Informal Caregiver Support Policies Change Use of Vocational Assistance Services for Individuals With Disabilities

Abstract

Support policies for caregivers improves care-recipient access to care and effects may generalize to nonhealth services. Using administrative data from the U.S. Department of Veterans Affairs (VA) for veterans <55 years, we assessed the association between enrollment in a VA caregiver support program and veteran use of vocational assistance services: the post-9/11 GI Bill, VA vocational rehabilitation and employment (VR&E), and supported employment. We applied instrumental variables to Cox proportional hazards models. Caregiver enrollment in the program increased veteran supported employment use (hazard ratio = 1.35, 95% confidence interval [1.14, 1.53]), decreased VR&E use (hazard ratio = 0.84, 95% confidence interval [0.76, 0.92]), and had no effect on the post-9/11 GI Bill. Caregiver support policies could increase access to some vocational assistance for individuals with disabilities, particularly supported employment, which is integrated into health care. Limited coordination between health and employment sectors and misaligned incentives may have inhibited effects for the post-9/11 GI Bill and VR&E.

Introduction

Due to advances in battlefield medicine, many veterans now survive combat-related injuries that were fatal in prior conflicts (Cifu et al., 2010). Thousands of veterans deployed during Operations Enduring Freedom, Iraqi Freedom, and New Dawn have returned home with disabling conditions that affect their ability to readjust to civilian life. An estimated 23% of these veterans have posttraumatic stress disorder (PTSD; Fulton et al., 2015) and 19% have a traumatic brain injury (TBI; Tanielian & Jaycox, 2008). These conditions are associated with affective, cognitive, and physical symptoms that reduce function (Meterko et al., 2012) and lead to difficulty finding/maintaining employment or continuing education and declines in health and quality of life (Kukla et al., 2012).

The Department of Veterans Affairs (VA) offers a range of medical, vocational, and education assistance services to eligible veterans to address their cross-cutting needs. However, these services reside within two distinct VA bureaus (i.e., Veterans Health Administration [VHA] and Veterans Benefits Administration [VBA]), which may lead to service fragmentation, poor service alignment with veteran needs, and variability in access to services across VA medical centers.

Individuals with disabilities may receive help with activities of daily living from informal caregivers, who are generally unpaid family members or friends. Informal caregivers often help individuals with disabilities access and coordinate medical care (Ramchand et al., 2014), and increasing caregiver skills could support use of vocational services (e.g., vocational rehabilitation and education assistance) by the individual with disabilities (Shepherd-Banigan et al., 2020). Therefore, policies and programs that emphasize the role of caregivers to help care-recipients access medical and vocational services could alleviate negative health, social, and economic outcomes.

In 2010, Congress established the nationwide VA Program of Comprehensive Assistance for Family Caregivers (PCAFC; P.L. 111–163) to support caregivers through education, training, and financial assistance. This legislation represented the most sweeping policy support for informal caregivers seen in the United States to date. Between May 2011 and May 2019, the program enrolled over 40,000 veteran families (program administrative records, May 2019). PCAFC support has led to higher use of primary, specialty, and mental health care by veterans (Van Houtven et al., 2017), but it is unclear if PCAFC support improves veterans’ access to vocational services. The purpose of this study is to examine the effect of VA’s caregiver support program on veteran use of three VA-provided vocational services: supported employment, vocational rehabilitation and employment (VR&E), and the post-9/11 GI Bill educational assistance.

We situate the hypothesized relationship between informal caregiver support and use of vocational and education services in intrahousehold welfare theory. This theory suggests that households are units in which individuals make joint decisions about how to generate and use household resources (time, money); as such, the well-being of each individual within that unit becomes an important factor for the overall well-being of the household. This theory has been used to explain how couples make decisions about division of labor and household spending (Becker, 1981). The allocation of tasks across household members in household production will be based on their comparative advantage. Any change in comparative advantage will affect household production decisions (e.g., an injury to the veteran means there are work reductions for the veteran and increases in caregiving/reductions in work for the spouse). For this study, we assume that return-to-work generates direct benefits in well-being for veterans and likely creates positive spillovers to informal caregivers. If the caregiver’s well-being depends on the veteran’s well-being—which is likely true for caregivers who are spouses/partners and share the same household—the increased household earnings from the veteran and other direct benefits for the veteran (health, human capital, quality of life) will positively affect the caregiver. Even for caregivers who are parents, benefits for the veteran will positively affect the parents’ well-being because they value their adult child’s good health (Becker, 1992; Van Houtven & Norton, 2004). Additional positive effects could accrue if veteran employment enables caregivers to return-to-work or engage in other non-caregiving activities that enhance their welfare, such as schooling, childcare, volunteer work, or leisure. Vocational rehabilitation increases return-to-work and should have a positive impact on household well-being.

Background

VA offers high-quality, evidenced-based medical, psychological, and vocational, and education assistance services that can address veterans’ cross-cutting needs. VHA’s Compensated Work Therapy Program offers numerous clinical vocational rehabilitation services. We focus on supported employment because it is the gold standard program, and in the past 10 years VA has adapted and expanded this program (Resnick & Rosenheck, 2007, 2008; Wewiorski et al., 2018). Supported employment is an evidence-based vocational rehabilitation program offered to individuals who experience the most significant employment barriers due to disabling mental illness. This program helps these individuals find and maintain competitive, community-based employment (Bond & Drake, 2014; Resnick & Rosenheck, 2007). Health care providers refer individuals to this program. Supported employment counselors are integrated into the veteran’s health care team, which helps individuals with disabilities manage symptoms and engage in successful workplace strategies (Bond et al., 2001). Supported employment is associated with competitive work attainment (Twamley et al., 2013), longer job tenure (Twamley et al., 2013), higher income (Crowther et al., 2001; Davis et al., 2012), and increased days of paid employment (Davis et al., 2012; Twamley et al., 2013). The VBA’s post-9/11 GI Bill was established in 2009 to provide education support, including tuition and housing, to qualifying veterans who wish to pursue higher education. This program is offered to veterans who served at least 90 days of active duty since September 11, 2001, and received an honorable discharge. Research indicates that the post-9/11 GI Bill drives veterans’ decisions to pursue higher education (Barr, 2015; Steele et al., 2011; Zhang, 2017). Furthermore, studies about the (pre-9/11) GI Bill (Barr, 2015; Steele et al., 2011) suggest that, among Vietnam-era veterans, VA education assistance was related to an increase in years of schooling and income (Angrist & Chen, 2011). The VBA also administers the VR&E program, which offers education and career counseling, educational assistance, and job placement services to veterans with a 10% or greater service-connected disability rating and a documented handicap (U.S. Department of Veterans Affairs, 2019). Evidence about the impact of participating in this program is limited.

VA is the only health system in the United States that has established a national program of comprehensive support for informal caregivers. Caregivers of veterans enrolled in PCAFC receive enhanced support from VA, including mandatory caregiver training, ongoing monitoring of the veteran and caregiver by a clinician through in-home and phone-based assessments, and financial stipends that can alleviate short-term financial burden and allow caregivers to be present to support their loved one. Stipends are tax-free and range from $600 to $2300 per month depending on the assessed intensity of caregiving required. We hypothesize that PCAFC support increases veteran use of vocational services by improving the caregiver’s ability to navigate VHA and VBA systems and providing the time flexibility to help the veteran use these services.

