In this cross-sectional study, we used VA administrative claims and electronic health record (EHR) data from the VA Corporate Data Warehouse that included demographics, clinical encounters, diagnoses, procedures, laboratory values, vital signs, echocardiography results, and pharmacy dispensing records. We used administrative claims from outside of the VA to capture healthcare services that were paid for by the VA or Medicare, including non-VA medication prescriptions. We used mortality data from the Corporate Data Warehouse and the VA Vital Status File. This study was approved by the Stanford University Institutional Review Board.

Veterans with prevalent HFrEF on January 1, 2022, were identified based on a diagnosis of HF and a documented ejection fraction (EF) ≤40%. We included Veterans with a primary inpatient diagnosis code or 2 or more total diagnosis codes for HF in the preceding 3 years across VA, Medicare, and community care data (diagnosis codes in Supplemental Table I). We identified patients with most recent documented left ventricular ejection fraction (LVEF) of 40%. We excluded Veterans with a ventricular assist device or heart transplant or a documented date of death prior to January 1, 2022. Patients without medication prescription fills or without VA primary care or cardiology visits in the preceding year were excluded.

We selected a broad range of patient characteristics that could have implications on GDMT use including demographics, vital signs, comorbidities (identified by ICD codes), and laboratory values. Race and ethnicity were classified using exclusive hierarchic categories consistent with prior studies on this topic: Hispanic of any race; non-Hispanic Black/African American (AA), non-Hispanic Asian, and non-Hispanic White. ^, For simplicity, we refer to each racial category without noting the exclusion of Hispanic ethnicity. Race and ethnicity data was over 98% self-identified from the VA EHR.

The most recent outpatient vital signs and laboratory variables were included within 2 years prior to the study date. Diagnosis codes for medical comorbidities within 3 years preceding the study date were captured (Supplemental Table I). LVEF data were previously extracted from clinical notes as well as echocardiography and radiologic reports using natural language processing, as has been previously described. ^, We collected data on outpatient visits to primary care and cardiology. HF hospitalizations included those within the VHA or paid for by Medicare or VHA in the community in the last year. VA facility characteristics were also captured.

To facilitate the comparison of rural and non-rural Veterans, we were interested in other sociodemographic patient information, the methodology of which has been previously described First, to denote rurality, we used 2010 RUCA (Rural–Urban Commuting Areas) codes, which categorize U.S. census tracts into codes by degree of rurality According to VA definitions, RUCA codes of 1.0 and 1.1 are classified as nonrural. Second, we analyzed patients’ drive time to VA primary care and VA specialty care, which is calculated based on distance, route options and speed limits. Third, we identified neighborhood social vulnerability, as defined by the Centers for Disease Control (CDC) Social Vulnerability Index (SVI) Fourth, we defined VA facility characteristics, including complexity level, teaching status, and geographic region

We identified treatment with GDMT for HFrEF: RASi, including angiotensin-converting enzyme inhibitors (ACEIs), angiotensin receptor blockers (ARBs), or ARNIs, BB, MRA and SGLT2i. ² The medications included are listed in Supplemental Table IV. We did not evaluate treatment rates for other HFrEF medical therapies, including hydralazine/nitrates or ivabradine. We then assessed whether patients were prescribed at least 50% of target doses for BBs and RASis For each class, we identified the most recent filled prescription within 120 days preceding the study date and calculated the estimated daily dose based on the prescribed dosage and the number of pills per day. For each therapy, we excluded patients with medication contraindications who were not prescribed that medication. As described previously, patients were considered to have a contraindication to therapy based on allergies, vital signs, laboratory data, or comorbidities. Details regarding medication contraindications are reported in Supplement Table III.

GDMT composite scores have shown promise as quality improvement and clinical research metrics to assess GDMT intensity and comprehensiveness. ^, We used a GDMT composite metric that is a continuous version of the 4-pillar score (C4P). This metric focuses on the 4 HFrEF therapies with clear mortality benefit (BB, RASi, MRA, and SGLT2i) taking into account dose with a maximum score of 18 and is associated with cardiovascular outcomes The scoring algorithm is described in Supplemental Table II. We first calculated the C4P score for all patients, regardless of contraindications, by dividing the numerator by the maximum score (18). We then adjusted the score (C4P score B) based on contraindications by reducing ineligible medications from the denominator. For the target dose metrics and C4P score, patients were deemed to reach their maximally tolerated dose if they developed hypotension, bradycardia, or worsening renal disease, as described in the list of contraindications in Supplemental Table III.

We compared categorical variables via chi-squared tests and continuous variables via t tests. We presented continuous variables as mean (standard deviation), and categorical variables as number (percentage). We calculated the C4P score as a composite metric, reporting a version excluding therapy contraindications (C4P Score A) and including contraindications (C4P Score B). A high GDMT score was considered a C4P score >12. We compared the C4P score between rural and non-rural Veterans overall and stratified by other patient and facility characteristics.

We used multivariable logistic regression to identify patient characteristics associated with a high C4P score, excluding (Score A) and including contraindications to therapy (Score B). We evaluated 4 models: (Model 1) adjustment for sociodemographic characteristics, (Model 2) Model 1 with facility adjustment, (Model 3) Model 1 with adjustment for medical comorbidities, and (Model 4) Model 3 with facility adjustment. We adjusted for facility with mixed effects logistic regression models with the facility as a random intercept. For models without the facility-level random intercept, we clustered standard errors by facility. We evaluated the facility-level variation by calculating the median odds ratio from the mixed effects logistic regression model. The median odds ratio represents the relative odds of high GDMT score at 2 different facilities for a similar patient We then evaluated the association between VA facility region and complexity with the odds of high composite GDMT score. We repeated the analysis with adjustment for patient characteristics and among the subgroup of rural Veterans. We used a 2-sided α = 0.05 to define statistical significance. Data analyses were performed with SAS, version 9.4 (SAS Institute Inc) and Stata, version 15.1 (StataCorp LLC).