalization costs and total medical costs. Total medical costs included medical costs from outpatient visits, hospitalizations, and ER visits. ASCVD-related HCRU and cost outcomes were measured using ICD-10-CM codes in the primary or secondary position (see Additional file 1). All-cause HCRU and medical costs included costs for any diagnosis, including ASCVD. The Optum CDM reports an estimated cost that is standardized based on a resource-based relative value scale derived from observed costs paid by the insurer, rather than the original paid amount.

Descriptive statistics were presented for all outcomes and covariates in the OW GLP-1 RA and DPP-4i groups. Counts and frequencies were used for categorical variables and means and standard deviations for continuous variables. Costs were reported as per person per month (PPPM) and adjusted to the year 2021. Incidence of first (or recurrent) stroke or MI in the follow-up (for those without and with a prior history of stroke/MI, respectively) was reported as number of events per 1000 person-years.

To reduce the observed selection bias between the 2 groups, inverse probability of treatment weighting (IPTW) [15] using average treatment effect weights was derived by conducting a logistic regression with the following variables: age, sex, race/ethnicity, geographic region, index year, Charlson Comorbidity Index, diabetes complications severity index, type of ASCVD history, comorbidities, glucose-lowering therapy, HbA1c, body mass index (BMI), and number of all-cause hospitalizations 60 days prior to the index date. For baseline characteristics, descriptive statistics were reported with and without IPTW. Standardized mean differences (SMDs) were presented. SMD < 10% was considered not significantly different for baseline characteristics. Weighted descriptive statistics were reported for all outcomes. For time-to-event outcomes, weighted cumulative incidence curves were generated and log-rank tests and Cox proportional-hazards (Cox-PH) regressions were conducted (proportional-hazards assumptions were met in these models). For HCRU and costs, generalized linear models with quasi-Poisson distribution and log-link function were used.

Interaction and stratified analyses were also conducted for the following variables: with and without history of ischemic stroke or MI, and end of follow-up before and after March 1, 2020, for HCRU and cost outcomes (to test the impact of COVID-19 on HCRU and costs). Sensitivity analyses excluding exenatide OW or including the prescriber type in weighting were also conducted. To assess residual unmeasured bias, negative control outcomes (ie, breast cancer and prostate cancer) and E-values were also examined [16]. To reduce bias due to informative censoring, inverse probability of censoring weighting (IPCW) was applied to assess clinical outcomes. To allow for a longer follow-up time, additional sensitivity analyses were conducted with the index date selection window restricted to between 2018 and 2020. Finally, to provide a complementary perspective, intention-to-treat (ITT) analyses were also performed for clinical outcomes.