New Contribution

Evidence about the effects of caregiver support policies has focused on health and health service outcomes for the caregiver and care recipient. However, this study examines how those policy effects may generalize to vocational rehabilitation services. This question is important for health services researchers, policy makers, and practitioners because return-to-work has demonstrated positive impacts on health (Ross & Mirowsky, 1995) and reduces high-cost health service use (Jackson et al., 2009; Kukla et al., 2012). Moreover, information about the role of informal care support policies in disabled individuals’ return-to-work could identify another mechanism to minimize the negative well-being and economic effects of long-term disabilities on families (Al-Janabi et al., 2016; Pacheco Barzallo, 2018).

Outside of VA, support and training of caregivers is ad hoc and limited for younger care recipients with functional limitations. However, recent momentum to develop and implement national caregiver support policies throughout the U.S. health care system suggests that systematic approaches to support informal caregivers will soon be available (Wolff et al., 2016). As such, results from this study can inform how supportive policies for informal caregivers outside of the VA system might enhance the use of services to address the cross-cutting health, social, and economic needs of individuals with disabilities.

Method

Data and Sample Selection

We use data from three sources: (1) national VA electronic health records (EHRs) that included information about VA and VA-purchased care, veteran demographics, ICD-9 clinical diagnoses, risk scores, facility-level factors, and use of supported employment; (2) VA Caregiver Application Tracker (CAT) data about PCAFC program application date, program eligibility determination, and caregiver relationship to the veteran; and (3) VBA data detailing the use of VBA services, including VR&E and the post-9/11 GI Bill. Veterans’ social security numbers were used to link data sets. As our study relies on VA health record and administrative data, we were unable to include detailed information about the caregiver. Due to VA regulations, we are also unable to share our data sets but can share codebooks, statistical code, and so on on request.

The sample included a national cohort of all U.S. veterans whose caregivers applied to the PCAFC between May 1, 2011, and March 31, 2014 (n = 24,942; Figure 1). To create this sample (Van Houtven et al., 2017), veterans were excluded if their identification number could not be matched to VA health records; their residence of record was outside the United States or Puerto Rico; they died within 3 months of application to the program, they enrolled in PCAFC for fewer than 90 days; they were older than 68 years of age at the time of PCAFC application or were older than age 65 years as of September 11, 2001; or they had a missing comorbidity score. For the primary analysis, we further constrained the sample to individuals younger than 55 years of age because we theorized that veterans who were 55 years or older when they applied for PCAFC might be less likely to subsequently seek vocational services.

Figure 1.

Sample flow diagram.

Note. PCAFC = Program of Comprehensive Assistance for Family Caregivers; IV = instrumental variable; VR&E = vocational rehabilitation and education; VBA = Veterans Benefit Administration.

We created a subsample for each vocational service outcome to capture incident appliers/users of each service. We excluded from the respective post-9/11 GI Bill and the VR&E samples veterans who applied for these programs prior to and including their PCAFC application date. We excluded from the supported employment sample veterans who used supportive employment at least one time between January 1, 2010, and their PCAFC application date. Each subsample employed the exclusion criteria only for the vocational service of interest, but individuals in each subsample could have used one of the other two vocational services. Additionally, we removed veterans from all samples if they did not have a social security number that matched across VBA and VHA records.

We also applied several exclusion criteria related to the instrumental variable (IV), which is defined in the section “Variables.” Veterans who applied during the first 6-month PCAFC application window and veterans whose “home” facility had no veterans in the sample in the prior 6-month period to their application were excluded because they would not have had a 6-month lagged IV value. Veterans whose home facility had a calculated IV of 0% or 100% were likewise excluded. Veterans who were excluded due to being in the first application period or because their facility had a 100% acceptance rate appeared to have more combat-related health conditions (e.g., TBI, pain) and used more VA services at baseline. Veterans whose home facility in a 6-month window had <10 veterans in the sample were also excluded. The final sample size for each outcome sample was the following: post-9/11 GI Bill (n = 9,776); VR&E (n = 9,390); and supported employment (n = 19,217). The Durham VA Institutional Review Board approved this study (Protocol No. 02115).

Variables

Outcome Variable.

The outcome variables are defined as “time to use” of each service between date of PCAFC application and the last day of the observation window for each outcome. Time to post-9/11 GI Bill application was calculated as the difference between the veteran’s application to post-9/11 GI Bill and application date to PCAFC. Veterans who did not apply for the post-9/11 GI Bill prior to August 26, 2016, were censored on this date, or date of death if deceased. Time to VR&E application was calculated as the difference between the veteran’s first application to VR&E and application date to PCAFC. Veterans who did not apply for VR&E prior to August 26, 2016 were censored on this date, or date of death if deceased. Of note, the application processes for these VBA programs are distinct from care received at the veteran’s home facility and from the teams who determine whether veterans are eligible for PCAFC. Time to supportive employment use was calculated as the difference between the veteran’s first use of supportive employment visit (defined using clinical stop codes: 222, 568; Twamley et al., 2013) and application date to PCAFC. Veterans who did not use supportive employment prior to January 31, 2017 were censored on this date or date of death if deceased. Compared with the VBA outcomes, we assessed supported employment use over a longer observation window because we had available VHA data over a longer time span.

Instrumental Variable.

We expect that unobserved baseline differences between the treatment and comparison groups might be correlated with the service use outcomes. For example, we are unable to account for baseline propensity to use vocational services or other unobserved confounding factors. We employed an IV to minimize potential endogeneity. The IV for each veteran was the proportion of veterans approved for PCAFC at the nearest VA Medical Center (VAMC)/Independent Outpatient Clinic (IOC; i.e., home facility) in the prior 6-month time period. Specifically, all veterans in the sample before exclusion restrictions were grouped into 6-month time periods based on their index application date to PCAFC. We assigned each veteran a home VAMC/IOC that was closest the veteran’s residence based on zip code at the time of index application. Then, within each application date-time period and for each VAMC/IOC, we calculated the proportion of veterans approved for PCAFC among all veterans with that home VAMC/IOC.

We contend that the IV we use is a theoretically strong IV that drives acceptance to PCAFC but does not relate to individual veteran or caregiver characteristics. To provide more context on why the lagged, facility-level PCAFC approval rate is a good candidate IV, it is important to know that PCAFC was rolled out to over 150 medical centers in a very short time period, and demand for the program was exponentially greater than expected. As such, PCAFC acceptance rates varied substantially over time and by VAMC. In response, the program procedures changed to increase the capacity of sites to efficiently process applications. For example, facilities standardized eligibility requirements over time and adopted the use of eligibility teams rather than individuals determining eligibility (Sperber et al., 2020). The IV therefore reflects the combined practice styles of the home facility examiners and the dynamic program constraints that occurred over time and so is not directly related to individual veteran or caregiver characteristics.

Primary Independent Variable.

