Health Care Utilization and Expenditures of Homeless Family Members Before and After Emergency Housing

Abstract

Objectives. To describe longitudinal health service utilization and expenditures for homeless family members before and after entering an emergency shelter.

Methods. We linked Massachusetts emergency housing assistance data with Medicaid claims between July 2008 and June 2015, constructing episodes of health care 12 months before and 12 months after families entered a shelter. We modeled emergency department visits, hospital admissions, and expenditures over the 24-month period separately for children and adults.

Results. Emergency department visits, hospital admissions, and expenditures rose steadily before shelter entry and declined gradually afterward, ending, in most cases, near the starting point. Infants, pregnant women, and individuals with depression, anxiety, or substance use disorder had significantly higher rates of all outcomes. Many children’s emergency department visits were potentially preventable.

Conclusions. Increased service utilization and expenditures begin months before families become homeless and are potentially preventable with early intervention. Infants are at greater risk.

Public Health Implications. Early identification and intervention to prevent homeless episodes, focusing on family members with behavioral health disorders, who are pregnant, or who have young children, may save money and improve family health.

More than a third of homeless individuals in the United States are part of a family unit.¹ Homeless families’ personal characteristics, health care needs, and concerns are different from those of single homeless individuals. For example, single homeless individuals are more likely to be adult males and have higher rates of addiction, severe mental illness, and other chronic disease.^2,3 By contrast, most homeless families are headed by women and include 1 or more young children.³ Pregnancy and childbirth are well-documented triggers for homelessness and critical health concerns for homeless families.^4,5

The health care needs and utilization patterns of homeless families received a significant amount of attention in the 1990s, but the ensuing 25 years have seen relatively few new studies.^6,7 During this time, health care coverage has improved significantly for poor children and, to a lesser degree, their parents.⁸ The Children’s Health Insurance Program, enacted in 1997, and subsequent Medicaid expansion for low-income adults under the Affordable Care Act extended insurance coverage to many more families at risk for unstable housing.⁸

Expanded coverage has also fueled concern about the cost and appropriate use of health care. Hopes of reducing emergency department (ED) visits, and ultimately costs, by increasing insurance coverage have not always been realized.^9,10 Concern that coverage alone may not lead to better access or more appropriate care, coupled with growing recognition of the role that unstable housing and other social determinants play in health, warrant a new look at the dynamics of health care utilization and homelessness.^9–11 Social determinants of health are increasingly seen as key targets in strategies to reduce costs and improve population health, but exactly how housing and other social factors affect consumers’ health care utilization remains poorly understood.¹²

Using administrative data from Massachusetts, we analyzed the health care utilization and Medicaid expenditures of families who experienced 1 or more homeless episodes during a recent 7-year period. We were particularly interested in how ED visits, hospital admissions, and total health care expenditures changed in the period leading up to and during an emergency shelter episode, as well as the impact of specific frequently occurring conditions, previous use of emergency shelter, and location of care. Our goal was to contribute knowledge that can be used in screening for social determinants and to shape interventions addressing the health care needs of homeless families.

METHODS

We linked administrative data for individuals living in families who received emergency housing assistance (EA) from the Massachusetts Department of Housing and Community Development with Medicaid (MassHealth) claims and enrollment data between July 1, 2008 and June 30, 2015. We used additional claims data from 2007 to describe service utilization before homelessness during 2008. In Massachusetts all qualified homeless families are entitled to publicly funded shelter; thus, families receiving EA represent most of the homeless families in the state during this period. To receive EA, families must be Massachusetts residents, meet income standards similar to those for Medicaid, have a child younger than 21 years, or be pregnant and be homeless owing to specific causes.¹³ During the study period, EA was primarily limited to shelter although ancillary placement services may also have been provided in a limited number of cases. Although the goal is to assign families to shelters near their previous residence, shelter constraints often require a more distant placement. EA is a benefit reserved for families and is managed separately from benefits for single homeless individuals.

Eligibility requirements for EA are like those for MassHealth, and most homeless families receive both benefits. We linked EA records for individual family members with MassHealth claims during the study period on the basis of birth date, social security number, and gender. These data were complete in both EA and MassHealth databases, making linkage mistakes unlikely but not impossible. It is likely that those who did not link had different insurance coverage.

