|Home | About | Journals | Submit | Contact Us | Français|
To address potential equity concerns about the U.S. Department of Veterans Affairs’ (VA) process for adjudicating military service-related disability claims.
Participants were a nationally representative sample of 20,048 veterans completing the 2001 National Survey of Veterans. Socio-demographic, access, and illness correlates of both the award and rate of general disability benefits awarded by the VA were examined using an established theoretical framework.
Socio-demographic, access, and illness variables were associated with both the award (“yes/no”) and rate of benefits (0-100%) awarded, with combat exposure, unemployment, and physical impairment accounting for the strongest model effects.
Veterans’ needs were not overshadowed by factors related to demographic background or access (e.g., race, gender, insurance), reducing concerns about disparities in general VA disability disbursements. These data are timely as disability claims/payments will likely increase dramatically in the near future due to current conflicts in the Middle East.
The U.S. Department of Veterans Affairs (VA) provides monetary compensation to veterans with military service-related disabilities (i.e., injuries and/or diseases that occurred and/or were aggravated during the course of military duty). Through its disability and pension programs, the VA paid approximately $34.5 billion in disability benefits to veterans and their survivors in fiscal year 2006, and claims for such disabilities have increased by approximately 50% from 2003 to 2006, totaling about 378,000 [1,2]. Despite the increase in payments allocated during this time period, there are concerns about the VA’s disability adjudication process, including the long wait for claims decisions (e.g., average of 657 days for appeals resolution), backlogs of claims, and the potential for inconsistencies or inequities in award decisions [1,2]. A recent report of the VA’s Inspector General examined variations in disability benefits allocation by states and/or regions and found inconsistencies in the process (e.g., variability in rating decisions) . To date, there has been very little systematic investigation of factors or variables that influence a veteran’s likelihood of being awarded disability-related benefits or in the rate at which benefits are awarded.
Only a few studies have examined correlates or predictors of disability benefits. These studies focused exclusively on the clinical syndrome of posttraumatic stress disorder (PTSD) and analyses were limited to a circumscribed set of predictor variables (i.e., race, gender, region, time). Nonetheless, these studies found that female veterans were less likely to receive VA disability payments for PTSD than male veterans, although this disparity ceased to exist after controlling for level of combat exposure [3,4]. With regard to racial differences, analyses of linked administrative and survey data found that African American male veterans were less likely to receive disability benefits for PTSD than Caucasian veterans . However, smaller studies of clinical populations did not find racial differences in disability benefits among treatment-seekers [6,7]. To date, there are no extant data examining disability benefits from a more general perspective (i.e., encompassing award of benefits for more than one disorder or condition), nor are there data that have examined a wide range of potential correlates and the significance of these variables in predicting adjudication of military service-related disability claims. Furthermore, none of these previous studies used nationally representative samples of veterans, limiting the generalizability of findings to the larger veteran population. The present study attempted to overcome all three of these important limitations.
Using data from the 2001 U.S. National Survey of Veterans (NSV)  we examined correlates of both the award (“yes/no”) and rate of disability benefits awarded (on a 0-100% scale) using the Behavioral Model of Healthcare Use  as a conceptual framework to guide analyses. This model is a well-validated theoretical framework for understanding determinants of healthcare use, and can also be useful in understanding determinants of disability benefit awards, since the award of such benefits is often linked to health problems and the use of health services. There are three components to the Behavioral Model: 1) “predisposing” variables, encompassing historical and socio-demographic characteristics, such as race, gender and education; 2) “enabling” variables, involving access-related factors, such as health insurance, employment, and geographical residence; and 3) “need” variables related to physical or mental health illness or impairment, such as patient-rated health functioning or an identified disabling condition. Using this model, researchers have identified a number of meaningful predisposing and enabling variables related to health service use in national studies of veterans, such as older age, male gender, minority racial/ethnic status, less education, single status, war zone and combat exposure, lower income, unemployment, and lack of health insurance [10-13]. Need variables predictive of health service use have included health problems, depression, and anxiety; and they have generally been the most robust predictors of service use [11-16].
In the present study we specifically examined the extent to which medical and mental health impairments (i.e., need variables) were associated with award (“yes/no”) and rate of benefits awarded (0-100%), controlling for predisposing and enabling variables. We hypothesized that need variables would be stronger predictors of disability benefits awarded than predisposing and enabling access variables. However, we also examined which, if any, predisposing and enabling variables were predictive of disability benefits awarded. This study is relevant and timely because there are increasing numbers of veterans returning from war-zone deployments in Iraq and Afghanistan, and many have significant medical and mental health difficulties [17-19]. Research is needed to identify gaps and/or disparities in the award and rate of disability benefits awarded to veterans and to address potential equity concerns and barriers to appropriate services among contemporary veterans.
