Abstract
Unmet social needs—including food, housing, and utilities—have been associated with negative health outcomes, but most prior research has examined the health associations with a single unmet need or analyzed samples that were homogeneous along one or more dimensions (e.g., older adults or patients with chronic health conditions). We examined the association between unmet social needs and psychosocial and health-related outcomes in a sample of Medicaid beneficiaries from 35 U.S. states. In 2016-2017, 1,214 people completed an online survey about social needs, demographics, and health-related and psychosocial outcomes. Seven items assessing social needs formed an index in which higher scores indicated higher levels of unmet needs. Participants were eligible if they were ≥18 years and had Medicaid. The sample was predominantly female (87%). Most (71%) lived with at least one child ≤18 years, and 49% were White and 33% were African American. Average age was 36 years (SD = 13). The most common unmet needs were not enough money for unexpected expenses (54%) and not enough space in the home (25%). Analyses controlling for recruitment method and demographics showed that increasing levels of unmet social needs were positively associated with stress, smoking, and number of chronic conditions, and negatively associated with future orientation, attitudes toward prevention, days of exercise/week, servings of fruits or vegetables/day, and self-rated health (all p < .01). Results add to the evidence about the relationship between unmet social needs and health. Interventions to help meet social needs may help low-income people improve both their economic situations and their health.
Extensive evidence suggests that health-related outcomes, including onset of illness and disease severity and progression, are associated with social determinants of health including educational attainment, income, occupation, neighborhood conditions, health-related knowledge and attitudes, and racism (Braveman, Egerter, & Williams, 2011; Braveman & Gottlieb, 2014; Marmot, Allen, Bell, Bloomer, & Goldblatt, 2012; Marmot, Friel, Bell, Houweling, & Taylor, 2008). Recent research on clinical practices to address social determinants of health among low-income populations has focused on a narrower set of material factors often called social needs or basic needs (e.g., Blazer, Sachs-Ericsson, & Hybels, 2007; McMullen & Katz, 2017).
These social needs—which include but are not limited to food, housing, and utilities—are associated with a range of negative health-related outcomes across a variety of contexts. For example, after abnormalities were detected on a cancer screening test, women who had more unmet social needs experienced a longer time to diagnostic resolution and were less likely to reach diagnostic resolution at all compared with women with fewer unmet needs (Primeau et al., 2014). Older adults reporting difficulty meeting social needs (financial, housing, and heat) at one point were more likely to report cognitive decline 3 years later (Sachs-Ericsson, Corsentino, Collins, Sawyer, & Blazer, 2010). In addition, older adults reporting higher levels of unmet social needs had higher levels of depressive symptoms (Blazer et al., 2007), poorer physical functioning (Sachs-Ericsson, Schatschneider, & Blazer, 2006), and higher mortality rates (Blazer, Sachs-Ericsson, & Hybels, 2005).
Despite evidence that low-income people often have multiple unmet needs (e.g., Kreuter, McQueen, Boyum, & Fu, 2016), researchers have often focused on the health correlates of a single unmet need at a time. Food insecurity has been the most widely studied need and has been associated with health outcomes such as disrupted sleep (Grandner et al., 2013), increased inflammation (Gowda, Hadley, & Aiello, 2012), lower self-reported health, and increased risk of chronic disease outcomes (e.g., hypertension and congestive heart disease; Gregory & Coleman-Jensen, 2017; Seligman, Laraia, & Kushel, 2010). Food insecurity has also been associated with greater cost-related medication nonadherence among older adults (Bengle et al., 2010; Sattler, Lee, & Bhargava, 2014) and increased risk of becoming a high-cost user of health care (Fitzpatrick et al., 2015). In addition, housing instability has been linked to poor health outcomes and reduced health care access (Burgard, Seefeldt, & Zelner, 2012; Charkhchi, Fazeli Dehkordy, & Carlos, 2018; Gibson et al., 2011; Keene, Guo, & Murillo, 2018; Shaw, 2004; Stahre, VanEenwyk, Siegel, & Njai, 2015).