Defining Shelter Episodes

Using beginning dates for EA, we identified episodes of emergency shelter for each person. We then constructed monthly (30-day) longitudinal measures of service utilization and expenditures for up to 12 months before each episode’s beginning. The Department of Housing and Community Development reports that rates of beginning EA are typically consistent with shelter entry within a few days. Department policy requires a new application if an EA request is not completed within 30 days. However, many families may experience a period of unstable housing before applying for EA. Ending dates of EA coverage were generally available for heads of households but not always included for other family members. Family identification numbers used to link individuals to specific families were missing for many members, which prevented analysis by family unit.

Variables

The International Classification of Diseases, Ninth Edition (Hyattsville, MD: National Center for Health Statistics; 1978) diagnoses associated with claims described the most frequent conditions for children and adults overall and separately for ED visits and hospital admissions. Outpatient claims included up to five 5-digit diagnoses per claim. Inpatient claims recorded up to 10 diagnoses. To simplify presentation, we grouped diagnoses into Clinical Classification System categories,¹⁴ with additional breakouts for some diagnoses reported at high frequencies in previous studies, such as pregnancy, depression, substance use disorders, and posttraumatic stress disorder (Table A, available as a supplement to the online version of this article at http://www.ajph.org, shows the diagnostic codes). We also used DxCG risk scores (a commercial risk adjuster used by MassHealth) to summarize the overall disease burden of the study group compared with all MassHealth members.¹⁵

We modeled ED visits not resulting in a hospital admission, hospital admissions, and total expenditures in each month using generalized estimating equations with covariates for time (months numbered sequentially from 1 through 24), year of episode start date, whether a month was before (= 1) or during (= 0) a homeless episode, whether the individual lived in the greater Boston metropolitan area, where provider concentrations and housing stock may be substantially different from other regions of the state (= 0) or elsewhere (= 1) before an episode, each person’s age at an episode’s beginning, and whether the current episode was their first (= 1), second, or third or greater episode (= 0). On the basis of a preliminary analysis of the most frequently occurring diagnoses and evidence from other studies that these conditions are associated with higher expenditures,^16,17 we also included variables indicating whether each individual was diagnosed with 1 of the following conditions at any time during the entire 24-month study period: anxiety, depression, a substance use disorder, or pregnancy.

Analysis

We designed the models to be exploratory and descriptive rather than to test a hypothesis. Models of visits and admissions used a logit link function and an exchangeable correlation structure. Expenditure models used a γ distribution with a log link function. Because of differences in utilization patterns, we conducted separate analyses for children (younger than 18 years) and adults.

We performed all analyses with SAS version 9.2 (SAS Institute, Cary, NC).

RESULTS

We describe the characteristics and utilization patterns of children and adults separately (Table 1). In Table 2, we present the results of multivariable analyses using generalized estimating equations. Results for ED visits and hospitalizations, which are bivariate, are presented as adjusted odds ratios (AORs). Continuous expenditure estimates are presented as marginal effects.

TABLE 1—

Characteristics of Homeless Children and Adult Medicaid Beneficiaries Who Received Emergency Family Shelter Between 2008 and 2015: Massachusetts

Characteristic	Children (n = 44 040), Mean (SD) or %	Adults (n = 34 783), Mean (SD) or %
Age, y	5.5 (5.0)	29.2 (8.7)
Gender
Female	49	78
Male	51	22
Race/ethnicity
White	22	28
Black	21	22
Hispanic	17	16
Other	3.6	2.3
Unknown	37	30
No. episodes
1	88	87
2	11	12
≥ 3	1.2	1.5
Living in Boston region	22	23
DxCG score, illness burden
< 1 (< average MassHealth member)	77	54
≥ 1 (≥ average MassHealth member)	23	46
ED visits before EA period
No. ED visits/month	0.052 (0.304)	0.129 (0.462)
% with any utilization	35	53
ED visits during EA period
No. of ED visits/month	0.074 (0.255)	0.129 (0.451)
% with any utilization	32	43
Hospitalizations before EA period
No. of hospitalizations/month	0.003 (0.057)	0.020 (0.140)
% with any hospitalization	3.5	21.0
Hospitalizations during EA period
No. of hospitalizations/month	0.004 (0.062)	0.020 (0.138)
% with any hospitalization	2.7	14.0
Mean expenditures/month
Before EA period	268.89 (2438.82)	484.75 (1906.24)
During EA period	306.58 (1911.23)	523.66 (1893.38)

Note. EA = Emergency Housing Assistance; ED = emergency department.

