Outcomes of Veterans Treated in Veterans Affairs Hospitals vs Non–Veterans Affairs Hospitals

Key Points Question How do outcomes compare in Veterans Affairs (VA) hospitals and non-VA hospitals for 6 conditions for veterans aged less than 65 years and veterans 65 years and older? Findings In this cohort study of 593 578 hospitalizations and 414 861 patients, VA hospitalizations compared with non-VA hospitalizations had significantly lower 30-day mortality for heart failure and stroke, lower 30-day readmission for coronary artery bypass graft, gastrointestinal hemorrhage, heart failure, pneumonia, and stroke, but longer mean length of stay and higher mean costs for most conditions. There were differences by age group. Meaning These findings suggest that veterans had better outcomes in VA hospitals for some conditions at the expense of higher costs.


Introduction
The Veterans Affairs (VA) health care system is the only national integrated delivery system in the US.
Many of the 9 million veterans enrolled in the VA have access to non-VA care through VA-purchased services from community clinicians or concomitant enrollment in insurance programs.The VA has long purchased community care when services could not be provided on site, but the Veterans Access, Choice and Accountability Act (Choice Act) in 2014 followed by the VA Maintaining Internal Systems and Strengthening Integrated Outside Networks (MISSION) Act in 2018 expanded the criteria to purchase care for veterans experiencing access barriers. 1,2The Patient Protection and Affordable Care Act further expanded access to Medicaid for low-income adults, including veterans, in many states beginning in 2014. 1 These policies increased use of non-VA care and decreased use of VA services. 2,3creased access to non-VA care can lead to better outcomes if patients receive higher-quality or more timely care. 4][10][11][12][13][14] However, the veteran enrollee population is more male and has worse health status, greater disability, and lower incomes compared with the nonveteran population. 15,16Moreover, younger veterans are typically not included in comparisons due to a lack of comprehensive data on non-VA use outside of Medicare, which reduces the generalizability of these comparisons.[19][20][21] Inpatient care is a core service provided by the VA in 140 hospitals with medical or surgical acute care beds, which range widely in volume and service capabilities.3][24] At a time when veterans have more access to non-VA hospital care, it is important to examine differences in outcomes between VA and non-VA hospitals.
This study compared mortality, readmission, length of stay (LOS), and costs of veterans hospitalized in VA and non-VA hospitals for acute myocardial infarction (AMI), coronary artery bypass surgery (CABG), gastrointestinal (GI) hemorrhage, heart failure (HF), pneumonia, and stroke.A lack of data on non-VA utilization often hinders comparisons between VA and non-VA care, but we used a comprehensive data set of VA and non-VA all-payer inpatient care records.No studies to date compared hospital outcomes for veterans of all ages with access to VA care.

Methods
The cohort study was approved by the institutional review boards (IRBs) at Stanford University, University of Utah, and Greater Los Angeles VA with a waiver of consent granted by the IRBs.We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for reporting cohort studies.

JAMA Network Open | Health Policy
Outcomes of Veterans Treated in VA and Non-VA hospitals

Study Cohort and Data Sources
We conducted a study using repeated cross-sections of hospitalizations for VA enrollees discharged January 1, 2012 to December 31, 2017.After reviewing availability and policies to request all-payer discharge data for research in all states, we obtained hospitalization records in 11 geographically diverse states (ie, Arizona, California, Connecticut, Florida, Illinois, Louisiana, Massachusetts, Missouri, New York, Pennsylvania, and South Carolina), which allowed linkage between discharge data and VA enrollment data.Our sample of states represented the Northeastern, Southeastern, Midwestern, and Western regions of the US; approximately 38% of VA enrollees live in these states. 25terans' VA use and cost records were obtained from the Inpatient Encounter files and the Managerial Cost Accounting (MCA) files in the VA Informatics and Computing Infrastructure. 26terans' non-VA use records were obtained from state inpatient discharge data linked with VA enrollment data using either deterministic or probabilistic methods with personal identifiers.
We obtained patients' sociodemographic characteristics from the VA Health Enrollment Files and the VA Observational Medical Outcomes Partnership Files. 27,28Veterans' and VA hospitals' addresses were obtained from the VA Geospatial Services Support Center Files. 29Non-VA hospitals' addresses were obtained from the Centers for Medicare & Medicaid Services (CMS) Provider of Service File. 30 Veterans' death information was obtained from the VA Vital Status File.VA hospital characteristics were obtained from the Veterans Integrated Service Network Support Services Center, and non-VA hospital characteristics were obtained from CMS hospital cost reports. 31,32

Acute Medical or Surgical Hospital Stays
VA acute hospital stays were identified from medicine and surgery bed sections and diagnosisrelated group (DRG).We excluded stays within 30 days of discharge from a previous admission and stays for more than 180 days (not considered acute).Hospitalizations for AMI, CABG, GI hemorrhage, HF, pneumonia, and stroke were identified from principal diagnosis codes.We focused on these conditions since the Agency for Healthcare research and Quality uses hospital mortality for these conditions as a quality indicator. 33Discharge records for Illinois could not be obtained for 2012, so hospitalizations in Illinois were excluded in that year.The Figure shows how the final sample was derived.