Other work has examined the effects of unmet social needs in particular populations. In patients with diabetes, unmet social needs were positively correlated with indicators of poor diabetes control, and researchers found an additive effect: outcomes were worse as the number of unmet needs increased (Berkowitz et al., 2015). A qualitative study of heart disease and diabetes patients found that 23% reported that housing difficulties contributed to their hospitalization, often because unstable housing made it more difficult for them to engage in self-care and eat healthy foods (Quensell, Taira, Seto, Braun, & Sentell, 2017).
Although the evidence base linking social needs and health outcomes has grown in recent years, gaps remain. Research examining associations between unmet needs and diverse health outcomes in a single study has been sparse. The purpose of this study was to examine associations between unmet social needs and psychosocial and health-related outcomes in a sample of adult Medicaid beneficiaries from multiple states. Prior theories have depicted various ways that unmet basic needs may have adverse consequences (Bassuk & Donelan, 2003; Maslow, 1970). Our conceptual model focuses on health and is based on the idea that “scarcity captures the mind” (Mullainathan & Shafir, 2013, p. 7) and renders people less able to focus on long-term goals, including good health. Figure 1 presents a broad conceptual model of multiple potential pathways by which unmet needs may affect health outcomes. We posited that unmet social needs adversely affect health through a pathway that includes increased perceived stress and lowered future orientation. In turn, this reduces engagement in health behaviors and worsens self-reported health (Kim, Evans, Chen, Miller, & Seeman, 2018; Senn, Walsh, & Carey, 2014). In particular, we hypothesized that, even after controlling for demographic variables and method of study recruitment, a higher level of unmet social needs would be associated with barriers to self-care (higher perceived stress, lower future orientation, and lower valuing of prevention). We also hypothesized higher unmet needs would be associated with worse health behaviors (smoking, as well as less frequent fruit/vegetable consumption and exercise) and worse health outcomes (lower self-reported health and higher number of chronic conditions).
Conceptual model pathways by which unmet social needs may affect health outcomes.
Method
Surveys were administered online using Qualtrics software (Provo, UT, 2015) and could be completed via computer or mobile phone. A small number of participants who experienced technical difficulties or who did not have computer access completed the survey by telephone. All participants gave informed consent and received a $10 gift card for their time. Surveys were completed between September 2016 and August 2017. All study materials and procedures were approved by the Human Research Protection Office at Washington University in St. Louis and by participating Medicaid health plans and state agencies responsible for the review and approval of member material.
Participants
A convenience sample of Medicaid beneficiaries was recruited (see Supplemental Material, available in the online version of this article). Medicaid beneficiaries aged 18 years and older who were able to complete the survey in English or Spanish were eligible to participate. Nearly all participants were recruited through their membership in one of six Medicaid health plans from five states (Georgia, California [two plans], Missouri, New Hampshire, and Mississippi) that chose to participate. Recruitment methods included mailed and e-mailed invitations, automated phone calls, community outreach events, and home visits from health plan staff. When used, home and e-mail addresses of enrolled plan members were provided by health plans. Participants who met the eligibility criteria were also recruited through two volunteer research databases, one national (ResearchMatch) and one maintained by Washington University in St. Louis.
Measures
Survey measures assessed unmet social needs and a range of psychosocial and health-related outcomes. Measures chosen included those the research team had used in prior work, as well as those commonly used in national surveys to measure health and health behaviors.
Social Needs
Social needs were assessed using seven items (Kreuter et al., 2016) based on Segal’s Personal Empowerment Scale (Segal, Silverman, & Temkin, 1993) and work by Blazer et al. (2005). Items assessed the likelihood that respondents would have enough money for necessities, money for unexpected expenses, enough food, physical safety, and a place to stay in the next 30 days (very likely/likely/unlikely/very unlikely); other items assessed housing space (too much/about right/not enough); and neighborhood safety (very unsafe/unsafe/safe/very safe).
For descriptive analyses, needs were dichotomized into met/unmet (very unlikely/unlikely = unmet; not enough space = unmet; very unsafe/unsafe = unmet). For regression analyses, a continuous social needs index (similar to one used by Blazer et al., 2005) was created by summing responses to all social needs items (α = .70). For example, “How likely is it that you and others in your home will get enough to eat this month?” was scored so that very likely = 0, likely = 1, unlikely = 2, and very unlikely = 3. (One item, about space in the home, had a 3-point response scale and was scored 0-2.) Scores on this index of seven items could range from 0 to 20, with higher scores indicating greater unmet social needs. Because one needs item had a different response scale than the others, an index using z scores was also created; however, because both ways of calculating the index yielded similarly significant bivariate and multivariate results for the needs index, as well as an identical Cronbach’s α, the 0 to 20 index is used in these analyses for ease of interpretation.