TABLE 2—

Analysis of Emergency Department Visits, Hospital Admissions, and Expenditures: Massachusetts, 2008–2015

Variable	Children, Estimate (95% CI)	Adults, Estimate (95% CI)
Emergency department visitsa
Age per 5 y	0.68 (0.67, 0.69)	0.97 (0.96, 0.98)
Before (= 1) vs during (= 0) emergency assistance	0.62 (0.60, 0.64)	0.90 (0.87, 0.92)
Other regions (= 1) vs Boston (= 0)	1.01 (0.98, 1.04)	1.23 (1.19, 1.26)
No previous episode (= 1) vs ≥ 1 (= 0)	0.93 (0.89, 0.97)	0.85 (0.82, 0.89)
Hospital admissionsb
Age per 5 y	0.74 (0.69, 0.79)	1.03 (1.02, 1.05)
Before (= 1) vs during (= 0) emergency assistance	0.90 (0.78, 1.03)	0.84 (0.79, 0.89)
Other regions (= 1) vs Boston (= 0)	1.03 (0.91, 1.17)	1.00 (0.96, 1.05)
No previous episode (= 1) vs ≥ 1 (= 0)	0.95 (0.79, 1.15)	1.00 (0.96, 1.06)
Expendituresc
Age per 5 y	−0.24 (−0.27, −0.21)	0.03 (0.02, 0.04)
Before (= 1) vs during (= 0) emergency assistance	0.15 (0.09, 0.20)	0.06 (0.03, 0.09)
Other regions (= 1) vs Boston (= 0)	−0.14 (−0.20, −0.07)	0.05 (0.02, 0.08)
No previous episode (= 1) vs ≥ 1(= 0)	0.10 (0.03, 0.17)	0.02 (−0.01, 0.06)

Note. AOR = adjusted odds ratio; CI = confidence interval. All analyses are generalized estimating equations. Bivariate outcomes, emergency department use, and hospital admission are presented as AORs. Continuous outcomes are marginal effects.

^aAOR estimates; ≥ 1 visit/month.

^bAOR estimates; ≥ 1 admission/month.

^cMarginal effects estimates. Natural log of monthly expenditures.

Cohort

Ninety-six percent of individuals receiving EA between 2008 and 2015 were linked to MassHealth records. The study group included 78 813 children (55.9%) and adults (44.1%). Almost 8 in 10 adults were women, with a mean age just younger than 30 years. Children averaged younger than 6 years. Although a substantial portion of adults and children were missing information on race and ethnicity, the available data suggest larger percentages of Black and Hispanic members than in the general population of Massachusetts. Approximately 1 in 5 members lived in the greater Boston area. Most participants (88%) had only 1 episode of homelessness during the study period, with a small group ranging from 2 to 6 episodes. The median episode lasted 193 days (mean 243), with an interquartile range from 70 to 337. Details of group characteristics are shown in Table 1.

The overall burden of illness was relatively low compared with the broader MassHealth population, with 77% of children and 54% of adults having lower than average DxCG scores (1 indicates average). The most common condition diagnosed for adults in the 12 months leading up to a homeless episode was pregnancy (28.8%), followed by abdominal pain (24.6%), depression (23.5%), back pain (22.3%), and anxiety disorders (22.1%.) These continued to be the most frequent conditions for adults after emergency housing but in a different order of frequency: depression (23.6%), anxiety (22.4%), pregnancy (19.5%), back pain (19.0%), and abdominal pain (18.7%). In the 12-month pre-episode period, children’s claims were most likely to list an administrative or social problem (58.4%), followed by upper respiratory infections (29%), lower respiratory disease (17.1%), otitis media (14.4%), and nutritional, endocrine, or metabolic disorders (13.3%). Administrative and social admissions cover a range of routine conditions or procedures that are not tied to a specific diagnosis. Issuing repeat prescriptions (54.0%), routine infant or child health checks (9.4%), and lack of housing (9.1%) were the most frequently cited conditions in this category. The same conditions in the same order persisted at slightly lower rates among children after entering shelter.

Outcomes

One third (35%) of children had at least 1 ED visit in the 12 months leading up to an EA episode, and more than one half (53%) of adults did (Table 1). Comparable data for other MassHealth members was not available for comparison. ED visits and hospital admissions (Figure 1) showed similar trends of accelerating utilization before the beginning of an EA episode followed shortly afterward by a gradual decline in utilization. Children’s utilization patterns were like those of adults but at consistently lower levels. Although adult ED use returned to pre-episode levels later in the homeless period, children’s ED visit rates declined somewhat but remained higher 12 months after the beginning of a shelter episode than at the beginning of the 12-month period before EA. ED visits and hospital admissions were consistently higher for children and adults with behavioral health or pregnancy diagnoses. Multivariable analyses confirm that these trends were statistically significant after adjusting for timing of episodes and region of residence. Younger children had significantly higher rates of ED use and hospital admissions and higher expenditures than did older children. Older adults had slightly more hospital admissions and higher expenditures but used EDs a little less frequently than did younger adults.