The sample included 20,048 non-institutionalized veterans who completed the 2001 NSV. The majority of participants were men (n = 18,767, 93.61%), Caucasian (n = 16,508, 82.98%), and married (n = 14,771, 73.83%). The average age of the sample was 59.28 years (SD = 14.93). About three-fifths of the sample had at least some post high school training or college education (n = 11,798, 58.99%), with 5,699 (28.49%) having a high school diploma/GED with no further education. Approximately half were employed (n = 9,773, 48.82%), with the remaining participants either retired (n = 6,576, 32.80%) or disabled (n = 2,467, 12.31%). Most participants had some form of health insurance coverage (n = 17,739, 88.94%), and were urban residents (n = 15,459, 78.37%).
Although not mutually exclusive, more than one-third of the sample served during the Vietnam War-era (n = 7,934, 39.58%), followed by the era between the Korean and Vietnam wars (n = 5,264, 26.26%), World War II (n = 4,565, 22.77%), and between the Vietnam War and 1991 Gulf War (n = 4,070, 20.30%). The sample was primarily composed of veterans from the Army (n = 10,501, 52.54%) and Navy (n = 4,589, 22.96%). Nearly one-half of participants (n = 9,167, 46.32%) endorsed having served in combat or a war zone.
Using 300 trained interviewers, the 2001 NSV  implemented computer-assisted telephone interviewing. Survey collection included a) random digit dial (RDD) methodology to identify the majority of veterans, and b) lists of patients enrolled in VA healthcare or receiving VA compensation or pensions, to identify remaining cases. Participation was voluntary, and responses were confidential. Only participants who consented to the interview after the procedures were explained were evaluated, and we obtained institutional review board approval for using these data. The survey resulted in 12,956 cases from the RDD sample and 7,092 from the patient lists. Undercoverage based on unlisted telephone numbers was nominal, corrected by “raking” procedures using U.S. Census estimates . Survey data were weighted based on the probability of selection, non-response and household size, thus ensuring that survey responses would generalize to the larger non-institutionalized veteran population. The survey’s response rate ranged from 62.8% (patient list) to 76.4% (RDD), which is an excellent response rate for epidemiological telephone-based surveys and is well above the 40% minimum requirement of other systematic population-based state and national telephone surveys (e.g., the Behavioral Surveillance Branch of the CDC). The final sample was demographically representative of the known veteran population collected in the 2000 U.S. Census. Refer to the final report on design and methodology for additional details on methodology and weighting strategies .
Numerous socio-demographic and background questions were queried. Those of relevance to the present paper included gender, age, race/ethnicity, educational level, marital status, health insurance possession, employment status, and rural-urban residence status.
Questions about participants’ military histories were gathered. These included inquiries about the branch of military and war era in which respondents served. Additionally, participants were asked if they had served in a war zone or combat.
Veterans were asked 1) whether they had been awarded a service-connected disability rating (i.e., yes/no) and, if they answered “yes,” 2) the rate of service-connection awarded on a 0% to 100% scale, with higher rates indicating more benefits coverage.
The Health Survey Short Form-12 version 1 (SF-12)  was administered, assessing physical (PCS) and mental health (MCS) functional impairment (scores ranging from 0 to 100, M = 50, SD = 10, with lower scores indicating more impairment). The validity and reliability of the PCS and MCS are well established [20,21]. MCS values ranged from 14.01 to 69.13 (M = 46.19, SD = 6.69) in the sample. PCS values ranged from 17.60 to 65.38 (M = 45.06, SD = 7.97).
Predisposing predictor variables included gender, age, race (Caucasian or minority), education level (some college education or less), marital status (married or unmarried), and combat exposure (endorsed or unendorsed). Enabling variables included health insurance possession (present or absent), employment status (employed or unemployed), and urban-rural residence. Need variables included the SF-12 MCS and PCS (See Table 1 for precise categorical variable codings). These variables were defined in a manner consistent with previous research.
Outcome variables included the two disability variables described above: Disability benefits awarded as a dichotomous outcome (“yes/no”), and rate of disability benefits awarded as a continuous outcome from 0 to 100%.