Perceived Stress
We used the four-item Perceived Stress Scale (Cohen, Kamarck, Mermelstein, 1983; Cohen & Williamson, 1988) to assess stress over the previous month (e.g., “In the last month, how often have you felt that you were unable to control the important things in your life?,” scored 0 = never, 1 = almost never, 2 = sometimes, 3 = fairly often, and 4 = very often). Scores could range from 0 to 16, with higher scores indicating greater perceived stress; α = .68.
Future Orientation
This subscale of a Time Orientation Scale (Kreuter, Lukwago, Bucholtz, Clark, & Sanders-Thompson, 2003; Lukwago, Kreuter, Bucholtz, Holt, & Clark, 2001) sums responses to five items assessing the degree to which respondents plan for the future (e.g., “I have a plan for what I want to do in the next 5 years of my life,” scored 1 = strongly disagree, 2 = disagree, 3 = agree, 4 = strongly agree). Scores range from 5 to 20, with higher scores indicating a stronger future orientation (α = .60).
Attitudes Toward Prevention
A scale was derived from the mean of responses to five items adapted from previous measures (Gebhardt, van der Doef, & Paul, 2001; Moorman, 1990). Items include “I try to prevent health problems,” “I don’t think about my health unless there is a problem or I am sick” (reverse scored), “Taking care of my health is important to me,” “I am willing to make sacrifices to have good health,” and “I try to avoid things that are not good for me” (response scale: strongly disagree = 1 to strongly agree = 5). Higher scores indicate greater endorsement of beliefs and behaviors favoring prevention of negative health outcomes (α = .76).
Health Behaviors
Participants were asked how many days a week they engaged in moderate exercise (0-7) and how many servings of fruits/vegetables they ate per day (0-5+). Exercise and fruit/vegetable consumption were treated as continuous in regression models. Participants were also asked whether they smoke cigarettes which we dichotomized to reflect smoking status (1 = smoke every day or some days, 0 = not at all).
Health Status
Two variables assessed health status. A single item assessed self-rated health (excellent, very good, good, fair, poor; treated as continuous in analysis). Participants were also asked whether a doctor had ever told them they had any of 11 chronic conditions (yes/no for diabetes, stroke, hypertension, cancer, COPD/lung disease, arthritis, heart disease, depression, osteoporosis, vision problems, or hearing problems). The number of chronic conditions was summed.
Demographics and Recruitment
Covariates included age in years, sex (male/female), employment status (employed full- or part-time vs. not), marital status (married/partnered vs. not), annual household income (<$20,000 vs. $20,000 or higher), and whether any children aged 18 years or younger lived in the home (yes/no). Participants also self-reported ethnicity (Hispanic or Latino yes/no) and race, for which they could choose all categories that apply to them. For analysis, two dummy variables were created for racial categories, one for participants who described their race only as White and one for those who described their race only as African American; the reference group was made up of participants who reported being another race (American Indian/Alaska Native, and Asian/Pacific Islander), who described themselves as multiracial, or who did not report their race. Method of recruitment was collapsed into a binary variable (1 = recruitment via e-mail, 0 = recruitment via other methods).
Analysis
Descriptive and bivariate statistical analyses were conducted in IBM SPSS Statistics Version 24 (Armonk, NY, 2016). Pearson and point-biserial correlations were used to examine bivariate associations among the needs index and the outcome variables. The value for significance was set at p < .05.
Multivariable regression models were conducted in Mplus Version 7 (Los Angeles, CA, 1998-2015). Linear regression was used for six models (stress, future orientation, positive attitudes toward prevention, days of exercise per week, daily servings of fruits or vegetables, and self-rated health); one model used Poisson regression (chronic conditions) and one used logistic regression (smoking). The robust maximum likelihood estimator was used because it can handle nonnormally distributed data and missing data. Over 13% of participants did not provide data about income; missing values for income were estimated using the robust maximum likelihood estimator by including the mean value for income as a parameter in the model (UCLA: Institute for Digital Research and Education, 2017). All models controlled for sex, race, ethnicity, employment, age, marital status, having children aged 18 years or younger, and recruitment method.
