Prevalence and Determinants of Unmet Social Needs Among Rural and Urban Veterans

Abstract

Purpose. To examine the prevalence and determinants of nine unmet social needs among rural compared with urban Veterans. Methods. Retrospective study using survey data collected in 2020 merged with Veterans Health Administration (VA) administrative data. For each unmet need, separate logistic regression modes were run predicting the odds of rural compared with urban Veterans endorsing the need adjusting for sociodemographic characteristics and comorbidities. Findings. 2,801 Veterans responded to the survey (53.7% response rate). Veterans experienced high rates of need (e.g., 22% reported food insecurity). Unmet need prevalence varied minimally between rural and urban Veterans and where they did, rural Veterans were less likely to endorse the need (e.g., loneliness). For many unmet needs, Black compared with White Veterans were at higher risk. Regional unmet need disparities were also observed. Conclusions. As VA considers expanding unmet need interventions, tailoring interventions to the sub-populations most at risk may be warranted.

Upwards of 70% of health outcomes result from social determinants of health, the conditions in the environments where people are born, live, work, and age that affect a wide range of health outcomes and risks.^1,2 Unmet needs (e.g., lack of stable housing) resulting from adverse social and economic conditions are associated with the onset and progression of disease.³ These associations are especially well-documented for cardiovascular disease (CVD), the leading cause of morbidity and mortality in the United States (U.S.).⁴ In recognition of these associations, numerous professional organizations, including the American Heart Association now encourage health care delivery systems to focus on unmet needs.⁴ This policy shift has accelerated unmet need-related activities within the health care sector and led to a small but growing evidence base in support of health care-based interventions to screen for and address unmet needs.^5–7 However, much of the evidence has tended to focus on urban populations resulting in limited information about unmet needs among rural populations, how the needs of rural and urban populations may differ, and best designs for a rural-focused health care-based unmet need intervention. This knowledge gap is especially significant for the Veterans Health Administration (VA) where over a third of the Veterans served reside in rural area.^8,9

Rural U.S. differs from urban U.S. with respect to the social and economic conditions that drive health and health outcomes. Rural regions tend to have higher levels of poverty and fewer job opportunities than urban areas, as well as differences in the presence of and access to (e.g., longer distances to travel, fewer public transport options, unavailability of broadband) resources to address needs.^10–12 These differences may partially explain why rural compared with urban Veterans have more comorbidities, higher disease prevalence, and greater rates of dying by suicide.^13,14 At the same time, rural communities may also have unique assets such as community connectivity and anchor institutions such as schools and faith-based organizations, which may provide valuable resources to address social needs.¹⁵ As health care providers seek to reduce urban-rural health disparities, accounting for the potentially unique needs and circumstances of rural populations is essential.

The VA provides care to over nine million Veterans and of these about 30% reside in rural and highly rural areas.¹⁶ In contrast to some health care delivery systems, addressing the unmet needs of rural and urban Veterans as part of the clinical care model is not new to the VA. The Veterans Administration already invests in social work services to address unmet needs (though social work staffing is highly variable across facilities, especially in rural settings), systematically screens for two unmet needs (homelessness and food insecurity), and has developed programs to assist select Veteran sub-populations (e.g., Veterans experiencing homelessness).^17,18 However, as VA seeks to mitigate a broader range of unmet needs, there is limited understanding about the prevalence of unmet needs overall and across Veteran sub-populations. Filling this knowledge gap can provide critical guidance on where to target limited resources for screening and addressing unmet needs.

In this foundational study, we examined unmet needs among Veterans with and at risk for CVD using a national mail survey fielded in the Fall of 2020. We focused on this population for several reasons: 1) the association between unmet needs and CVD outcomes is especially well-established; 2) Veterans are at especially high risk for these clinical conditions;^4,19 3) for the prior two reasons, the survey was conducted in part to inform a rural-tailored unmet need intervention for Veterans with and at-risk for CVD. In addition to examining unmet needs among rural compared with urban Veterans, we assessed the determinants of unmet needs and if and how they differed between rural and urban Veterans.