The treatment variable of institutional support for informal caregivers is defined as having applied to and been accepted into PCAFC between May 1, 2011, and March 31, 2014, and having at least 90 consecutive days in PCAFC by November 1, 2014. To be eligible for PCAFC, the veteran must be receiving informal care for injuries sustained or exacerbated during military service. Enrollment in PCAFC was based on clinical assessments made by individual providers at program inception and then moved to multidisciplinary clinical team assessments by social workers, mental health providers, physicians, nurses, and occupational and physical therapists. The comparison group comprised veterans whose informal caregivers applied to but never enrolled in PCAFC between May 1, 2011, and March 31, 2014. Not meeting administrative eligibility criteria for PCAFC was the most common reason that veterans/caregivers were not enrolled in PCAFC. For example, caregivers were not enrolled in PCAFC if they cared for a veteran who did not serve in the U.S. military since September 11, 2001, or the veteran’s health conditions were deemed to be unrelated to military service. Veterans and caregivers who were not enrolled in PCAFC received usual care or standard VA benefits and health care per veteran-specific clinical eligibility, including case management, medical and mental health care services, disability compensation payments, and services and supports from home and community-based services. Standard VA Caregiver Support Program services were available to all caregivers of veterans who were determined to be in need of support for activities of daily living for at least 6 months not only those enrolled in PCAFC. These services included access to the Caregiver Support Line (a telephone hotline for caregivers of veterans), a caregiver listserv, self-care classes, and an interactive website. Caregivers were also eligible for mental health care and respite care when determined to be part of the veteran’s treatment plan.

Other Independent Variables.

Independent baseline variables in the analytical models included sociodemographic and economic factors, comorbidity indicators, access to and use of health care, and VA region and facility-level factors. Sociodemographic and economic characteristics included veteran age, gender, race, and means test status (an indicator of whether the veteran qualifies for copay exemptions due to low income). Health-related characteristics included percent service connection (an indication of disability status), comorbidities (e.g., PTSD, TBI, musculoskeletal conditions), VA enrollment priority, and Nosos comorbidity score (Wagner et al., 2016). Access to care included distance to VA care and prior use of primary and mental health care. Caregiver variables included the relationship between the veteran and caregiver. Fixed effects for the home facility or Veterans Integrated Service Network (VISN) were also included to control for local facility and regional market-level factors. Home facility fixed effects were included in the first stage models and VISN fixed effects were included in the second stage models because the prevalence of the outcomes was low enough to preclude regression with facility fixed effects. Indicator variables for the 6-month (5 months for latest appliers) application date time periods were constructed to control for time trends. Independent variables in Table 1 are presented as they were specified in the models.

Table 1.

Baseline Descriptive Characteristics of Veterans and Caregivers in Each Outcome Sample: Veterans Aged <55 Years at PCAFC Application.

Baseline characteristics	Post-9/11 GI Bill sample	VR&E sample	Supported employment sample
Number of observations in samplea	9,776	9,390	19,217
Caregiver approved for PCAFC	62.8%	61.3%	61.6%
Age at PCAFC application
19–30	28.5%	34.2%	31.8%
31–40	37.2%	33.8%	36.3%
41–50	27.4%	26.0%	26.3%
51–54	6.9%	6.0%	5.6%
Male gender	91.6%	91.9%	90.5%
Race
White/unknown	74.0%	73.2%	70.0%
Black/other	26.0%	26.8%	30.0%
Hispanic/Latino(a) ethnicity	11.0%	12.1%	12.3%
Married	67.4%	66.0%	66.7%
Caregiver is a veteranb	11.0%	11.0%	11.4%
Veteran homelessnessc	6.2%	6.6%	6.2%
Caregiver’s relationship to veteran
Spouse/partner	80.0%	80.5%	80.4%
Mother or father	8.6%	8.8%	8.1%
Other	11.4%	10.7%	11.5%
Enrollment priority group
Group 1	80.5%	76.7%	81.5%
Groups 2–4	10.6%	12.1%	10.6%
Groups 5–8 or missing	8.9%	11.2%	8.0%
Service connected
High (≥70%)	67.1%	62.1%	66.6%
Medium high (50% to 69%)	12.3%	13.5%	13.8%
Medium low (10% to 49%)	6.2%	8.0%	6.9%
Low (<10%) or missing	14.4%	16.4%	12.7%
Had a mental health visit in the 6 months prior to PCAFC applicationd	68.9%	69.3%	68.5%
Had a primary health care visit in the 6 months prior to PCAFC applicationd	69.8%	68.1%	69.2%
Diagnosesc
Physical comorbidities
Musculoskeletal disorders/diseases	61.9%	59.1%	61.0%
Vision loss	2.1%	1.7%	1.7%
Hearing: loss, pain, other	16.8%	16.7%	16.0%
Diabetes	8.3%	6.6%	7.5%
Neoplasm	6.5%	5.8%	6.1%
Hyperlipidemia	27.6%	24.0%	25.5%
Hypertension	24.2%	21.3%	22.5%
Obesity	18.0%	15.9%	17.7%
Pain, not including back or joint	44.8%	41.5%	43.2%
Traumatic brain injury	27.0%	27.6%	26.6%
Headache	18.2%	17.0%	18.6%
Joint pain and effusion, not including back	37.0%	36.2%	37.5%
Chest pain/acute myocardial infarction	6.8%	5.9%	6.3%
Mental health comorbidities
Adjustment reaction	10.3%	10.6%	10.0%
Anxiety	25.3%	26.1%	25.1%
Bipolar disorder	9.9%	10.0%	9.8%
Depression	49.6%	48.2%	48.7%
Other mental health	15.5%	14.4%	14.7%
Posttraumatic stress disorder	67.7%	68.5%	68.1%
Tobacco use	22.9%	22.3%	21.0%
Alcohol or substance use disorder	20.8%	20.8%	19.9%
Has non-VA insurance	13.5%	12.2%	12.4%
Must pay copay	12.6%	14.8%	12.2%
Copay unknown	20.1%	18.7%	21.1%
Copay exempt	67.3%	66.5%	66.7%
Nosos score is greater than 1	39.3%	33.2%	35.3%
Lives at least 30 miles from nearest VAMC	53.5%	53.4%	50.6%

Note. PCAFC = VA Program of Comprehensive Assistance for Family Caregivers; VR&E = Vocational Rehabilitation and Education Program; VAMC = Veterans Affairs Medical Center; VA = U.S. Department of Veterans Affairs.

^aThe Post-9/11 GI Bill, VR&E, and Supported Employment samples are all derived from the same parent sample. Veterans who applied for or used the service prior to or including PCAFC application date are excluded from the respective sample. Veterans remaining in the sample may or may not have applied for or used the service after the PCAFC application date, and unique veterans may be present in more than one of the samples depending on their own histories. Additional exclusions made based on the constructed instrumental variable (IV).

^bThis variable was constructed based on caregivers’ social security number matching in the Vital Status Mini File. It may not include caregivers who themselves are Veterans who do not utilize the VA health system (per VA electronic medical records).

^cAssessed in the year prior to and including application date.

^dNumber of visits in the 6 months prior to and including application date. Mental health care was VA provided or purchased; primary care was VA provided.

Empirical Strategy

We studied the effect of PCAFC on time to application/use of three VA vocational services among veterans with service-related disabilities who have an identified informal caregiver. We examined time to application of VBA post-9/11 GI Bill and VR&E services and time to first use of VHA supported employment. Generally, we write the vocational service outcomes as follows:

$$y_i=f\left(PCAF{C}_i,{\mathit{X}}_i,{\mathit{S}}_i,{\mathit{M}}_i,v_i\right),$$

where y_i represents the time to vocational service outcome for each veteran; PCAFC_i is a measure of whether each caregiver was enrolled in PCAFC for at least 90 days versus not enrolled; X_i is a vector of veteran and caregiver demographic, socioeconomic, and health variables; S_i is a vector of facility/VISN, or VA region, level indicator variables representing the closest VAMC to individual i; M_i is a vector of time indicators for the 6-month interval in which the veteran applied to PCAFC; and v_i is an individual error component.

Model Specification.