Results
Due to the small number of people completing the survey in Spanish (n = 22), these analyses include only participants who took the survey in English. Of the 1,571 participants who consented to complete the English survey, 46 were ineligible (age < 18 years). When address records used for gift card mailings indicated that multiple people from the same household had taken the survey or that respondents had taken the survey more than once (e.g., by responding to both an initial e-mail and a reminder e-mail), only the first survey from an address was included in the analyzed data; this led to the exclusion of 68 surveys. An additional 239 respondents were excluded due to an error assessing eligibility (i.e., rather than identifying as Medicaid beneficiaries, they described their health insurance as “other” and added the name of their health plan or a descriptor such as “Obamacare,” which led the survey software to categorize them as ineligible). The final sample includes 1,214 eligible participants who answered at least one social needs question.
Most participants were recruited via e-mail (74%), mail (16%), or volunteer registries (7%). Compared with people recruited from health plans, volunteer registry participants were more likely to be older, male, and not married/partnered (p < .05); the basic needs index, however, was not significantly different between groups (p = .43). The 1,106 participants who provided an address to receive a gift card came from 35 states, the most common of which were those served by five of the six health plans: Georgia (49%), California (29%), New Hampshire (7%), and Missouri (7%).
Table 1 provides descriptive statistics. The mean age of participants was 36 years, and most described themselves as either White (49%) or African American (33%). Most participants were female (87%) and had children aged 18 years or younger in the home (71%). Of the 87% who reported household income, the majority had incomes below $20,000. The average number of chronic conditions was 1.0 (SD = 1.4), most commonly depression (23%), vision problems (19%), and hypertension (18%).
Characteristics of Participants (N = 1,214) Who Responded to at Least One Item About Social Needs.
Note. Percentages may not sum to 100% due to rounding. Data were missing as follows: children (n = 16), income (160), marital status (19), working (16), exercise (14), fruit/vegetable consumption (14), smoking (15), health status (14), perceived stress (1), prevention (7), and future orientation (4).
Of the seven social needs measured (Table 2), the most common unmet needs were lack of money for unexpected expenses in the next month (54%) and not enough space in the home (25%). About two thirds of participants reported one or more unmet needs. The average number of unmet needs was 1.3 (SD = 1.3), and the mean score on the social needs index was 5.7 (SD = 3.2).
Unmet Social Needs for Survey Participants (N = 1,214).
Note. Percentages may not sum to 100% due to rounding. Data were missing for needs items as follows: not enough space (n = 1), unsafe neighborhood (2), not enough to eat (1), someone likely to hurt you (1), and likely to have no place to stay (2). For the mean number of unmet social needs and the social needs index, n = 1,212.
Results of bivariate analyses (Table 3) demonstrated support for our hypothesis that level of unmet needs would be associated with health outcomes in bivariate analyses. Higher levels of unmet social needs were positively correlated with perceived stress scores, number of chronic conditions, and being a smoker (p < .01). Higher levels of unmet needs were also inversely associated with future orientation, positive attitudes toward prevention, days of exercise per week, daily servings of fruits or vegetables, and self-rated health (p < .01).
Bivariate Correlations Between the Social Needs Index and Outcome Variables for Survey Participants.
p < .05. **p < .01.
Results of multivariable analysis also supported our hypotheses that these relationships would remain significant after controlling for demographic covariates and method of recruitment (Table 4). Even with these additions to the models, the social needs index remained significantly associated with all eight health-related and psychosocial outcomes (all p < .01). The standardized parameter estimates for the needs index provide an indication of the strength of the effect. For perceived stress, for example, 1 standard deviation change in the needs index (3.2 points) is associated with a change in stress score equal to the product of the parameter estimate (.364) and the standard deviation of the Perceived Stress Scale (3.0). In other words, a 3.2-point increase in the needs index would be associated with a 1.1-point increase in stress score.