FIGURE 1—

Emergency Department (ED) Visits and Hospital Admissions: Massachusetts, 2008–2015

Adults and children also differed materially in their reasons for using EDs and inpatient care (Table B, available as a supplement to the online version of this article at http://www.ajph.org). Pregnancy and related complications, abdominal pain, back pain, and respiratory disease were the most frequent reasons for adult ED visits. Upper respiratory infection, lower respiratory disease, otitis media, asthma, and allergic reactions accounted for most ED visits by children.

Pregnancy and childbirth (40.0%, overlapping with pregnancy diagnoses) were the diagnoses listed most often for adult hospital admissions, followed by alcohol or drug use disorders, depression, asthma, and anxiety disorders. Complications of birth, asthma, upper respiratory infections, alcohol or drug use disorders, and allergic reactions were the conditions most frequently associated with children’s hospital admissions.

Most family members used at least some health care during the study period: 92.4% of children and 93.3% of adults had 1 or more claims. Total health expenditures per month varied widely from person to person (Table 1). Figure 2 shows rising expenditures leading up to shelter entry and a gradual tapering afterward for children and adults, with consistently higher spending for adults. These trends before and during homeless episodes resemble those observed for ED visits and hospital admissions. Expenditures for individuals with behavioral health conditions (anxiety, depression, or substance use disorders) and pregnancy followed a similar pattern before and after EA but were significantly higher throughout the 24-month observation period. Expenditures for women with a pregnancy or pregnancy complication diagnosis at any time during the observation period declined more sharply after EA.

FIGURE 2—

Total Health Expenditures: Massachusetts, 2008–2015

Note. SUD = substance use disorder.

Other multivariable results for other locations, compared with the greater Boston region, showed significantly higher odds of hospital admission and higher expenditures for adults and lower total expenditures for children who lived outside Boston before entering shelter. Adults and children with no previous shelter episodes had lower odds of an ED visit than did those with previous episodes of homelessness. Children with no previous episodes also had slightly higher expenditures than did those with 1 or more previous visits.

DISCUSSION

Adults and children relied heavily on EDs for care that is typically addressed in less intensive settings.^18,19 Diagnoses associated with children’s visits—respiratory infections, otitis media, and asthma—have all been identified as conditions in which emergency care can be prevented with appropriate management in less intensive settings.^20,21 Diagnoses associated with adults’ use of EDs included pregnancy complications, respiratory conditions, back pain, and other conditions that may also have been treatable in less costly primary care or urgent care settings. Infants and younger children were more likely than were older children to have an ED visit or hospitalization, perhaps because of parents relying on EDs for routine care or their heightened concern for a young child’s well-being. Placing families in shelters away from their home community can considerably disrupt regular sources of care, resulting in higher levels of ED use.²²

Prioritizing families with special health care needs for local shelter placement would allow them to maintain critical health care relationships. When a family must be placed in a region that is distant from their home, efforts to strengthen communication and information sharing across health care sites as well as training shelter staff to reach out quickly to establish health care access may improve care continuity.²² Recent evidence suggests that parents’ health literacy may also play a role in frequent ED visits.^23,24

Pregnancy and childbirth were leading conditions associated with adults’ ED visits and had the strongest association with hospital admissions. Increasing stress and disrupted access to usual care sources in the periods before and during homelessness may lead to more pregnancy and childbirth complications and greater reliance on emergent treatment settings for prenatal and postpartum care.²⁵ Although our understanding of how best to intervene would improve with further study of the dynamics of homelessness and perinatal care, these findings clearly show that pregnancy and childbirth are among the most important drivers of health care utilization and cost in this population. Several programs provide housing and intensive case management for pregnant women at risk for homelessness, but the literature offers few well-tested interventions that target homeless pregnant women. Health care–based interventions that aim to provide pregnant women with stable housing in conjunction with prenatal care may improve birth outcomes and reduce costs.

Rates of behavioral health disorders, although lower than those reported by single homeless populations,^2,22,26 also contributed to hospital admissions, ED use, and expenditures for adults and children. Stress related to housing instability, pregnancy concerns, and family disruptions may exacerbate these conditions and disrupt ongoing treatment. The subgroup of adults and children with these conditions seems particularly at risk for hospitalization and high costs during periods of unstable housing. These conditions among adult family members, particularly risky or higher levels of substance use, may also increase families’ vulnerability to becoming homeless and delay their exit from temporary shelter.²⁶ Families with behavioral health conditions are important targets for intervention.