We estimated the model first treating PCAFC enrollment as exogenous. We assessed this effect using the naive regression form to estimate the hazard ratios (HRs) using a Cox proportional hazards regression, adjusting for baseline covariates, nearest facility or regional fixed effects, and 6-month application period fixed effects (Table 1). This equation takes the following form:

$$h_i(t)=h_0(t)\exp \left({\text{β}}_0+{\text{β}}_{1}PCAF{C}_i+\gamma {\mathit{X}}_i+\delta {\mathit{S}}_i+\lambda {\mathit{M}}_i\right),$$

where h_i (t) is the “hazard” for individual i initiating the service at time t; h₀(t) is the baseline hazard at time t, eventually conditioned out for model fitting and defined as the hazard when all covariates are equal to zero; and PCAFC_i and X_i are as defined above.

We next treat PCAFC enrollment as endogenous. To estimate the causal effect of PCAFC enrollment on the use of each of the three vocational services, we used a two-stage residual inclusion (2SRI) instrumental variable approach because the Cox model is nonlinear (Martinez-Camblor et al., 2019). As a reminder, the IV is the 6-month PCAFC approval rate at a veteran’s home facility from the 6 months prior to his/her application to the program, W_j. Assuming that the IV is strongly related to the primary independent variable and is not correlated with the dependent variable except through the primary independent variable, the inclusion of the residual term should eliminate the association between unobserved variables, the outcome, and the primary predictor. We modeled each outcome separately as a Cox proportional hazards model and estimated the proportional hazards of PCAFC enrollment on time to application/use of each vocational service outcome. In the first-stage model, enrollment in PCAFC is regressed on the IV and all observed confounders. This equation takes the following form:

$$logit\left(p_i\right)={\mathbf{\alpha }}_0+{\mathbf{\alpha }}_1{\mathit{W}}_j+{\mathbf{\alpha }}_2{\mathit{X}}_i+{\mathbf{\alpha }}_3{\mathit{S}}_i+{\mathbf{\alpha }}_4{\mathit{M}}_i,$$

where p_i is the probability of enrollment for individual i and W_j, X_i, S_i, and M_i are as defined above.

We then applied a Cox proportional hazards model and regressed time to use of each vocational service outcome on enrollment in PCAFC; all observed covariates that were included in the first stage (except the facility fixed effects); VISN-level fixed effects; 6-month application period fixed effects to control for time trends; the residual term from the first stage equation; and a normally distributed random effect for each individual (Martinez-Camblor et al., 2019). We included the random effect term in the second-stage equation because the inclusion of the residuals from the first stage induces variation in survival times in the second-stage Cox regression equation even when the true survival model does not include unmeasured confounding (Martinez-Camblor et al., 2019). This induces an association between the independent predictor and the unmeasured predictors of service use at that time point (Martinez-Camblor et al., 2019). Martinez-Camblor and colleagues show that including a random effect for each individual in the second stage produces consistent estimates under 2SRI (Martinez-Camblor et al., 2019). Individuals who did not apply for or use the service by the end of the outcome-specific observation window were censored at the end of the observation window, or date of death if deceased prior to that date. The second stage equations take the following form:

$$h_i(t)=h_0(t)\exp \left(\begin{array}{l}{\text{β}}_0+{\text{β}}_{1}PCAF{C}_i+\gamma {\mathit{X}}_i\hfill \\ +\mathbf{\delta }{\mathit{S}}_i+\lambda {\mathit{M}}_i+\text{δ}{\widehat{ϵ}}_i+b_i\hfill \end{array}\right),$$

where ${\widehat{ϵ}}_i$ is the estimated residual from the first stage equation for individual i, b_i is individual is random effect, and other terms are as described above.

We estimated the 95% confidence intervals of the IV models by taking the 2.5% and 97.5% percentiles from 100 bootstrapped samples. The IV model produces the local average treatment effect of PCAFC enrollment conditional on the unmeasured confounders and the random effect.

Evaluating Assumptions.

We evaluated model assumptions for each outcome sample. We visually inspected the Schoenfeld residuals and graphed the log cumulative hazard curves by treatment condition to assess whether our models violated the proportional hazards assumptions.

We checked the IV assumptions by examining the strength of the instrument through the partial F test of the IV in the first stage equation. A strong instrument was indicated by a partial F- statistic greater than 10 (Staiger & Stock, 1997). We also examined conditionally whether covariates were balanced between treatment and comparison groups on a median split of the instrument. For each outcome sample, two-groups were created; one group had IV values below the median and the other group had IV values at or above the median. We then examined the standardized differences for each covariate comparing the treatment versus comparison groups and above versus below median values of the IV groups to identify if the standardized differences were improved across median values of the IV compared with across the treatment/comparison groups. A standardized difference of 0.10 or below indicated satisfactory balance (Austin, 2009).

Sensitivity Analyses

In each respective sample, we applied the naive regression and IV regression approaches to examine the relationship between PCAFC enrollment and the outcome of interest in the sample that included veterans 55 years of age or older at the time of application to assess whether including them changed the observed relationship between PCAFC enrollment and time to application/use of VA vocational services. We also conducted a robustness check by running the outcomes in the naive and IV models as binary variables.

Results

Descriptive Results

Table 1 presents the baseline descriptive statistics of each outcome sample. The supported employment sample contained the most veterans (n = 19,217), partially because very few people were excluded for having used supported employment during the look-back period (January 1, 2010, to PCAFC application date). A higher proportion of veterans applied to VR&E (19.2%) compared with the post-9/11 GI Bill (14.7%). Use of supported employment was the lowest (1.7%) of all three services. In all samples, just over 60% of veterans had caregivers who were enrolled in PCAFC. Descriptive characteristics were similar across the three outcome samples. Average age was 35 years, and the veterans were primarily male and White/non-Hispanic. Across samples, most caregivers were married to the veteran, and veteran service connection percentage was high. Close to 70% of veterans had used mental health care and primary health care in the 6 months prior to PCAFC application. Prevalence of diagnoses of musculoskeletal disorders, pain, depression, and PTSD was also high.

Evaluating Assumptions

The proportional hazards assumption held for all three samples based on visual inspection of the Schoenfeld residuals. The untestable assumption of noninformative censoring was justified as censoring only arose due to study end, which was unrelated to patient outcomes and death. We have no reason to suspect a strong association between death and receiving supportive services; the proportion who died by the end of the follow up period was 3% for the post-9/11 GI Bill and supported employment and 4% for VR&E.

For all three outcomes, the F statistic in the first stage equation was greater than 10. For the post-9/11 GI Bill the F statistic was 11.3; for VR&E it was 17.2; and for supported employment it was 36.2. The IV improved balance across covariates versus the treatment/comparison groups. The absolute value of the standardized differences across levels of the IV was 0.10 or under for most variables; variables that were greater than 0.10 included race, ethnicity, TBI diagnosis, PTSD diagnosis, and proximity to a VAMC.

Cox Model Results

The naïve model (Table 2) demonstrated PCAFC enrollment was not associated with time to application to the post-9/11 GI Bill (HR = 0.94; 95% confidence interval [CI: 0.85, 1.0]). Enrollment in PCAFC was associated with a 16% decrease in the rate of applying to VR&E (HR = 0.84; 95% CI [0.76, 0.93]). Veterans whose caregivers enrolled in PCAFC engaged in supported employment services at a 29% higher rate compared with veterans whose caregivers were not enrolled (HR = 1.29; 95% CI [1.00, 1.67]).

Table 2.