Results of Multivariable Linear and Poisson Regression Analyses for Survey Participants (N = 1,192) Predicting Health-Related and Psychosocial Outcomes.
Note. Est. = parameter estimate. Partially standardized coefficients are provided as parameter estimates for categorical variables and fully standardized coefficients are provided for continuous variables. Bold values are significant at p < .05.
In addition, the logistic regression model of smoking yielded a significant odds ratio (OR) for the social needs index (OR = 1.09, 95% confidence interval [CI: 1.04, 1.14]). Every 1-point increase in the social needs index was associated with a 9% increase in the odds of smoking. Several covariates were also significantly associated with smoking: being female (OR = 0.51, 95% CI [0.33, 0.80]), having children aged 18 years or younger (OR = 1.56, 95% CI [1.03, 2.36]), having an income below $20,000 (OR = 1.58, 95% CI [1.07, 2.32]), and being White (OR = 1.71, 95% CI [1.04, 2.80]).
Discussion
In this convenience sample of Medicaid beneficiaries, we found a consistent association between unmet social needs and a diverse set of health-related and psychosocial variables including perceived stress, future orientation, attitudes toward prevention, fruit/vegetable consumption, exercise, smoking, self-rated health, and number of chronic conditions. These results show support for our hypotheses. Prior work has shown an additive effect of multiple unmet social needs on health outcomes in samples including older adults (Blazer et al., 2005) and people with diabetes (Berkowitz et al., 2015). Our results demonstrate that these associations hold among a sample of Medicaid beneficiaries drawn from 35 states.
Although the effect sizes for the influence of social needs on these outcomes were not large, the fact that effects were seen consistently across a range of outcomes has implications for public health, since small-to-moderate effects on many health-related outcomes may add up to bigger effects, especially over time. Similarly, increasing levels of unmet needs were associated with all examined health and psychosocial outcomes, including those that are proximal (e.g., perceived stress) and those that are more distal (e.g., number of chronic conditions). Unmet social needs showed the strongest association with perceived stress, which is consistent with our conceptual model suggesting that stress would be a short-term outcome of unmet needs. This finding is also consistent with previous work showing that stress is negatively associated with health behaviors (Ng & Jeffery, 2003) and overall health (Watson, Logan, & Tomar, 2008). Our findings that unmet needs were associated with poorer self-reported health and a greater number of chronic conditions is consistent with previous findings in older adults that unmet needs are associated with decreased physical functioning (Sachs-Ericsson et al., 2006) and increased risk of mortality (Blazer et al., 2005).
Two thirds of participants reported one or more unmet social needs out of the seven we assessed, which is lower than rates from other studies conducted among low-income individuals in a range of health care and social service settings (Garg, Toy, Tripodis, Silverstein, & Freeman, 2015; Gottlieb et al., 2016; Kreuter et al., 2016). One obvious difference between our study and prior work is the primary mode of data collection; we used an online survey, whereas many other studies have collected data in person or by telephone. A second difference is that all our participants had been enrolled in Medicaid; none was uninsured. It is plausible that having access to Medicaid may have freed up financial resources for participants that enabled them to meet their social needs. Since enrolling in Medicaid requires people to take proactive steps, it is also plausible that resourcefulness as well as access to and comfort with technology may have been greater in our sample compared with other low-income samples. These findings have practical implications for assessing social needs. As more health care providers seek to assess social needs (Alley, Asomugha, Conway, & Sanghavi, 2016; Andermann, 2016; Crossley, Tyler, & Herbst, 2016; Fierman et al., 2016; Gottlieb, Wing, & Adler, 2017; McMullen & Katz, 2017), reliance on proactive use of online tools may lead to sampling bias. Providers and health plans must ensure that tools to assess social needs are accessible to all patients, even those not comfortable with technology. Strategies such as incorporating screening via tablet computer into usual care (e.g., Gottlieb, Hessler, Long, Amaya, & Adler, 2014) may help more patients use technology and may also improve reporting of sensitive social needs.