Outcome Measures
Hospital outcomes included 30-day hospital mortality, 30-day readmission, inpatient costs, and LOS.
Hospital mortality was indicated for all-cause deaths occurring within 30 days of admission.
Thirty-day all-cause readmission was indicated for stays followed by another admission within 30 days of discharge regardless of where the stays occurred.Thirty-day mortality could not be measured for non-VA hospital stays in California and Pennsylvania since admission and discharge dates were not provided, and 30-day readmissions could not be measured for non-VA stays in California because no readmission indicator was provided in the discharge data.
VA costs included direct and indirect costs after subtracting national administration costs. 34n-VA costs included estimated professional fees 35 and facility charges which were adjusted by hospital cost-to-charge ratios. 31Costs were adjusted for inflation to 2017 dollars. 36,37LOS was calculated as the number of days between admission and discharge, inclusive.

Statistical Analysis
The unit of analysis was the hospital stay.Since patients who were more sick may potentially choose 1 hospital system over another, comparing outcomes in a traditional regression may produce biased results.9][40] In IPWRA models, we estimate Walvraven index from all recorded diagnosis codes. 44We indicated post-Choice Act period beginning in 2015, the first full year of implementation, because it reduced VA use.Median income was obtained for patients' zip code from US Census data.

Patient Measures in Outcomes Equation
In outcomes equations, we adjusted for factors potentially influencing outcomes that included patients' age, marital status, priority for VA care, nonelective admission, overall comorbidity score, specific medical comorbidities, mental health comorbidity, and area-level income.Models for mortality and readmission used a probit model, and models for LOS and log-transformed costs used a linear model.We estimated average treatment outcomes of VA hospitals as the difference between estimated probabilities and means for all observations assuming treatment in VA hospitals and all observations assuming treatment in non-VA hospitals along with 95% CIs.Standard errors were adjusted for each unique patient-hospital combination. 45 sensitivity analyses, we estimated in-hospital mortality because we had complete data for all states.We also conducted analysis limited to nonelective hospitalizations because treatment and outcome patterns may vary by admission type and analysis with only 1 observation per patient throughout the 6-year period.

JAMA Network Open | Health Policy
Outcomes of Veterans Treated in VA and Non-VA hospitals

Characteristics of Patients and Hospitals by System
The study sample included a total of 593 578 hospitalizations and 414 861 veterans with a mean (SD) age 75 (12) years, 405 602 males (98%), 73 155 hospitalizations of non-Hispanic Black individuals (12%), and 442 297 hospitalizations of non-Hispanic White individuals (75%) overall.The mean age was similar for younger veterans but higher for older veterans in non-VA hospitalizations compared with VA hospitalizations (Table 1).Most patients were male in all groups.Non-VA hospitalizations had higher mean comorbidity scores.VA hospitalizations were more likely to be for patients who were Black individuals or Hispanic individuals, not currently married, had a service-connected disability, and lived in urban areas and closer to a VA hospital than non-VA hospitalizations.Patients traveled farther when admitted to a VA hospital vs non-VA hospital.
VA hospitalizations were more likely to be nonelective and for HF and pneumonia compared with other study conditions than non-VA hospitalizations.Rates of medical comorbidities were generally lower among VA hospitals compared with non-VA hospitals (Table 1).

Average Treatment Outcomes of VA Hospitals
In models balancing covariates between patients in VA and non-VA hospitals, there were no significant treatment effects of VA hospitals on probability of 30-day mortality for most conditions (Table 3

JAMA Network Open | Health Policy
Outcomes of Veterans Treated in VA and Non-VA hospitals In sensitivity analyses with all states, VA hospitals had significantly higher probability of in-hospital mortality for pneumonia but significantly lower probability for stroke and no other differences (eTable 19 in Supplement 2).Including only nonelective admissions, results were similar to all hospitalizations (eTable 20 in Supplement 2).When analysis was limited to only 1 observation per patient, results were similar to all hospitalizations, except that mortality was higher in VA hospitals for AMI for patients younger than 65 years (eTable 21 in Supplement 2).