Methods

Data sources and sample

This was a cross-sectional study using data obtained from the VA Corporate Data Warehouse (CDW) (9/1/2019–03/31/2021) and a national mail survey. Using CDW, we identified a cohort of 3,967,145 eligible Veterans based on select diagnostic codes from January 2019 to March 2020. Participants were included in one of two mutually exclusive cohorts based on ICD-10 codes: Veterans with CVD and those at risk for CVD. Veterans with CVD were defined as having a primary diagnosis related to coronary artery disease (I20–I25), cerebrovascular disease (I160–I69), and/or peripheral artery disease (I70–I72) excluding those without any other coronary artery or cerebrovascular-related diagnoses. Veterans at risk for CVD were defined as having diagnoses of hypertension (I10–I16), diabetes (E08–E13), or hyperlipidemia (E78) excluding those without more than one hyperlipidemia diagnosis not related to lipoprotein disorders or metabolism and without any of the previously defined CVD diagnoses. A minimum of one encounter with one or more of the previously mentioned ICD-10 codes determined a Veteran as eligible. We selected a stratified random sample of 5,204 Veterans to survey. Key strata were cohort (CVD vs. CVD-risk), geographic region (Continental, Northeast, Pacific, and Southeast), and race (White, Black, Other).²⁰ We oversampled small strata to improve the precision of our estimates. We fielded the survey in September–December 2020 following the Dillman approach.²¹

This human subjects study was reviewed by the VA Boston Healthcare System (VABHS) Institutional Review Board in November 2019, which made a determination of exemption (under category 2). The study received a notice to proceed from the VABHS Research and Development Committee in January 2020.

Measures

We developed a survey that included 10 items representing nine measures of unmet needs. The nine measures represent two dimensions of unmet need: 1) economic needs (food, housing, utility, financial, and employment); and 2) social and community-level needs (loneliness, legal, transportation, and neighborhood safety). We also obtained select sociodemographic information to use as predictors (i.e., educational status).

The survey measures were reproduced or adapted from survey items in existing validated tools. Five of the nine measures were drawn from the Accountable Health Communities Health-Related Social Needs survey and the Protocol for Responding to and Assessing Patients' Assets, Risks and Experiences;^22,23 the remaining measures were drawn from the English Longitudinal Study on Ageing, Roots to Health Survey, Survey of Household Economics and Decision-Making, and Black Rural and Urban Caregivers Mental Health/Functioning Survey.^24–27 We chose to create as opposed to use an existing survey tool because no one tool included items that mapped precisely to domains relevant to Veterans (see Supporting Information, Table A for the measure sources; supplementary material available from the authors upon request). Five of the nine measures were drawn from the Accountable Health Communities Health-Related Social Needs survey and the Protocol for Responding to and Assessing Patients' Assets, Risks and Experiences;^22,23 the remaining measures were drawn from the English Longitudinal Study on Ageing, Roots to Health Survey, Survey of Household Economics and Decision-Making, and Black Rural and Urban Caregivers Mental Health/Functioning Survey.^24–27 Five measures asked about current need (housing, employment, finance, lonely, neighborhood safety) and four measures asked about need in the past 12 months (utility, transportation, legal, food).

Survey measures included 1) food insecurity ("worried sometimes or often that food would run out before you got money to buy more" and/or "the food you bought just didn't last and you didn't have money to get more"); 2) housing ("worried about losing housing" or "don't have a steady place to live"); 3) utility ("electric, gas, oil, or water company threatened to shut off services or already shut off services"); 4) financial (financially "just getting by" or "finding it difficult to get by"); 5) employment (current work situation "unemployed" [as distinct from "otherwise unemployed but not seeking work"]); 6) loneliness (sometimes or often "feel lonely or isolated from those around you"); 7) legal ("need or want help with legal problems"); 8) transportation ("lack of reliable transportation kept you from medical appointments, meetings, work, or getting things you need for daily living"); and 9) neighborhood safety (strongly disagree, disagree, or mostly disagree with the statement "My neighborhood is a safe place to live"). Survey measures were recoded into binary variables to indicate whether the need was endorsed. Based on the frequency distribution of the number of needs per respondent, we also created one additional outcome measure representing having four or more of any of the nine unmet needs assessed (hereafter: "4+ Needs").