Estimates of Hazard Ratios (HRs) and 95% Confidence Intervals (CIs) of PCAFC Approval on Engagement in Vocational and Education Assistance Services, Results From Naïve and IV Models.

Outcome Cohort	Sample size	Naïve HR [95% CI]	IV HR [95% CI]
Time to application to post-9/11 GI Billa	9,776	0.94 [0.85, 1.03]	0.93 [0.85, 1.05]
Time to application to VR&E	9,390	0.84 [0.76, 0.93]	0.84 [0.76, 0.93]
Time to first supported employment visit	19,217	1.29 [1.00, 1.67]	1.35 [1.14, 1.53]

Note. PCAFC = Program of Comprehensive Assistance for Family Caregivers; IV = instrumental variable; VR&E = Vocational Rehabilitation and Education Program. Brackets indicate 95% confidence interval constructed using standard errors for IV models from 100 bootstrapped replications using sampling with replacement. Use of supported employment is defined as at least one documented session of supported employment. IV is defined as proportion in the sample who were accepted into PCAFC in the previous 6-month time period within each “home facility” in the two-stage residual inclusion (2SRI) results. Comparison group = Veterans of caregivers who applied to but were denied entry into PCAFC; treatment group = Veterans of caregivers approved into PCAFC.

^aN = 16 veterans were in a home “VAMC” that did not have any veterans in the sample applying for the post-9/11 GI Bill in the assessed timeframe and were excluded from the analysis.

All models controlled for indicator if approved for PCAFC, age, gender, marital status, whether the caregiver was a veteran (per VA electronic health records), veteran received a homeless diagnosis or clinic stop in year prior to and including application date, race, Hispanic/Latino ethnicity, veteran/caregiver relationship, any mental health visits in 6 months prior to PCAFC application, any primary care visits in 6 months prior to PCAFC application, enrollment priority group, service connected, physical comorbidities including: musculoskeletal disorders/diseases, pain (not including back or joint), joint pain (not including back), hyperlipidemia, hypertension, traumatic brain injury, obesity, headache, hearing (loss, pain, or other), diabetes, neoplasm, and chest pain/acute myocardial infarction, mental health comorbidities including: PTSD, depression, anxiety, tobacco use, alcohol or substance use disorder, other mental health, adjustment reaction, and bipolar disorder, non-VA insurance, means test status (must pay copay, copay unknown), lives within 30 miles of VAMC, Nosos score ≥1, fixed effects for 6-month application date time periods, fixed effect for VAMC. Supported employment models controlled for VISN and not VAMC because 16 “home” VAMCs did not have any veterans in the sample with supported employment use in the assessed timeframe.

In the IV models, PCAFC enrollment was associated with a 35% increase in average time to first supported employment visit (HR = 1.35; 95% CI [1.06, 1.79]) and a 16% decrease in time to VR&E application (HR = 0.84; 95% CI [0.76, 0.92]; Table 2). We did not observe a statistically significant effect of PCAFC enrollment on average time to application for the post-9/11 GI Bill (HR = 0.93; 95% CI [0.85, 1.05]). See the appendix for full results.

Sensitivity Analysis Results

We assessed how sensitive our results were to sample restrictions by estimating the naïve and IV models in the samples that were not constrained to veterans younger than 55 years of age. The following number of veterans were excluded from the main models because they were 55 years or older: post-9/11 GI Bill sample (n = 1,291); VR&E sample (n = 1,027); supported employment sample (n = 1,508). In the sensitivity analyses, the effect of enrollment in PCAFC on time to application for the post-9/11 GI Bill was not significant in the naïve model (HR = 0.97; 95% CI [0.89, 1.07]) or the IV model (HR = 0.96; 95% CI [0.88, 1.06]). For VR&E, the naïve and IV models estimated a statistically significant lower HR (naïve HR = 0.86; 95% CI [0.77, 0.96]) and (IV HR = 0.85; 95% CI [0.75, 0.97]). For the supported employment sensitivity analyses, the naïve model estimated a statistically significant HR (HR = 1.29; 95% CI [1.01, 1.67]), but the wider IV CI rendered the IV-modeled HR not significant (HR = 1.30; 95% CI [1.00, 1.68]). The point estimates and 95% coverage rates in the sensitivity analyses that included veterans 55 years of age or older were similar to the point estimates from the main models.

All binary models yielded similar effect sizes as the Cox proportional hazard models of PCAFC approval on vocational rehabilitation and education assistance service use or application. Coefficients for other the variables in the models were also similar.

Discussion

Nearly 20% of veterans used VR&E, 15% used the post-9/11 GI Bill, and less than 2% used supported employment. The low use of supported employment is consistent with findings from prior research (Twamley et al., 2013) and may be related to programmatic differences between supported employment and VBA services (e.g., supported employment may increase barriers to entry by requiring a provider referral and focusing on veterans with serious mental illness). Despite this low use, veterans whose caregivers enrolled in PCAFC had a 35% increased use of VHA-provided supported employment. On the other hand, we did not detect a positive association between VA institutional support for informal caregivers through PCAFC and time to use of the VBA services—the post-9/11 GI Bill or VR&E. Lack of structural integration and coordination between VBA and VHA may have limited the ability of informal caregivers to help veterans use VBA services. We may have observed positive effects for supported employment as this service and the Caregiver Support Program are both clinical programs housed in VHA that have a direct focus on health and shared medical records. There are two channels through which this effect might occur: (1) informal caregivers may have learned about supported employment through interactions with the VA Caregiver Support Program staff and then communicated with the veteran’s health care team to encourage them to provide referrals to this service; and/or (2) the VA Caregiver Support Program staff could have communicated directly with the veteran’s health care team to encourage referrals to supported employment after meeting with the caregiver.

Another reason that we may not have observed positive effects of caregiver support on VBA outcomes is because of the added burden of coordinating services that are not structurally integrated in VHA. Intrahousehold welfare theory suggests that caregiver and veteran well-being is linked. As caregivers play a large role in helping their care recipients to manage health care, we expected that a similar relationship would exist for vocational services provided under the VA system. After all, engaging in vocational services has the potential to lead to long-term gains in income and personal well-being for each individual and the dyad, partly due to the ability for each member of the dyad to return to spending time on tasks for which they had comparative advantage prior to the injury. This could be work for the veteran, for example, and work or household production for the caregiver (e.g., raising children). However, caregivers dedicate substantial time to providing logistical support and coordinating veteran VA health care (Shepherd-Banigan et al., 2020). Therefore, the additional burden of also coordinating VBA vocational services—especially when these services are perceived to be functionally distinct from the veteran’s health care—may overwhelm the caregiver’s available time and capacity. The caregiver may perceive that efforts to support endeavors with uncertain, long-term future gains are less worthwhile, particularly when facing acute health care needs. Caregivers might also face personal barriers to coordinating care across health and social services, including lack of knowledge (i.e., social service literacy) or skills (i.e., ability to seek out social services online).