STATA Version 9.0 (Stata Statistical Software Release 9.0. 2005, Stata Corporation: College Station, TX) was used for statistical analyses to control for the complex survey design of the 2001 NSV and to provide estimates that would generalize to the U.S. veteran population. Infrequent missing continuous variable responses were replaced with series means (age and SF-12 items), generally appropriate in large samples (here, roughly 20,000 subjects) . Approximately 1% of cases were replaced, mostly from the SF-12. Missing categorical data resulted in the list-wise exclusion of only 5% of participants in multivariate analyses. Analyses were two-tailed with significance set at p-value of 0.05.
We used logistic regression to examine multivariate associations between predictor variables and award of benefits (“yes/no”). Award of benefits was entered as the dependent variable; predisposing and enabling variables were entered as predictor variables in the first step of the models, and need variables were entered in the second step.
The relationship between predictor variables and rate of benefits awarded (0-100%) was examined using linear regression (this outcome variable was normally distributed). Using a similar modeling approach, rate of benefits awarded (0-100%) was entered as the dependent variable and predisposing, enabling, and need variables were entered sequentially as dependent variables. Entry of variables was based on clinical relevance and previous work on the most robust determinants of service use [23,24].
About one-third (n = 6,652, 33.57%) of veterans were rated with a service-connected health-related disability from the VA, with a median disability rating of 30. The three highest frequency counts were for 10% (n=1,689), 30% (n=981), and 100% (n=768) disability ratings, respectively.
Multivariate logistic regression analysis revealed that the predisposing/enabling model was significantly related to an award of disability benefits, χ2(9, N = 18937) = 666.45, p < .001, accounting for a modest amount of variance (Nagelkerke’s R2 = .06). The need model was also significantly associated with an award of disability benefits, χ2(11, N = 18937) = 1198.59, p < .001, and incrementally added a modest amount of variance over the first model, LR χ2change(2, N = 18937) = 609.18, p < .001 (R2change = .06). In the final model (R2 = .12), an award of disability benefits was associated with younger age, minority race, having a college education, combat exposure, unemployment (predisposing/enabling variables), and poorer mental and physical health functioning (need variables), with the strongest effects for combat and physical health impairment (Table 1).
Multivariate linear regression analyses revealed a pattern similar to the results found for the dichotomous award of disability benefits (“yes/no”) analyses. The predisposing/enabling model was significantly related to the rate of benefits awarded, F(9, 5980) = 41.30, p < .001, accounting for a moderate amount of variance (R2 = .11). The need model was also significantly associated with rate of disability benefits awarded, F(11, 5978) = 48.81, p < .001, adding a modest, but significant amount of variance over the first model, Fchange(2, 9) = 63.45, p < .001 (R2change = .05). The final model (R2 =.16) yielded several significant associations with rate of benefits awarded. Significant variables included female gender, younger age, having a college education, combat exposure, unemployment (predisposing/enabling variables), and poorer physical health functioning (need variable), with the strongest effects for unemployment and physical health impairment (Table 2).
As additional exploratory analyses, we also conducted the above analyses (award yes/no) and rate (0-100%) of disability benefits using only those veterans who endorsed combat (n=9,167); and the pattern of results remained largely unaltered. All statistically significant relationships remained significant, and two other variables emerged as significant predictors. Marital status became a significant predictor of award (“yes/no”) of disability benefits (z=2.34, p=.02), and poorer mental health became a significant predictor of rate (0-100%) of disability benefits (t=−2.06, p=.04); albeit each accounted for a modest amount of variance in outcomes. Physical health impairment remained the most robust predictor of award (yes/no) of disability benefits (z=−16.42, p=<.001) and unemployment and physical health impairment remained the most significant predictors of rate of benefits awarded (0-100%; t=−10.13, p=<.001 and t=−8.68, p=<.001.
In this study we addressed equity concerns by identifying predictors of general VA disability benefit awards in a national sample of 20,048 veterans, using the Behavioral Model of Healthcare Use as a framework to guide our analyses. Results revealed that a number of socio-demographic and need variables were associated with both award of benefits (“yes/no”) and rate of benefits awarded (0-100%). Specifically, younger age, minority race, having a college education, combat exposure, being unemployed, and poorer mental and physical health were predictive of being awarded disability benefits (“yes/no”), with need models contributing statistically significant additive effects beyond predisposing and enabling models. Although the predisposing and enabling model accounted for a comparable amount of variance in award of benefits as the need model, this was largely due to the inclusion of combat exposure, which was the most robust predictor in both models. Reassuringly, combat exposure and mental and physical impairment were the most robust predictors of disability benefits awards, and there was little evidence of unfair race or gender disparities in the VA’s adjudication of disability claims in this sample. In fact, contrary to what might be anticipated, younger age, female gender, and minority status were all associated with a greater likelihood of VA disability benefits awards.