Discussion
To our knowledge, this is the first study to compare outcomes for veterans of all ages in VA and non-VA hospitals for 6 common conditions.After accounting for selection of patients into VA or non-VA hospitals, patients treated in VA hospitals had significantly lower probability of 30-day mortality than those in non-VA hospitals for HF among older patients and stroke for both younger and older patients.Patients treated for CABG, GI hemorrhage, HF, pneumonia, and stroke in VA hospitals had lower probability of readmission compared with patients in non-VA hospitals; however, differences for GI hemorrhage and HF were found only in younger In contrast, younger patients hospitalized for AMI in VA hospitals had higher probability of readmission than non-VA patients.Mean hospitalization costs were mostly higher, and mean LOS was longer in VA hospitals for the study conditions.Costs of AMI hospitalizations for younger patients were lower in VA hospitals than non-VA hospitals.
Our findings showing lower mortality in VA hospitals for 2 of the 6 conditions suggests that there was a mortality advantage associated with VA hospitals but not for all types of care.Recent studies of inpatient surgery and emergency department care also found associations between lower mortality and better quality in VA hospitals compared with non-VA hospitals. 5,46,47More research is

JAMA Network Open | Health Policy
Outcomes of Veterans Treated in VA and Non-VA hospitals needed to determine what aspects of VA care, such as postdischarge care, can improve mortality and whether there are differences for other clinical outcomes.
Our findings on mortality are similar to some previous findings but diverged from others.We did not find differences in mortality for CABG between VA and non-VA hospitals similar to another study. 17We found lower mortality for HF and stroke but not AMI in VA hospitals, while another study found lower mortality for AMI and HF in VA hospitals but did not include stroke in a study of adults aged 65 years and older. 13That study included veterans and nonveterans in an earlier period, which may explain different findings. 8 found lower readmissions in VA hospitals for CABG, GI hemorrhage, HF, pneumonia, and stroke but higher readmissions for AMI in younger patients.In contrast, Nuti et al 8  found lower VA costs over 28 days, so focusing on inpatient costs does not account for postdischarge costs that may be lower in the VA.
Our findings are especially relevant given that the Centers for Medicare & Medicaid Services now publicly reports the performance of VA hospitals in addition to non-VA hospitals on its Care Compare website.Veterans may be more likely to choose VA hospitals that perform comparatively better than other hospitals in their service area.

Limitations
This study has limitations.These data precede the MISSION Act of 2018, so our findings may not be generalizable to veterans currently accessing non-VA care.Our findings were based on hospitalizations from 47% of VA hospitals from diverse states, but they may not be generalizable to all VA hospitals.Our methods used many observed patient characteristics to account for patient selection, but there may have been unobserved factors which influenced patients' use of VA hospitals and outcomes.Undercoding of comorbidities was previously documented in the VA, 49,50 so differences in outcomes may have been underestimated.We did not distinguish between potentially avoidable readmissions and unavoidable or planned readmissions which may have led to overestimates of the observed readmission rates; however, planned readmissions only account for roughly 7% of all readmissions, so it is unlikely to materially affect our results. 51Non-VA hospitalization costs were estimated from cost-adjusted charges, which is less accurate than production costs, so cost differences may have been underestimated.Finally, we included hospitalizations for patients who were discharged against medical advice because these hospitalizations typically represent a small proportion (1%) of hospitalizations. 52

Conclusions
Expanding access to non-VA care may improve timeliness and reduce travel costs for many veterans; however, there are tradeoffs with higher mortality and readmissions in non-VA hospitals observed across age groups.As more veterans use care in the community paid for by the VA due to the MISSION Act, our findings suggest there may be reasons for concern

Patient Measures in Treatment Equation
geographic region (Northeast, South, Midwest, West), rural or urban location, area-level income (mean standardized), and post-Choice Act period.Race and ethnicity were included to adjust for sociodemographic factors.Comorbidity score was measured for each stay using the Elixhauser-van For descriptive purposes, hospital characteristics were measured in VA and non-VA hospitals, including number of staffed beds, bed occupancy rate, academic affiliation, and patient experience rating using percent of patients likely to recommend their hospital.Patient and hospital characteristics by VA and non-VA hospital and age group were compared in Pearson χ 2 and analysis of variance tests.
a Observations summarized here are hospitalizations.bP values reported for Pearson χ 2 tests for categorical variables and analysis of variance tests for continuous variables.cIncludes Alaska Native, Asian American, Native Hawaiian, and Pacific Islander.younger

Table 2 .
Unweighted Acute Hospitalization Outcomes in VA and Non-VA Hospitals, 2012-2017 Abbreviations: AMI, acute myocardial infarction; CABG, coronary artery bypass surgery; GI, gastrointestinal; HF, heart failure; LOS, length of stay; VA, Veterans Affairs.aNos.varied by outcome and reported only for LOS.b P values reported for analysis of variance tests.