Perverse incentives could have played a role in the inverse relationship between PCAFC enrollment and use of VR&E services. Most veterans who receive disability benefits do not face the same limits on reentry to work such as those faced by Social Security Disability beneficiaries because VA disability benefits are not earnings tested (Rutledge & Wu, 2014). There are limits on work for veterans who receive monthly benefits due to the “total Disability Based on Individual Employability” designation, and VA benefits are protected while veterans participate in the supported employment program. However, families may be concerned that veteran employment obtained through VR&E might prompt the VA to reevaluate the veteran’s health status. In this case, if VA determines that the veteran is not as disabled as classified or not in need of much (or any) caregiving support, veterans could lose their service-connected disability classification and associated income and/or financial support from PCAFC. These concerns might discourage veterans from pursing vocational services (Duggan, 2014). VA offers vocational services to help veterans integrate into the workforce and VR&E services require veterans to put in place a job plan. However, supported employment jobs tend to have low wages and may not offset disability payments. The post-9/11 GI Bill does not require a job plan and veterans may view this benefit as a way to increase skills for leisure versus work (Shepherd-Banigan et al., 2020). Furthermore, the potential gains of education assistance offered through VR&E and the post-9/11 GI Bill are far in the future, and uncertainty about the ultimate outcome could discount the current-day perceptions about future gains, particularly amid concerns about the immediate loss of the PCAFC caregiver stipend or disability-related income.

Limitations

First, as we are unable to observe selection into PCAFC of dyads who did not apply, and the decision to apply to PCAFC is not random, our inferences may not generalize to dyads who might have been eligible had they applied to PCAFC. However, as this sample includes individuals who applied to PCAFC, our estimates may be conservative in that informal caregiver support could have a more pronounced impact for veterans who are less well informed about VA programs and services. Second, as with all caregiver research in a VA context, we are unable to observe important caregiver characteristics beyond their relationship to the care recipient. Third, our results estimate the effect for veterans who are at the margin of selecting into VA vocational service programs; yet, it is unclear exactly who is included in this group. Instrumental variable estimation delivers the local average treatment effect of PCAFC, which is likely different from the average treatment effect among the treated. Using the naïve results against the IV results can provide a range of potential average effects of PCAFC on the vocational service outcomes though our point estimates were so similar between the two approaches suggesting that endogeneity may have been minimal. Finally, given the low frequency of use of supported employment, we defined this outcome as having at least a single encounter with the program. This single encounter could represent a screening visit and not service engagement.

Conclusions

Return-to-work for people with disabilities can have positive impacts on family-level health, social, quality of life, and economic outcomes (Pacheco Barzallo, 2018). While informal caregivers are well positioned to help care recipients engage in vocational services, the structure of vocational services may matter for how to effectively leverage the efforts of caregivers. For example, our findings suggest that caregiver support had a positive impact on use of supported employment, which is incorporated into the veteran’s health treatment plan—health care is a domain that many caregivers successfully navigate (Shepherd-Banigan et al., 2020). To extend this impact to non-health-related vocational services, health systems may need to do more to integrate vocational and education assistance services into health care. This will include addressing misaligned incentives generated by disability/benefit programs and expected outcomes of vocational rehabilitation services. For example, elements of the supported employment model (i.e., one-on-one coaching, treatment plans that incorporate patient employment goals, protecting disability payments) could be adapted for VR&E and the post-9/11 GI Bill services. Consequent structural changes will require perspective shifts in health care systems to explicitly define employment and education as priority determinants of health. The VA Caregiver Support Program can provide a model for actions that can strengthen service integration and the role of caregivers. The program recognizes the potential for caregivers to help veterans achieve their life goals and promotes a psychosocial recovery-oriented model of veteran rehabilitation that acknowledges vocational rehabilitation as a component of health recovery. Moreover, program leadership and staff have strengthened relationships with VBA counselors and have conducted joint outreach activities with VBA to caregivers. The program has also worked to improve data access across VBA and VHA services (correspondence with VA Caregiver Support Program).

Models for extending caregiver inclusion beyond health care is an area for future research that will require data from in-depth interviews and surveys collected from caregivers, veterans, and health system administrators to understand how to improve caregiver knowledge and skills in social and health service integration, the role that veterans want caregivers to play, the bandwidth caregivers have to play this role, and the extent to which vocational rehabilitation services allows caregivers and veterans to return to intrahousehold production tasks (Becker, 1981).