Outcome measures

In multivariate models predicting unmet need, we focused on the six most prevalent needs, defined as a prevalence rate of 10% or more overall or among either the rural or urban cohort. This included food, housing, finance, employment, legal needs, loneliness, and 4+ Needs. We report on all nine unmet needs in the descriptive analyses.

Predictor measure

The independent variable of primary interest was Veterans' rurality status. The VA Planning Systems Group designates a Veteran as urban, rural, highly rural, or islander based on census block population density. We excluded islanders due to small sample size and aggregated rural with highly rural, resulting in a two-category variable (rural/urban).

Covariates

We selected additional variables to include in the models based on work showing that they may be related to unmet needs.^5,6 Variables included sex (male, female), race (White, Black, Other), ethnicity (Hispanic, non-Hispanic), age in years, geographic region (Continental, Northeast, Pacific, Southeast), education (high school or less, more than high school), VA priority group based on June 2020 values, and Charlson Comorbidity Index (CCI) based on an 18-month period corresponding to the year prior to the earliest survey mail-out date and covering the time to respond to the survey.^20,28 All variables were derived from the CDW except for education which was derived from the survey data. Other race includes Veterans who are Asian, Native Hawaiian/Pacific Islander, and American Indian/Alaskan Native. Priority group indicates a Veteran's priority level for enrollment in VA. Veterans are assigned a VA priority group number indicating their priority for enrollment in VA based on specific eligibility criteria, including severity of service-connected disabilities and income level. Veterans with lower numbers have the highest enrollment priority and are exempt from copayments for health care visits, whereas copays are required for Veterans with higher numbers. We grouped these priority group numbers into three categories for analysis: 1–4 (greater service-connected disability and lower income), 5–6 (non-compensable service connected-disability and lower income), 7–8 (non-service-connected disability and higher income). The CCI is based on several categories of ICD diagnosis codes that are each assigned an associated weight from one to six and then summed resulting in a single comorbidity score; higher scores are predictive of increased mortality.²⁹ We specified CCI as a three-part ordinal variable (0, 1, 2 ≥).

Analysis

We compared the sociodemographic and clinical characteristics of survey respondents and non-respondents. We used weighting techniques to account for sampling stratification and oversampling. Next, we compared the characteristics and unmet needs of the rural and urban sub-samples. For all bivariate comparisons, we used chi-square tests of association, and mean differences were tested by two-tailed t-tests. The values of p <.05 were considered statistically significant. To test correlations among the nine outcome measures and by extension possibly inform combining some needs in subsequent modelling, we calculated Pearson correlation coefficients. For each of the five most prevalent outcome measures (defined as ≥ 10% of study sample endorsing the need), we ran separate weighted logistic regression models predicting the odds of endorsing the need or not. The models included all the independent variables described in the prior section. For these analyses, we report estimated odds ratios (OR) and 95% confidence intervals (CIs). We interpreted the estimated ORs in terms of different effect size: small (1.68), medium (3.47) and large (6.71).³⁰ For OR < 1, this translates to small (OR = 0.6), medium (OR=0.29) and large (OR = 0.15). In a secondary analysis, we ran the same modeling stratifying urban and rural Veterans for each of the needs and compared the estimated ORs for each independent variable, identifying non-overlapping 95% CIs. All analyses were completed using SAS 9.2.³¹

Results

Descriptive results

Study population

We received survey responses from 2,801 Veterans (53.7% response rate). The rural Veteran response rate was 51.8% and the urban Veteran response rate was 48.2%. Compared with non-respondents, survey respondents were on average older and more likely to be White race and non-Hispanic; they were also more likely to be in higher priority groups and to have less clinical complexity (see Table B, Supporting Information; supplementary material available from the authors upon request). Among survey respondents, rural compared with urban Veterans were significantly more likely to be White (91% vs. 80%; p < .0001), non-Hispanic (98% vs. 94%; p < .0001), and to have lower educational levels (44% vs. 33%; p < .0001) (see Table 1). Rural Veterans were also more likely to be in VA priority groups 5–6 (35% vs. 30%; p = .0087) and less likely to reside in the Pacific region (11% vs. 16%; p = .0004).