Several study limitations must be acknowledged. We assessed seven social needs, but additional needs (e.g., transportation and child care) may also be important in understanding health and psychosocial outcomes. The internal consistency of the Future Orientation Scale and the Perceived Stress Scale in this sample were lower than the Cronbach’s α reported in prior work (e.g., Cohen et al., 1983; Lukwago et al., 2001). Future research may benefit from including additional or revised items to measure these constructs, as well as from assessing whether low levels of reading or computer literacy affect the reliability of these measures when they are administered online. Measures with lower reliability may attenuate the associations with other variables. We focused on Medicaid beneficiaries and did not assess involvement with other social programs (e.g., the Supplemental Nutrition Assistance Program) that may have helped participants meet social needs; such holistic assessment of engagement in social programs will be important as the field advances. Although we found support for our hypotheses, the cross-sectional nature of the data precluded us from examining processes over time.
It is also possible that people who chose to respond to the survey differed from those who did not, as well as the broader group of people served by these health plans. As noted in the Supplemental Material, the response rates for most methods of contact were quite low. Contact information obtained from health plans was taken from enrollment data, which may have biased the sample toward participants whose contact information remained consistent over time (e.g., due to stable housing).
Strengths of this study include a large convenience sample of Medicaid beneficiaries from across the United States. Unlike many previous studies of unmet social needs, our sample spanned multiple geographic areas and health plans, and we examined associations with multiple health-related and psychosocial variables.
Implications for Theory, Policy, and Practice
Although we found preliminary support for our conceptual model, future longitudinal work is needed to investigate how social needs change over time and how they affect diverse health-related outcomes (and vice versa). Intensive longitudinal data collection could shed light on the processes and systems underlying our conceptual model and help generate effective solutions. Multilevel modeling would allow for examination of multiple levels of influence (e.g., potential differences in the influences of unmet needs on health-related outcomes among patients nested within health plans). Systems science research also holds promise for developing solutions more efficiently through simulations that model the effects of meeting social needs on different health outcomes over time. These techniques may enhance current theoretical understandings of the relationship between unmet needs and health-related outcomes, and longitudinal data collection would allow for the investigation of mediators in the causal pathway.
Our results also have implications for practice and suggest that interventions to help people with Medicaid meet social needs may yield a range of positive downstream consequences. This idea is consistent with findings that, for example, participation in the Supplemental Nutrition Assistance Program has been associated with lower health care expenditures (Berkowitz, Seligman, Rigdon, Meigs, & Basu, 2017). Past research has found that screening families at well-child visits and providing referrals for unmet needs increased the likelihood that parents would later report receiving community resources (Garg et al., 2015). Such screening and navigation programs could be expanded to community health centers and included as part of primary care practice with adults (Alley et al., 2016; Gottlieb et al., 2016; Gottlieb et al., 2017).
Finally, these results have implications for policy. As programs such as needs navigation are developed and implemented, it is important to acknowledge that availability of resources within a community may be a limiting factor in receiving help (Boyum, Kreuter, McQueen, Thompson, & Greer, 2016). To promote a wide range of positive health-related outcomes, public health practitioners and health professionals should advocate for increased support resources at local, state, and federal levels, as well as increased coordination across sectors. Our work adds to the growing body of evidence that suggests increasing such resources holds potential to improve the health of low-income people.
Supplemental Material
HEB822724_Supplemental_Material – Supplemental material for Social Needs and Health-Related Outcomes Among Medicaid Beneficiaries
Supplemental material, HEB822724_Supplemental_Material for Social Needs and Health-Related Outcomes Among Medicaid Beneficiaries by Tess Thompson, Amy McQueen, Merriah Croston, Alina Luke, Nicole Caito, Karyn Quinn, Jennifer Funaro and Matthew W. Kreuter in Health Education & Behavior
Footnotes
Acknowledgements
The authors would like to thank the health plans that participated in this research: Home State Health, Magnolia Health, Peach State Health Plan, California Health & Wellness, Health Net, and NH Healthy Families. The authors would also like to thank Balaji Golla and Katie Childs for their work on the online survey.
Declaration of Conflicting Interests
The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Karyn Quinn and Jennifer Funaro are employed by Envolve PeopleCare.
Funding
The authors disclosed receipt of the following financial support for the research and/or authorship of this article: This research was funded by the Envolve Center for Health Behavior Change, a collaboration between the Brown School at Washington University in St. Louis, The Center for Advanced Hindsight at Duke University, and Centene Corporation.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