Appendix

	Post-9/11 GI Bill (n = 9,776)		VR&E estimates (n = 9,390)		Supported employment (n = 19,217)
Baseline variable	IV model	Naïve model	IV model	Naïve model	IV model	Naïve model
Caregiver approved for PCAFC	0.93 [0.85, 1.05]	0.94 [0.85, 1.03]	0.84 [0.76, 0.92]	0.84 [0.76, 0.93]	1.35 [1.14, 1.53]	1.29 [1.00, 1.67]
Veteran age at PCAFC application
19–30—Reference
31–40	3.72 [2.96, 5.13]	3.52 [2.74, 4.54]	2.45 [1.93, 3.40]	2.41 [1.81, 3.21]	1.92 [1.08, 3.03]	1.98 [1.03, 3.80]
41–50	2.58 [2.12, 3.57]	2.49 [1.94, 3.19]	2.05 [1.60, 2.95]	2.03 [1.53,2.70]	1.39 [0.77, 2.17]	1.44 [0.76, 2.75]
51–54	1.83 [1.50, 2.50]	1.80 [1.41, 2.31]	1.61 [1.20,2.26]	1.60 [1.20, 2.13]	0.93 [0.60,1.52]	0.95 [0.49, 1.84]
Veteran male gender	0.98 [0.81, 1.17]	0.97 [0.83, 1.14]	0.97 [0.83, 1.13]	0.97 [0.82, 1.16]	1.07 [0.86, 1.36]	1.09 [0.74, 1.60]
Veteran married	1.09 [0.99, 1.23]	1.09 [0.99, 1.20]	0.94 [0.83, 1.05]	0.94 [0.84, 1.05]	0.81 [0.68, 1.00]	0.82 [0.64, 1.06]
Caregiver is a Veterana	1.07 [0.96, 1.23]	1.07 [0.94, 1.21]	0.92 [0.79, 1.10]	0.93 [0.79, 1.08]	1.12 [0.79, 1.41]	1.12 [0.79, 1.57]
Veteran homeless in past 12 monthsc	1.06 [0.85, 1.25]	1.05 [0.88, 1.25]	1.03 [0.82, 1.26]	1.03 [0.85, 1.24]	1.85 [1.63, 2.39]	1.78 [1.30, 2.45]
Veteran race
White/unknown	0.78 [0.69, 0.86]	0.79 [0.71, 0.87]	0.69 [0.62, 0.76]	0.70 [0.63, 0.78]	0.54 [0.44, 0.60]	0.54 [0.42, 0.70]
Black/other—Reference
Veteran Hispanic/Latino	1.18 [1.01, 1.34]	1.17 [1.03, 1.33]	1.07 [0.89, 1.24]	1.07 [0.92, 1.24]	0.88 [0.72, 1.28]	0.89 [0.61, 1.29]
Caregiver relationship to Veteran
Spouse/partner	1.01 [0.85, 1.19]	1.01 [0.88, 1.16]	1.04 [0.00, 1.21]	1.04 [0.89, 1.23]	0.80 [0.67, 0.89]	0.84 [0.59, 1.20]
Mother/father	0.69 [0.57, 0.83]	0.71 [0.59,0.86]	0.90 [0.69, 1.09]	0.90 [0.73, l.l1]	1.23 [1.02, 1.48]	1.26 [0.84, 1.89]
Other relationship—Reference
Veteran enrollment priority group
Group 1	1.08 [0.80, 1.46]	1.07 [0.77, 1.48]	1.03 [0.63, 1.48]	1.03 [0.69, 1.55]	0.78 [0.10, 1.02]	0.82 [0.25, 2.63]
Group 2–4	1.16 [0.98, 1.39]	1.14 [0.94, 1.38]	0.96 [0.68, 1.21]	0.97 [0.75, 1.24]	1.00 [0.35, 1.74]	1.06 [0.54, 2.07]
Group 5 or missing—Reference
Veteran service connection
High (≥70%)	0.99 [0.63, 1.49]	0.99 [0.69, 1.44]	0.63 [0.40, 1.06]	0.65 [0.41, 1.03]	0.73 [0.26, 3.97]	0.79 [0.22, 2.86]
Low (< 10% or missing)	1.84 [1.46, 2.34]	1.76 [1.41,2.19]	0.86 [0.64, 1.01]	0.87 [0.67, 1.12]	0.86 [0.29, 1.55]	0.94 [0.49, 1.80]
Medium high (50% to 69%)	0.91 [0.58, 1.33]	0.91 [0.62, 1.35]	0.75 [0.46, 1.18]	0.76 [0.48, 1.21]	1.09 [0.34, 6.39]	1.14 [0.31,4.21]
Medium low (10% to 49%)—Reference
Had a mental health visit in 6 months prior to PCAFC applicationb	0.93 [0.82, 1.03]	0.93 [0.83, 1.04]	1.02 [0.91, 1.13]	1.02 [0.90, 1.16]	2.15 [1.64, 3.75]	2.19 [1.52, 3.17]
Had a primary care visit in 6 months prior to PCAFC applicationb	1.02 [0.93, 1.13]	1.02 [0.92, 1.13]	1.08 [0.98, 1.18]	1.08 [0.96, 1.21]	0.96 [0.74, 1.02]	0.96 [0.74, 1.26]
Diagnosesc
Diabetes	0.85 [0.72, 1.04]	0.86 [0.71, 1.04]	0.81 [0.65, 1.05]	0.82 [0.64, 1.04]	1.59 [1.16, 2.16]	1.57 [1.03, 2.39]
Musculoskeletal disorders/diseases	0.90 [0.81, 1.00]	0.91 [0.83, 1.00]	0.92 [0.84, 1.03]	0.92 [0.83, 1.03]	0.80 [0.63, 0.94]	0.81 [0.63, 1.04]
Alcohol or substance use	1.10 [0.95, 1.22]	1.09 [0.97, 1.22]	0.87 [0.76, 0.98]	0.86 [0.76, 0.99]	1.21 [1.05, 1.58]	1.18 [0.90, 1.54]
Vision loss	0.94 [0.69, 1.24]	0.93 [0.70, 1.25]	1.04 [0.60, 1.38]	1.04 [0.72, 1.50]	0.58 [0.14, 0.79]	0.58 [0.22, 1.57]
Hearing loss	1.10 [1.01, 1.23]	1.09 [0.98, 1.22]	1.09 [0.97, 1.26]	1.09 [0.96, 1.24]	0.68 [0.52, 0.91]	0.68 [0.49, 0.94]
Hyperlipidemia	0.89 [0.82, 0.98]	0.89 [0.80, 1.00]	0.74 [0.63, 0.82]	0.74 [0.64, 0.85]	0.91 [0.75, 1.18]	0.90 [0.67, 1.22]
Hypertension	0.91 [0.80, 1.03]	0.91 [0.81, 1.03]	0.90 [0.76, 1.01]	0.90 [0.78, 1.04]	0.90 [0.71, 1.16]	0.91 [0.67, 1.24]
Obesity	1.09 [0.98, 1.22]	1.09 [0.97, 1.23]	1.02 [0.88, 1.17]	1.01 [0.88, 1.17]	0.94 [0.78, 1.16]	0.94 [0.69, 1.28]
Pain, including back or joint	0.95 [0.88, 1.04]	0.95 [0.87, 1.05]	1.02 [0.92, 1.14]	1.02 [0.92, 1.14]	1.02 [0.85, 1.36]	1.02 [0.80, 1.30]
Traumatic brain injury	1.07 [0.98, 1.18]	1.06 [0.96, 1.17]	1.15 [1.01, 1.27]	1.15 [1.02, 1.29]	1.29 [l.l1, 1.59]	1.29 [1.01, 1.66]
Headache	1.13 [0.98, 1.25]	1.12 [1.01, 1.25]	1.08 [0.97, 1.23]	1.08 [0.95, 1.23]	0.90 [0.76, 1.07]	0.91 [0.68, 1.22]
Joint pain and effusion, not including back pain	1.05 [0.94, 1.16]	1.05 [0.96, 1.15]	0.97 [0.87, 1.05]	0.97 [0.87, 1.08]	0.97 [0.80, 1.16]	0.96 [0.76, 1.22]
Adjustment reaction	1.16 [1.02, 1.35]	1.15 [1.01, 1.31]	1.25 [1.05, 1.43]	1.24 [1.07, 1.43]	1.29 [1.17, 1.60]	1.28 [0.93, 1.74]
Anxiety	1.00 [0.89, l.l1]	1.00 [0.91, l.l1]	0.95 [0.86, 1.10]	0.95 [0.85, 1.07]	1.21 [1.05, 1.59]	1.20 [0.94, 1.53]
Bipolar	0.90 [0.78, 1.08]	0.90 [0.77, 1.05]	0.85 [0.71, 0.98]	0.85 [0.72, 1.02]	1.29 [1.06, 1.60]	1.31 [0.97, 1.77]
Depression	0.94 [0.86, 1.04]	0.94 [0.86, 1.04]	1.03 [0.91, 1.12]	1.03 [0.93, 1.15]	0.93 [0.77, 1.02]	0.93 [0.73, 1.19]
Other mental health	0.81 [0.73,0.90]	0.82 [0.72, 0.93]	1.00 [0.87, 1.20]	1.00 [0.93, 1.15]	1.27 [1.12, 1.53]	1.26 [0.96, 1.67]
Posttraumatic stress disorder	0.98 [0.89, 1.09]	0.99 [0.89, 1.10]	0.89 [0.77, 1.02]	0.90 [0.79, 1.01]	1.03 [0.82, 1.18]	1.05 [0.78, 1.42]