With regard to rate of benefits awarded (0-100%), significant predictors included female gender, younger age, having a college degree, combat exposure, unemployment, and poorer physical health. The strongest predictors were unemployment and physical health, with employment status yielding a slightly higher association than physical health. Again, these data suggest that appropriate need variables, rather than predisposing variables (e.g., race, gender, marital status) or most access variables (e.g., health insurance, rural/urban residence), are the most robust predictors of disability benefit disbursements. Further, one could logically argue that unemployment, the one significant access variable to emerge in these analyses, is largely conflated with need. That is, there is a strong relationship between unemployment and medical and/or mental health functioning [25, 26].
This study has a number of limitations worth noting. First, these data are cross-sectional and all of the limitations regarding this type of study design apply to the current study. Second, we did not have access to information on the number of veterans who applied for and were subsequently denied benefits, and such data could yield different associations than those found in the present study. Third, there may be several sources of sampling bias. All data were based on veterans’ self-report, which may introduce unknown sources of error. The survey’s response rate ranged from 63% to 76%, which also leaves room for response bias error but is generally considered a strong response rate for epidemiological telephone-based surveys. Nonetheless, approximately 37% and 24% of those contacted did not respond, and we cannot speak to the characteristics of these individuals. Additionally, although the NSV is generally demographically representative of U.S. Census statistics for veterans [27, 28], 78% of veterans in the NSV were urban residents. This is discrepant with previous census data indicating that the highest concentrations of veterans are in rural and non-metropolitan counties .
Quite possibly, those veterans who are not included in the NSV represent those who are most ill and in need of services or disability benefits. That is, the most vulnerable or at-risk veterans (i.e., the geographically isolated, homeless, antisocial, and incarcerated) are underrepresented in the NSV, and inclusion of these individuals may have altered the pattern of our study findings and yielded less optimistic findings. Additionally, although these findings provide some insight into the decision-making process for service-connection through the VA system at the national level, results may not generalize to U.S. veterans returning from more recent deployments such as Afghanistan and Iraq. Indeed, our sample includes a higher percentage of WWII and Vietnam era veterans than is reflected in more recent U.S. Census data . However, these data are relevant and timely because recent administrative initiatives have focused on identifying disability claims process problems and finding opportunities for improvement [1,2,29], yet there is little written on the topic. Finally, both a strength and a limitation of the current study is that disability status is not linked to a specific disorder. These data are reassuring with regard to the allocation of benefits across different types of conditions. However, inequities within specific conditions (e.g., PTSD) may persist and warrant further investigation.
In the current analyses, physical health functioning was one of the strongest predictors of disability benefits, and combat exposure and employment status were strong predictors in the separate models. All of these variables are potentially relevant and appropriate for making disability award decisions. Similar analyses using the subset of veterans who endorsed combat exposure yielded a similar pattern of results, further suggesting that need variables are the most salient predictors of disability benefits. These findings suggest that veterans’ needs are not overshadowed by factors related to predisposing or access variables, thus reducing concern about potentially unfair disparities among racial minorities and females in general VA disability disbursements. Further, this study is well-timed as disability claims and payments will likely increase dramatically in the next few years due to current conflicts in Iraq and Afghanistan, and there is a need to systematically examine factors that may adversely and/or favorably affect the VA’s disability claims process. Elsewhere  we have expressed concern that VA psychiatric disability policies for PTSD are outdated and problematic. However, these data are reassuring with regard to the equitable adjudication of military service-related disability claims.
This work was partially supported by grant CD-207015 from Veterans Affairs Health Service Research and Development and grant MH074468 from the National Institute of Mental Health. All views and opinions expressed herein are those of the authors and do not necessarily reflect those of our respective institutions or the Department of Veterans Affairs.
Authorship Statement: All authors participated in the conceptual development of the opinions and conclusions expressed herein, including review of literature, writing, data analyses, and editing, and all have seen and approved the final version. There is no one else who fulfills the criteria but has not been included as an author.
Conflict of Interest Statement: There are no disclosures or conflicts of interest to report.