Association of Disability Compensation With Mortality and Hospitalizations Among Vietnam-Era Veterans With Diabetes
Introduction
People with lower incomes have worse health and die sooner than those with higher incomes.1–3 The association between income and health is evident throughout the life course, across the gradient of incomes, and in many countries, including those with universal health insurance and robust social welfare programs.1–10
Although income and health are associated, outside of randomized studies conducted in low-income countries, it remains unclear whether policy interventions that increase income improve health outcomes.11–21 This evidence gap has high-stakes consequences. In 2019, approximately 8% of federal government spending ($361 billion) provided income support, tax credits, or other benefits (besides health insurance) to low-income and disabled individuals.22 Limited understanding about the health consequences of income assistance programs prevents policy makers from understanding their full value to society.23 Moreover, if income assistance averts serious and costly health events, such as acute hospitalizations, then the costs of income support programs may be partially offset by less health care spending.
The Department of Veterans Affairs (VA) oversees the second largest source of disability-related income assistance in the US.24 In 2020, more than 5 million veterans received a total of $91 billion in compensation for disabling conditions related to military service.25 These payments align with a core principle that predates the nation’s founding: that the government should care for and compensate veterans who have sustained injuries or developed medical conditions during military service. The payments are large in magnitude (reaching an annual maximum of $37 757 in 2021 for a veteran without dependents), not subject to federal or state income tax, typically made in perpetuity, and apply to individuals with lower socioeconomic status and worse health than the general population.26–31 Therefore, while prior research has largely focused on the influence of disability compensation on veterans’ employment decisions,24,32,33 disability compensation may also have important benefits for health.
This quasiexperimental study evaluated a change in VA disability policy in July 2001 that qualified some Vietnam-era veterans with diabetes for disability compensation. Specifically, we examined the association between eligibility for disability compensation with mortality and hospitalizations among veterans with diabetes. Prior studies have established large socioeconomic gradients in mortality and hospitalizations among persons with diabetes, prompting growing interest in interventions to address social determinants of health for this population.34–36
Methods
Policy Context and Study Design
On July 1, 2001, the VA added diabetes to the list of conditions presumptively connected to military service for all Vietnam-era veterans who served with “boots on the ground” (BOG) in Cambodia, Laos, or Vietnam during the Vietnam War. The change in policy was motivated by an Institute of Medicine report that found a “limited/suggestive” association of Agent Orange, an herbicide used by the US government during the Vietnam War, with diabetes.37 As a result of the VA’s decision, a diagnosis of diabetes would qualify previously nondisabled BOG veterans for disability compensation and lower copayments for outpatient and hospital care. Vietnam-era veterans who were “not on ground” (NOG) remained ineligible for disability compensation due to diabetes.
We conducted a difference-in-differences study that examined changes in outcomes for BOG and NOG Vietnam-era veterans following the VA’s July 2001 policy decision. We used the period from October 1, 1996, to September 30, 1998, to establish the presence of diabetes, from January 1, 1999 (October 1, 1999 for hospitalizations and outpatient visits), to June 30, 2001, to ascertain trends in prepolicy outcomes, and from July 1, 2001, to December 31, 2018, to measure postpolicy changes (eFigure 1 in the Supplement). Requiring an established diagnosis of diabetes prior to October 1, 1998, avoided the potential bias that the VA policy may have led some BOG veterans to seek a diagnosis of diabetes. To understand how differences between BOG and NOG veterans evolved after the policy change, we divided the postpolicy period into early, middle, and later periods of approximately 6 years each (period 1: July 1, 2001, to December 31, 2006; period 2: January 1, 2007, to December 31, 2012; period 3: January 1, 2013, to December 31, 2018). Data analysis was performed from October 1, 2020, to December 1, 2021. The VA Central IRB approved the study and waived the need for informed consent because the research involved no more than minimal risk to study participants. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Sources of Data/Study Population
We merged data from the Veterans Health Administration, Veterans Benefits Association, the Centers for Medicare & Medicaid Services, and Vietnam-era military deployment records from the Department of Defense. Further details are available in eMethods in the Supplement.
The primary study population included male veterans (99.8% of BOG veterans in the study cohort were male) with a diagnosis of diabetes38 in the Veterans Health Administration prior to October 1, 1998, and military service during the Vietnam-era (February 28, 1961, to May 7, 1975) (see eFigure 2 in the Supplement for flowchart). We excluded veterans with a history of disability compensation payments prior to 1999 and/or who served in the military for 20 or more years, the qualifying period for a military pension. Until January 2004, the Department of Defense deducted any VA disability compensation from the military pension.