Tobacco use	0.90 [0.82, 1.02]	0.91 [0.81, 1.01]	0.85 [0.75, 0.94]	0.85 [0.75, 0.97]	1.01 [0.75, 1.17]	1.02 [0.78, 1.32]
Chest pain/acute myocardia infarction	0.82 [0.66, 0.99]	0.82 [0.67, 1.00]	0.97 [0.77, 1.19]	0.97 [0.78, 1.21]	1.49 [0.96, 2.12]	1.47 [1.01, 2.14]
Neoplasm	1.00 [0.83, 1.16]	1.00 [0.83, 1.21]	1.10 [0.88, 1.31]	1.09 [0.88, 1.35]	1.29 [1.20, 1.64]	1.29 [0.84, 1.97]
Has non-VA insurance	0.98 [0.85, l.l1]	0.98 [0.87, 1.10]	1.13 [0.97, 1.31]	1.13 [0.98, 1.29]	1.31 [1.19, 1.55]	1.34 [0.99, 1.81]
VA copay exemption
Copay exempt—Reference
Must pay VA copay	1.06 [0.94, 1.22]	1.05 [0.91, 1.20]	1.05 [0.90, 1.00]	1.06 [0.92, 1.22]	1.13 [0.83, 1.20]	1.14 [0.80, 1.63]
VA copay unknown	1.14 [1.02, 1.23]	1.13 [1.02, 1.24]	1.04 [0.92, 1.16]	1.04 [0.92, 1.18]	0.68 [0.58, 0.87]	0.67 [0.48, 0.94]
Lives at least 30 miles from nearest VAMC	0.82 [0.77, 0.91]	0.83 [0.76, 0.90]	0.84 [0.75, 0.93]	0.85 [0.77, 0.94]	0.61 [0.46, 0.84]	0.62 [0.49, 0.78]
Nosos score is greater than 1 VISN	0.84 [0.75, 0.91]	0.84 [0.77, 0.93]	0.88 [0.78, 1.00]	0.88 [0.79, 0.99]	l.l1 [0.92, 1.16]	1.13 [0.88, 1.44]
10	0.86 [0.55, 1.26]	0.88 [0.56, 1.38]	0.64 [0.34, 0.96]	0.65 [0.40, 1.06]	0.36 [0.13, 0.58]	0.36 [0.15, 0.88]
11	0.79 [0.53, 1.22]	0.80 [0.56, 1.15]	0.73 [0.48, 1.01]	0.73 [0.51, 1.07]	0.11 [0.03, 0.15]	0.11 [0.04, 0.32]
12	1.00 [0.64, 1.50]	1.00 [0.69, 1.45]	0.73 [0.47, l.l1]	0.74 [0.50, 1.09]	0.40 [0.26, 0.74]	0.40 [0.20, 0.79]
15	0.89 [0.61, 1.36]	0.90 [0.64, 1.28]	0.53 [0.38, 0.78]	0.54 [0.38, 0.79]	0.44 [0.21, 0.77]	0.42 [0.21, 0.81]
16	1.12 [0.85, 1.59]	1.12 [0.82, 1.55]	0.57 [0.36, 0.78]	0.57 [0.40, 0.80]	0.42 [0.29, 0.66]	0.37 [0.20, 0.67]
17	1.31 [0.98, 1.79]	1.30 [0.96, 1.76]	0.66 [0.48, 0.93]	0.66 [0.48, 0.91]	0.44 [0.31, 0.55]	0.40 [0.23, 0.69]
18	0.84 [0.60, 1.27]	0.85 [0.61, 1.19]	0.88 [0.64, 1.20]	0.89 [0.65, 1.23]	0.30 [0.07, 0.55]	0.30 [0.17, 0.55]
19	1.44 [1.02, 2.17]	1.42 [1.02, 1.98]	0.93 [0.62, 1.31]	0.91 [0.65, 1.29]	1.46 [0.81, 2.24]	1.28 [0.76, 2.16]
2	1.18 [0.75, 1.63]	1.18 [0.77, 1.81]	0.68 [0.40, 1.05]	0.69 [0.41, 1.16]	0.34 [0.16, 0.75]	0.33 [0.10, 0.96]
20	0.83 [0.62, 1.16]	0.84 [0.61, 1.16]	0.64 [0.46, 0.88]	0.64 [0.46, 0.90]	0.37 [0.23, 0.57]	0.36 [0.20, 0.65]
21	1.04 [0.66, 1.40]	1.04 [0.74, 1.47]	0.91 [0.63, 1.28]	0.92 [0.66, 1.28]	0.29 [0.17, 0.37]	0.30 [0.16, 0.57]
22	1.28 [0.95, 1.77]	1.26 [0.92, 1.73]	0.99 [0.75, 1.34]	1.00 [0.73, 1.37]	0.09 [0.05, 0.15]	0.09 [0.04, 0.20]
23	1.23 [0.82, 1.75]	1.23 [0.84, 1.82]	0.72 [0.47, 1.18]	0.73 [0.45, 1.19]	0.85 [0.46, 1.61]	0.82 [0.41, 1.64]
3	1.06 [0.64, 1.54]	1.05 [0.67, 1.64]	1.07 [0.69, 1.49]	1.08 [0.73, 1.60]	0.30 [0.15, 0.32]	0.32 [0.15, 0.69]
4	0.91 [0.59, 1.45]	0.92 [0.62, 1.36]	0.50 [0.29, 0.77]	0.52 [0.32, 0.83]	0.16 [0.08, 0.35]	0.16 [0.06, 0.46]
5	1.25 [0.78, 1.83]	1.25 [0.84, 1.87]	1.03 [0.72, 1.54]	1.04 [0.69, 1.56]	0.26 [0.18, 0.67]	0.26 [0.11, 0.65]
6	1.07 [0.79, 1.46]	1.08 [0.79, 1.46]	0.58 [0.45, 0.78]	0.59 [0.43, 0.80]	0.30 [0.19, 0.47]	0.29 [0.16, 0.52]
7	1.21 [0.99, 1.65]	1.21 [0.89, 1.63]	0.71 [0.51, 0.95]	0.71 [0.52, 0.96]	0.14 [0.06, 0.23]	0.13 [0.07, 0.26]
8	1.16 [0.86, 1.53]	1.17 [0.85, 1.59]	0.78 [0.60, 1.07]	0.79 [0.58, 1.08]	0.41 [0.28, 0.57]	0.39 [0.23, 0.68]
9	0.93 [0.68, 1.34]	0.95 [0.69, 1.30]	0.65 [0.50, 0.86]	0.66 [0.48, 0.90]	0.21 [0.11, 0.33]	0.21 [0.10, 0.41]
Applicant group
Group 1—Reference
Group 2	1.02 [0.87, 1.18]	1.01 [0.89, 1.16]	1.19 [1.00, 1.40]	1.17 [0.98, 1.40]	0.91 [0.78, 1.37]	0.87 [0.60, 1.25]
Group 3	1.16 [0.99, 1.38]	1.15 [1.01, 1.32]	1.26 [1.03, 1.51]	1.23 [1.04, 1.47]	0.94 [0.72, 1.41]	0.87 [0.60, 1.26]
Group 4	1.13 [0.96, 1.34]	1.11 [0.97, 1.28]	1.24 [1.03, 1.53]	1.21 [1.01, 1.43]	1.16 [1.06, 1.68]	1.04 [0.73, 1.49]
Group 5	1.09 [0.91, 1.35]	1.08 [0.92, 1.27]	1.33 [1.03, 1.73]	1.28 [1.05, 1.55]	1.29 [1.07, 1.87]	1.06 [0.70, 1.61]
Residual estimated from first-stage equation	1.00 [0.93, 1.05]	NA	0.99 [0.93, 1.06]	NA	0.92 [0.86, 1.02]	NA

Note. PCAFC = VA Program of Comprehensive Assistance for Family Caregivers; VR&E = vocational rehabilitation and education; VAMC = Veterans Affairs Medical Center; VISN = Veteran Integrated Service Network; VA = U.S. Department of Veterans Affairs. “Post 9/1 l-GI Bill,” “VR&E,” and “Supported Employment” samples are all derived from the same parent sample. Veterans who applied for or used the service prior to or including PCAFC application date are excluded from the respective sample. Veterans remaining in the sample may or may not have applied for/used the service post PCAFC application date, and unique veterans may be present in more than one of the samples depending upon their own histories. Additional exclusions made based upon the constructed instrumental variable (IV).

^aThis variable was constructed based upon caregivers’ social security number matching in the Vital Status Mini File. It may not include caregivers who themselves are Veterans who do not utilize the VA health system (per VA electronic medical records).

^bNumber of visits in the 6 months prior to and including application date. Mental health care was VA provided or purchased; primary care was VA provided.

^cAssessed in the year prior to and including application date.