Prevalence of needs among rural and urban veterans

For rural and urban Veteran

Table 1.

CHARACTERISTICS OF RURAL VS. URBAN VETERAN SURVEY SAMPLE, U.S. VETERANS HEALTH ADMINISTRATION, UNITED STATES, 2020

Table 2.

RURAL VS. URBAN VETERAN UNMET NEEDS: WEIGHTED SURVEY DATA, U.S. VETERANS HEALTH ADMINISTRATION, UNITED STATES, 2020

respondents combined, the most prevalent needs were financial insecurity (23%), food insecurity (22%), legal needs (17%), and loneliness (14%) (see Table 2). Rural compared with urban Veterans were significantly less likely to endorse three of the nine needs that were assessed: they were less likely to be lonely (12% vs. 15%; p = .04), to be unstably housed or homeless (6% vs. 10%; p < .001), and to have transportation needs (6% vs. 8%; p = 04). Rural Veterans were also less likely to have 4+ needs (9% vs. 13%; p = .01).

Correlation among needs

Figure 1 illustrates correlations among social needs. The largest correlation was between needs related to finance and food (r = .54). Other pairs ranking among the highest correlations were finance and housing (r = .38) and housing and food (r = .34). All other paired needs had correlations below .30.

Multivariate results: Factors associated with the most prevalent unmet needs

Economic needs

Table 3 presents the logistic regression models indicating factors associated with the most prevalent economic needs (food, financial, housing, and employment). In all but one model, the rural coefficient was not significant. In the one model where rurality was significant (housing), rural compared with urban Veterans were significantly less likely to report the need, though the effect size was small (OR 0.67; CI=0.49, 0.92). Black compared with White Veterans were significantly more likely to report food needs (1.82; CI= 1.33. 2.50) and housing needs (2.65; CI –1.71, 4.10). Greater comorbidity was associated with significantly higher risk of food (1.72; CI=1.33, 2.23) and finance needs (1.72; CI= 1.34, 2.22).

Figure 1.

Correlation matrix of unmet social needs.

Veterans who reside in the Pacific compared with Northeast region were significantly more likely to experience food (OR 1.59; CI=1.14, 2.20), housing (OR 2.17; CI = 1.38, 3.43), and employment needs (OR 2.10; CI=1.40, 3.13) though the effect size for food was negligible. Veterans in higher priority groups for enrollment were significantly more likely to experience food and financial needs compared with Veterans in lower priority groups for enrollment. For example, Veterans in priority groups 5–6 compared with groups 7–8 were associated with an OR of 4.02 (CI=2.26, 7.15) for financial need. Lower educational levels were also associated with a greater risk for food needs (OR 1.79; CI=1.44, 2.22). For needs related to food, finance, and housing, increasing age was associated with reduced odds of having the need but the effect size was negligible (OR ranged from 0.95 to 0.96)

Social and community-level needs

Table 4 presents the logistic regression models for factors associated with the most prevalent social and community-level needs (loneliness and legal). The rural coefficient was not significant in any model. Black compared with White Veterans were significantly more likely to report both these needs (small to medium effect sizes). Veterans in priority groups 1–4 compared with groups 7–8 were significantly more likely to be lonely (OR 1.97; CI=1.08, 3.60). For both legal needs and loneliness, increasing age was associated with reduced odds of having the need. Female Veterans were significantly more likely to have legal needs (OR 1.67; CI=1.04, 2.66).

Four+ needs

With respect to the overall burden of needs, Black compared with White Veterans were significantly more likely report having 4+ needs (OR 1.97; CI=1.29, 2.96) (see Table 4). Compared with Veterans residing in the Northeast, Veterans residing in

Table 3.

PREDICTORS OF ECONOMIC NEED: WEIGHTED ADMINISTRATIVE AND SURVEY DATA, U.S. VETERANS HEALTH ADMINISTRATION, UNITED STATES, 2020

Table 4.

