We conducted an observational analysis of Veterans’ ratings of health care experiences in community care settings by race and ethnicity using the VA SHEP-CCS for the period 2016–2021.20 The SHEP-CCS is a mixed-mode (internet/mail) survey administered to Veterans who received VA-funded community care over the prior 3 months. The survey asks Veterans to rate their experiences across nine domains.19,21,22 Sampling for the SHEP-CCS is random within strata, which reflect the type of care received (e.g., primary care, psychiatric care, other subspecialty care). We linked respondent-level data from the SHEP-CCS to data from the Centers for Medicare and Medicaid Services (CMS) to identify Medicare and Medicaid enrollment, the VA Corporate Data Warehouse (CDW) to obtain information on demographics, VA priority group status, health conditions, geography, and the Veterans Integrated Services Network (VISN) where care was received. VISNs are regional divisions of the Veterans Health Administration that manage VA medical centers and other medical facilities (nationally, there are 18 VISNs).23 The VA Pittsburgh Healthcare System Institutional Review Board approved this study.

SHEP-CCS had a response rate of 30.7% during the study period, in line with other surveys of patient-reported care experiences.24 A total of 233,634 respondents to the SHEP-CCS were identified from 2016 to 2021. We excluded 188 respondents who could not be linked to VA CDW data; 1093 respondents who resided outside of the 50 US states or Washington, D.C.; and 484 respondents without county-level geographic identifiers (needed to measure county-level covariates). From this sample, we analyzed differences in community care experiences based on ethnicity (n = 16,490 Hispanic and 200,725 non-Hispanic Veterans) and race (n = 24,306 Black/African American and 180,313 White Veterans).

We examined experiences in nine domains: overall satisfaction with community care, overall rating of the provider, eligibility determination for VA community care, first appointment access, scheduling a recent appointment, provider communication, care coordination, non-appointment access (e.g., after-hours access to providers, waiting time in the office), and billing.21 We followed domain-item groupings for the VA-SHEP survey to combine responses to individual survey items into domain scores (see Appendix for details). Respondent-level scores were constructed as equally weighted means of ratings of domain items. Items were linearly converted to 0–100-point scales before aggregation into scores. Higher scores represent greater satisfaction with care.

We analyzed SHEP respondents according to their self-reported race and ethnicity. Race and ethnicity are social constructs and reflect the influence of social, political, and economic forces that lead to institutional inequity and interpersonal discrimination.25,26,27 We compared community care experiences of Veterans by race (comparing those who identified as Black or African American vs. White, regardless of ethnicity) and ethnicity (comparing those who identified as Hispanic vs. non-Hispanic, regardless of their race). Veterans who identified as Black/African American in addition to other racial groups were analyzed as Black. We did not separately analyze care experiences among racial or ethnic groups with smaller representation in the SHEP, such as American Indian or Alaska Native and Asian Veterans, because smaller sample sizes limited our ability to make meaningful comparisons.

We assessed age, gender, health, insurance and socioeconomic status, rural residence, county-level supply of health care providers, and type of community care received. To measure health status, we used the Elixhauser Comorbidity Index28 along with separate indicators for the presence of a serious mental illness (i.e., bipolar disorder, major depression, post-traumatic stress disorder, schizophrenia, or psychosis) and substance use disorder (i.e., substance use disorder related to drug or alcohol use). We identified these conditions using diagnoses on Veterans’ health care records in the VA CDW (care provided within VA) and in VA Program Integrity Tool files (administrative claims for community care) in the 2 years preceding the SHEP-CCS survey. We measured socioeconomic status and insurance using VA priority group status (which reflects Veterans’ income and service-connected disabilities29) and with indicators for enrollment in Medicaid, Medicare, Medicare Part D, and Medicare Advantage.30,31 Insurance is correlated with socioeconomic status and may impact access to care outside of the VA healthcare system. We included county-level measures of urban vs. rural residence (large metropolitan area, small metropolitan area, micropolitan area, and rural) and supply of community physicians per 1000 county residents.32 Finally, we included indicators for category of community care received: primary care, subspecialty care, surgical care, eye care, acupuncture, psychiatric care, and other care.

We plotted unadjusted ratings of community care experiences to examine trends and racial and ethnic differences in ratings over the study period. Next, for each domain score, we ran three sets of respondent-level linear regression models to estimate racial and ethnic differences in community care experiences. We constructed sequential models, guided by the National Academy of Medicine’s framework for examining health care disparities. According to this framework, disparities represent racial or ethnic differences that are not explained by group differences in health status or care needs. The framework considers how geographic, socioeconomic, and insurance factors may mediate racial and ethnic disparities in care.33

Accordingly, Model 1 adjusted for demographic factors (age and sex), health status, indicators for the category of community care received, and year fixed-effects. Model 2 adjusted for all variables in the first model as well as rurality and county-level supply of physicians. Model 3 further adjusted for education, along with socioeconomic and insurance factors (included together because many differences in insurance coverage are income-related). Sequential adjustment for covariates allowed us to quantify the extent to which differences in ratings persisted after adjustment for geography, socioeconomic status, and insurance. We also conducted a sensitivity analysis that controlled for individual Elixhauser comorbidities instead of a linear comorbidity index.

We conducted a secondary analysis to explore whether Veterans differed in their likelihood of reporting high vs. low ratings of care by race or ethnicity.9,34 For each domain score, we estimated logistic regression models to test for differences by race or ethnicity in the probability of rating care on that domain at or above 90th percentile (high rating) or at or below 10th percentile (low rating). Percentiles were constructed for each patient experience domain among all SHEP-CCS respondents. We estimated marginal differences in the probability that Black vs. White or Hispanic vs. non-Hispanic Veterans reported high vs. low ratings of care, adjusting for all covariates in Model 3.

All estimates were weighted to account for survey sampling and non-response using STATA version 15. Statistical tests were conducted using a two-tailed 5% type-I error rate.