Although there are numerous descriptions of interventions for homeless families, there is limited rigorous research testing specific treatment approaches that target homeless family members. However, a recent study tested an adapted primary care–based collaborative care model for homeless mothers with depression, demonstrating improved use of primary care and case management along with improvements in depression symptoms.²⁷ Another adapted the evidence-based critical time intervention that has been successful with single homeless adults with mental illness for families, demonstrating improvements in children’s mental health and school outcomes.²⁸

ED use, hospital admissions, and expenditures began rising months before the beginning of a homeless episode for many individuals, suggesting that interventions to reduce utilization will be substantially more effective if risk of housing instability can be identified before families are literally homeless. Recent research demonstrates how life events can interact with local housing conditions and individual characteristics to increase the risk of homelessness.^29–31 It seems feasible that information on neighborhood housing markets and individual characteristics could be combined to identify and streamline access to preventive services for families who are at high risk of becoming homeless when faced with challenging life events, such as loss of a job, a relationship breakup, or an unexpected pregnancy. The ability to identify need and deploy resources rapidly is likely a key to the success of any such intervention.

The association between being located outside the Boston area and likelihood of an adult having an ED visit may be attributable to the greater availability of health care alternatives in the Boston area compared with other areas or to other unmeasured social determinants.

Limitations and Strengths

Some features of our data and analysis may limit the generalizability of findings to other areas. In particular, Massachusetts’ Medicaid and housing policies likely provide more extensive coverage and support than do those of other states, especially states with more restricted health care insurance and access. Although we believe our data capture most homeless families, it is likely that a small number did not qualify for EA or MassHealth. These could have different characteristics than did our study population. Although EA benefits are available only to families, EA records are kept at the individual level and could not be reliably aggregated into family groups in many cases, preventing a family-level analysis.

Missing data on shelter exit dates for a number of family members prevented us from precisely identifying any changes in health care utilization that might have been associated with transition to long-term housing. Finally, it is important to note that families qualified for MassHealth benefits at different points in the study. Although most were enrolled in MassHealth before receiving EA benefits, some became eligible at the time of homelessness, thus resulting in a potential underidentification of preexisting health conditions and utilization. Any care reimbursed by a source other than MassHealth was not available for inclusion in our analysis. Despite these concerns, our study has several strengths, including a large, longitudinal sample of homeless family members with detailed information on health utilization and expenditures.

Public Health Implications

Declining expenditures after entering emergency shelter suggest that this assistance helps to lower health care costs. But opportunities for improvement remain. Homeless families begin to rely more heavily on EDs for routine care up to several months before they become literally homeless and eligible for emergency shelter. The most effective strategies for preventing these patterns likely require early detection of families at high risk of becoming homeless, coupled with rapid intervention. Although accurate identification of high-risk families can be difficult, screening to identify women who are pregnant or parenting young children and individuals with behavioral health disorders may be a productive starting point. A number of brief screening instruments for common mental health and substance use problems, some of which may be administered by nonclinicians, can be used as a first step in identifying high-risk groups in health care settings and emergency shelters.³²

The Center for Medicare and Medicaid Innovation is now conducting demonstration projects that include screening for a variety of social determinants of health, including housing stability.^33,34 Evaluations of these tools and interventions may provide further data on optimal ways to identify and intervene on behalf of homeless families. Pregnancy has been shown to increase the risk of homelessness and was the most prevalent health condition observed among adults in our study. Young children were particularly likely to use EDs for routine care and were at greater risk for hospitalization. Screening for housing instability in prenatal care settings as well as EDs might effectively identify a significant proportion of at-risk families at an early stage.

Efforts to prevent homelessness have grown considerably in the past 5 years, although there is limited evidence that specific prevention efforts reduce homelessness among families. Program efforts include, for example, placing housing workers in community health center waiting rooms to assist patients with housing applications and problem solving for housing issues; others train community health workers to detect and address legal issues related to housing instability.²² Shinn et al. developed a promising empirical model using administrative data to help identify families most at risk for homelessness in New York City,⁵ paving the way for more efficiently targeting prevention services.

Clearly, additional research is needed to identify the best ways to identify risk of homelessness at an earlier stage. More research incorporating comparison groups and focusing on the circumstances leading up to a homeless episode is also essential for understanding what drives health care utilization in the early stages of housing instability and, ultimately, developing a more timely response.