Outcomes
The primary outcomes were all-cause mortality and acute care hospitalizations (eTable 1 in the Supplement). Secondary outcomes included receipt of disability compensation, amount of disability compensation (in 2018 dollars), outpatient physician visits, and acute hospital days. The receipt and amount of disability compensation were recorded in monthly snapshots available annually each December. Because the use of VA data alone would not capture services financed by other payers, we conducted analyses of hospitalizations and outpatient visits after restricting to veterans who were enrolled in traditional (fee-for-service) Medicare in 1999 and only considered utilization that occurred in months when veterans were concurrently enrolled in traditional Medicare.39–41
Exposure and Covariates
The primary exposures were BOG status, indicators for 3 postpolicy periods, and the interaction of BOG status with each postpolicy period. Demographic covariates included indicators for year of birth, race and ethnicity (non-Hispanic Black, Hispanic, non-Hispanic White, and other), urban residence, and the zip code–level proportion of persons with college attendance and with income below the federal poverty limit as reported in the 2000 US Census. Race and ethnicity were derived from VA administrative data. Veterans who were American Indian/Alaska Native, Asian, Native Hawaiian/Pacific Islander, other, or who were designated as multiple races were classified as other owing to small sample sizes. Clinical covariates included baseline blood pressure and glycated hemoglobin level (the values closest to January 1, 1999); the number of Elixhauser comorbidities identified between October 1, 1996, and September 30, 1998; and receipt of an angiotensin-converting enzyme inhibitor or angiotensin receptor blocker, sulfonylurea, statin, or insulin prior to July 1, 2001. For continuous covariates, we converted values to quintiles with a category for missing. The proportion of missing covariates is reported in eTable 2 in the Supplement. We did not include postpolicy covariates because they may lie on the causal pathway between increased income and improved health.
Statistical Analysis
We constructed generalized linear models of the outcomes after including the independent variables and covariates as described above. To account for differences in age between BOG and NOG veterans, the primary model included fixed effects for birth year (eg, 1950) and fully interacted fixed effects for each combination of birth year and postpolicy era. The main coefficients of interest, the interactions between BOG status and postpolicy period, therefore reflected differences in outcome means for BOG vs NOG veterans in a specific postpolicy era, relative to the difference between BOG vs NOG means in the prepolicy era, among veterans born in the same year and after controlling for other covariates. The use of a concurrent control group of NOG veterans also accounts for the large secular decline in hospital admissions over the study period, particularly among older adults.42 For mortality, outpatient visits, and hospitalizations, the unit of analysis was the person-quarter. Analyses of receipt of disability compensation and monthly compensation payment amount (in 2018 dollars) were conducted at the patient-year. We clustered standard errors at the person-level using generalized estimating equations. The eMethods in the Supplement provide the regression specification.
We conducted stratified analyses by race, zip code–level poverty, and number of comorbid conditions. Exploratory analyses also focused on deaths due to cardiovascular diseases and hospitalization for cardiovascular disease, diabetes, and heart failure, the 3 most common reasons for hospitalization in the cohort. We also separately modeled hospitalizations financed by VA and by Medicare and modeled mortality for the Medicare-enrolled cohort.
Analyses of prepolicy trends are described in the eMethods in the Supplement. We examined changes in the composition of BOG and NOG veterans by modeling the baseline covariates as outcomes in difference-in-differences models.43 In another sensitivity analysis, we included interactions of each covariate with postpolicy time period. To examine how enrollment in managed care may have affected our findings, we estimated the annual number of months of managed care enrollment and examined the use of hospital care among switchers to managed care vs those who remained in traditional Medicare. To explore the sensitivity of our results to missing covariates, we fitted a model that excluded all covariates other than birth year. Finally, we conducted a falsification test of BOG and NOG veterans who were already receiving maximal disability compensation payments prior to 1999. Statistical analyses were performed using SAS statistical software (Enterprise Guide 7.1; SAS Institute, Inc).
Results
The study population included 70 471 Vietnam-era veterans, of which 14 247 were BOG veterans (mean [SD] age, 51.2 [3.8] years as of December 31, 1998; 25.7% were Black; 3.3% were Hispanic; 63.6% were White; and 6.9% were of other race), and 56 224 were NOG veterans (mean [SD] age, 54.2 [6.3] years; 21.7% were Black; 2.1% were Hispanic; 67.1% were White; and 8.2% were of other race) (Table 1). Absolute differences between BOG and NOG veterans in other characteristics were minimal. eTable 3 in the Supplement provides the prevalences of Elixhauser comorbidities. eTable 4 in the Supplement describes the 3623 BOG veterans and 19 174 NOG veterans enrolled in traditional Medicare in 1999.