Table 3 .
Average Treatment Outcomes of VA Hospitals Compared With Non-VA Hospitals for 30-Day Mortality and 30-Day Readmission Average treatment outcomes of difference in predicted probability for VA hospitals vs non-VA hospitals estimated from inverse probability weighting regression adjustment models with probit models for treatment and outcomes.
a b Nos.varied by outcome.

Table 4 .
48cumented higher readmissions in VA hospitals for AMI, HF, and pneumonia in older patients prior to access expansions.VA hospitals may be more successful in reducing readmissions due to an integrated delivery system, implementation of the patient-centered medical home, and an electronic medical record system.It is unclear why younger patients who were hospitalized for AMI were more likely to be readmitted in VA hospitals even though patients often travel longer distances to VA hospitals, potentially affecting their outcomes.Both VA and non-VA hospitals have recently emphasized reducing readmissions through the use of performance measures in the VA and payment policies in the private sector and Medicare.Mean LOS was longer and costs were higher in VA hospitalizations for most conditions compared with non-VA hospitalizations.Medicare and private insurance payment policies (eg, bundled payment programs) have focused on efficiency and may have influenced hospitals to discharge patients sooner while VA hospitals were unaffected by such policies.VA hospitals may keep patients longer to ensure they are stable before discharging them.Higher VA hospitalization costs may be partly explained by longer LOS.There may be other differences due to staffing and overhead between VA and non-VA hospitals leading to greater resource use.A study48about ED care Average Treatment Outcomes of VA Hospitals Compared With Non-VA Hospitals for Length of Stay and Costs Abbreviations: AMI, acute myocardial infarction; CABG, coronary artery bypass surgery; GI, gastrointestinal; HF, heart failure; LOS, length of stay; VA, Veterans Affairs.aAverage treatment outcomes of difference in predicted probability for VA hospitals vs non-VA hospitals estimated from inverse probability weighting regression adjustment models with probit models for treatment and linear models for outcomes.b Nos.varied by outcome.
. Veterans could experience worse outcomes for some types of care without well-developed community care networks based on quality standards and sufficient care coordination between VA and non-VA clinicians.In an era of greater choice, veterans' often benefit by choosing VA care.AMI Patients Less Than 65 Years of Age: Coefficients From IPWRA Models eTable 2. AMI Patients 65 Years and Older: Coefficients From IPWRA Models eTable 3. CABG Patients Less Than 65 Years of Age: Coefficients From IPWRA Models eTable 4. CABG Patients 65 Years and Older: Coefficients From IPWRA Models eTable 5. GI Hemorrhage Patients Less Than 65 Years of Age: Coefficients From IPWRA Models eTable 6. GI Hemorrhage Patients 65 Years and Older: Coefficients From IPWRA Models eTable 7. HF Patients Less Than 65 Years of Age: Coefficients From IPWRA Models eTable 8. HF Patients 65 Years and Older: Coefficients From IPWRA Models eTable 9. Pneumonia Patients Less Than 65 Years of Age: Coefficients From IPWRA Models eTable 10.Pneumonia Patients 65 Years and Older: Coefficients From IPWRA Models eTable 11.Stroke Patients Less Than 65 Years of Age: Coefficients From IPWRA Models eTable 12. Stroke Patients 65 Years and Older: Coefficients From IPWRA Models eTable 13.Balance of Covariates in Mortality Model for AMI eTable 14.Balance of Covariates in Mortality Model for CABG eTable 15.Balance of Covariates in Mortality Model for GI Hemorrhage eTable 16.Balance of Covariates in Mortality Model for HF eTable 17.Balance of Covariates in Mortality Model for Pneumonia eTable 18. Balance of Covariates in Mortality Model for Stroke eTable 19.In-Hospital Mortality: Average Treatment Effects (ATE) and Predicted Probability/Means for VA Hospitals From IPWRA Models eTable 20.Results From IPWRA Models Including Only Nonelective Hospitalizations eTable 21.Results From IPWRA Models With One Observation per Patient and Standard Errors Adjusted for Clustering Within Hospital eTable 22. ICD-9 and ICD-10 Codes for Admitting Condition and Comorbidities