PREDICTORS OF SOCIAL AND COMMUNITY NEEDS, AND MULTIPLE NEEDS: WEIGHTED ADMINISTRATIVE AND SURVEY DATA, U.S. VETERANS HEALTH ADMINISTRATION, UNITED STATES, 2020

the Pacific region were significantly more likely to report 4+ needs (OR 1.86; CI=1.16, 2.97). Increasing age was associated with reduced odds of having 4+ needs but again the effect size was negligible (OR 0.96).

Rurality effect

Our analysis of sub-group effects when stratifying rural compared with urban Veterans revealed a few significant differences. These results are not reported in manuscript tables but can be found in Table C, Supporting Information (supplementary material available from the authors upon request). For example, the effect of region on transportation needs differed: rural Veterans in the Pacific compared with Northeast region were significantly less likely to report transportation needs (OR 0.33; CI=0.12, 0.96); while this sub-group in the urban model was significantly more likely to report transportation needs (OR 1.92; CI=1.02, 3.61). The effect of race on employment needs also differed between rural and urban Veterans: rural Veterans defined as Other race were significantly more likely to be unemployed (OR 2.20; CI=1.18, 4.12); while this race group in the urban model were less likely to report being unemployed (0.45; CI=0.18, 1.10). Finally, Hispanic rural Veterans compared to their White counterparts were significantly less likely to report housing needs (OR 0.08; CI=0.02, 0.37); while in the urban model, Hispanic compared to White Veterans were significantly more likely to report housing needs (OR 2.41; CI=1.26, 4.62).

Discussion

We examined the prevalence of nine social needs to understand social needs and their impacts among rural compared with urban Veterans. There are several important findings from this study.

First, we found that Veterans, regardless of rural or urban setting, experienced relatively high rates of need. Almost one in four (23%) experienced financial insecurity, which is consistent with other studies reporting that approximately 1.5 million U.S. Veterans live below the federal poverty level (approximately 8%) and an additional 2.4 million are thought to live paycheck to paycheck (approximately 13%).^32–34 Food insecurity was also highly prevalent (22%). Prior estimates of food insecurity among Veterans vary widely, ranging from 6% to 24%.³⁵ One study that was similarly based on a national representative sample (using 2005–2013 data) found that 8% of Veterans were food-insecure.³⁶ Our study thus updates these earlier findings and suggests a higher prevalence rate than previously reported. Finally of note, were the relatively high rates of legal needs and loneliness. A recent systematic review identified only one study examining perceived loneliness among Veterans: Kuwert et al. found that 10.4% of Veterans reported often feeling lonely.^37,38 The comparatively high rates of needs observed in our study may reflect the timing of our survey, which coincided with the COVID-19 pandemic, a period marked by increases in economic and social needs, including loneliness.³⁹

Second, contrary to our expectation, we observed relatively minimal differences between rural and urban Veterans in the type and prevalence of unmet needs. Where we did observe significant differences (housing), rural Veterans were significantly less—not more—likely to endorse the need. Our expectations of more prevalent needs among rural Veterans were based on the literature describing higher poverty rates, fewer job opportunities, transportation barriers, and challenges accessing resources in rural areas.⁴⁰ Possible explanations of our findings of minimal difference include that the VA invests in social work and programs to address some needs among some populations, and these programs may effectively reduce social need disparities between urban and rural Veterans. Additionally, there is some evidence that rural Veterans perceive social needs differently and by extension underreport having social needs.^41–43 At the same time, our finding of minimal difference in unmet social needs among rural and urban Veterans begs the question of what else is determining the known health disparities between rural and urban Veterans. Our study cannot address this, but it is possible that limited access to high-quality health care, well-documented as a more prevalent challenge in rural compared to urban settings explains some of the rural/urban health disparity.⁴⁴ This merits further study.

Third, certain Veteran sub-groups were especially vulnerable to social needs, including Black Veterans, Veterans in lower priority groups, and Veterans who reside in the Pacific region. Our finding that Black compared with White Veterans were significantly more likely to endorse most of the needs we examined is consistent with the literature on social needs.⁴⁵ However, it is somewhat counter to studies that find fewer racial health disparities within the VA compared with the general U.S. population.^41,46 Reduced racial disparities in the VA have been attributed in part to the VA's unique characteristics as an equal-access health care system (i.e., it provides care to eligible Veterans without the requirement to pay insurance premiums) and its efforts to address nonfinancial barriers to care such as transportation and housing.^41,46 Further research is needed to understand our observation of widespread racial disparities in unmet need in the context of a delivery system associated with relatively few racial disparities in health outcomes, as well as to understand whether there are disparities in access to VA social services and if so, how to minimize these barriers.

Our finding that Veterans in priority groups 5–6 and those in the Pacific region had higher needs are of note. With respect to the former, relative to priority groups 7–8, both groups 1–4 and 5–6 were associated with comparatively higher odds for financial and food needs. However, the effect size for groups 5–6 relative to groups 1–4 was substantially larger. The especially elevated risk among groups 5–6 may be explained in part by VA service-connected disability compensation (SDC): Veterans in groups 1–4 receive these benefits adjusted according to percent of service-connected disability, while Veterans in groups 5–6 do not. These benefits have been credited with Veterans in groups 1–4 having a significantly lowered poverty rate when compared with non-Veteran disabled.^32,46 This finding speaks well for the VA's efforts to serve Veterans in the priority groups with the highest priority for enrollment (i.e., Groups 1–4) not only in terms of health services but also in terms of economic security. With respect to the role of region, we found that Veterans in the Pacific relative to the Northeastern region had higher food, housing, and employment needs. We can't fully explain what might account for this, but some insight is gained by understanding that homeless Veterans in California represent 31% of the national homeless Veteran population.⁴⁷

Fourth, we found that some sub-group effects differed by rurality. For example, Other race rural Veterans relative White rural Veterans had higher odds of being unemployed; this effect was not observed among Other race relative to White urban Veterans. Overall, findings underscore the importance of accounting for how rural and urban conditions may affect different sub-populations differently when developing social need interventions. For example, a follow-up study could explore the conditions that explain the differing employment needs of Other race rural compared to urban Veterans (e.g., is it explained by differing transportation barriers, differences in education and training, etc.), which in turn could inform an intervention tailored to the potentially unique employment needs of Other race rural Veterans.

Our study has strengths but is not without limitations. Strengths include the high response rate, use of stratified sampling from broad regions in the U.S., and the consistency of our findings across models with respect to the significant determinants of endorsing unmet need. Limitations include the timing of our data collection. With certainty, the COVID-19 pandemic caused increases in many of the social needs we assessed for Veterans and non-Veterans alike. Thus, our findings may not represent non-pandemic times. Another limitation is the accuracy of responses, a common concern with surveys, as well that respondents may perceive social needs differently from one another. While all the survey items used in the study were drawn from existing validated tools and minimally adapted in only a few instances, the lack of formal development and testing of validity of the final tool is a limitation. Our focus on Veterans with and at-risk for CVD may limit the generalizability to Veterans without these conditions. At the same time, given the prevalence of these conditions among VA-enrolled Veterans, our findings represent a large and important sub-population. Relatedly, our sample, comprising Veterans with at least one encounter in the 15-month window may limit generalizability to VA-enrolled Veterans who do not use VA health services; it may also bias the results. For example, our findings may undercount unmet need if having a greater number of unmet needs is a barrier to utilization. Yet, we designed this study to inform possible health care-based interventions to address unmet needs among Veterans served by the VA. Such interventions largely depend on a patient encounter, which means our sample does represent the population that would be the target of an in-clinic unmet need screening and referral intervention.^5–7 Assisting Veterans who do not come in for health care services is critically important but will require a different kind of intervention; it was not the focus of this study.

In sum, the VA can advance population health by systematically screening for and addressing a wider range of unmet needs than the current screening practices for housing needs and food insecurityamong the Veterans they serve. The high prevalence of legal and financial needs relative to housing needs suggests two needs that might be prioritized for future screening efforts. Furthermore, the VA may consider expanding unmet need interventions, while tailoring these interventions to the unique needs of sub-populations and carefully considering equitable access to social services.