Factors Associated With the Health Care Utilization of Homeless Persons FREE

Homeless persons face numerous barriers to receiving appropriate health care. Prior research has documented that homeless persons have high rates of physical illness, mental illness, substance abuse,^{1 - 3} and early mortality.^{4 - 7} They are more likely to be admitted to the hospital⁸ and to have increased length of hospitalization⁹ and may present a substantial burden on the resources of safety-net hospitals and clinics.¹⁰ Despite having a higher burden of illness, homeless persons have fewer encounters with ambulatory care than nonhomeless persons.¹¹ For homeless persons, health care competes with more immediate needs, such as obtaining adequate food and shelter.¹² However, given the opportunity, homeless persons are willing to obtain health care for chronic conditions if they believe such care to be important.¹³

While prior research has described homeless persons' dependence on acute care, such as hospital-based services,^{8 ,14 - 17} the factors associated with utilization of health care in this population have not been explored from a national perspective. Most research has focused on homeless persons residing in specific cities. Wide variations in characteristics and utilization patterns between different regions make national policy implications difficult to discern from local studies. Homelessness has proven to be a durable problem in diverse regions of the country, creating a large toll of human suffering as well as a complex burden on the safety-net health care system. National policy decisions geared toward decreasing barriers to appropriate health care utilization are needed.

We performed a secondary data analysis of the factors associated with the use of ambulatory care services, emergency departments, and acute care hospital services and self-perceived barriers to medical care among currently homeless persons included in the National Survey of Homeless Assistance Providers and Clients (NSHAPC).¹⁸ The survey, performed in October and November 1996 by the Census Bureau, was designed to enroll a nationally representative sample of persons who used homeless services in the United States. The Census Bureau used criteria established in the McKinney Act of 1987 to establish homelessness.¹⁹ Representative areas were chosen for urban, suburban, and rural regions. The 28 largest metropolitan statistical areas were chosen to represent urban areas. Twenty-four areas were randomly sampled from the small- and medium-sized metropolitan statistical areas to represent suburban and urban fringe areas, and 24 groups of rural counties or parts of counties were randomly sampled to represent rural areas.

Subjects were selected among people who used homeless services. The Census Bureau documented and surveyed homeless service providers (hereafter includes homeless assistance programs, including shelters, transitional housing programs, permanent housing programs for fomerly homeless people, food pantries, soup kitchens, etc) in the designated areas and then selected a systematic sample of programs.²⁰ Programs were chosen with a probability proportional to size. Program visits for data collection were scheduled at randomly chosen times and dates, and subjects were randomly selected to complete in-person interviews. Interviews were conducted on-site by census field workers. All subjects were interviewed in person by trained interviewers and were paid $10 for their participation. For our analysis, we excluded formerly and never homeless subjects.

Factors associated with the use of ambulatory care, emergency departments, and acute care hospital services and perceived barriers to health care were analyzed in 5 domains: sociodemographic characteristics, physical health, mental health, substance abuse, and history of homelessness. Sociodemographic characteristics included age (<25 years, 25-49 years, and ≥50 years), sex, race/ethnicity (non-Latino White, non-Latino African American, Latino, or Native American), and veteran status (yes or no). Type of locale (ie, urban, suburban, or rural) was determined by the interviewer based on location of site. Medical insurance (yes/no) was determined by self-report.

Physical health was represented by self-reported chronic comorbidity. Subjects were asked to report if they had diabetes, anemia, hypertension, heart disease or stroke, liver disease, arthritis or joint conditions, cancer, physical disability, or human immunodeficiency virus (HIV) infection or acquired immunodeficiency syndrome (AIDS). The number of positive responses were summed and categorized as 0, 1, 2, and 3 or more.

Mental illness was screened by using a modified version of the Addiction Severity Index (ASI).²¹ According to convention, a score greater than 0.25 on the psychological scale was considered positive for mental illness. The ASI has been validated in homeless populations as a tool for screening for the likelihood of having mental health conditions and drug and alcohol abuse.^{22 - 23} As in the report of the NSHAPC,²⁴ subjects who did not meet the criterion for mental health conditions on the ASI were characterized as having a mental health condition in the past year if they reported mental health service use (ie, medication, hospitalization, or outpatient treatment) or repeated episodes of disabling symptoms. Similarly, alcohol abuse was screened for by the modified ASI (score ≥0.17 in the alcohol realm) by self-reported heavy drinking in the past year or use of alcohol treatment services in the past year. Drug abuse in the past year was screened for (ASI score ≥0.10 in the drug use realm) by self-report of injection drug use, heavy drug use, or treatment for drug abuse in the past year.

Respondents were asked where they had spent the majority of the time since becoming homeless. Responses were categorized as "literally homeless" if they had spent the majority of their time homeless staying on the street or in shelters and as "marginally housed" if they had spent the majority of their time staying in transitional housing, hotel, or a friend's or relative's home.

The outcome variables included 3 service utilization variables and 2 measures of barriers to care. The utilization variables were use of ambulatory care services, emergency departments, and inpatient medical and/or surgical hospital services during the past year. The measures of perceived barriers to care were self-reported inability to receive needed care and the inability to comply with needed prescription medicine in the past year.

Subjects were asked to report all places where they had received medical care in the past year, including inpatient hospitalization and the emergency department. The number of times visited and the length of stay were not obtained. Thus, history of hospitalization and emergency department use were each dichotomized yes or no. The following responses were included as indicating receipt of ambulatory care: hospital outpatient department, Veterans Affairs (VA) outpatient department, physician or nurse in homeless shelter, health care for the homeless clinic, community health clinic, migrant health clinic, or private physician's office. A respondent who did not report having received care at 1 of those sites was considered to not have had an ambulatory care visit in the preceding year. The subject's self-report was not validated by medical records.

Respondents were asked if they had needed care from a physician or nurse in the past year but were unable to obtain it. If they responded affirmatively, they were classified as being unable to receive needed care in the past year. Respondents were asked whether they were supposed to be taking a prescription medication. Only persons who reported being prescribed medication were included in the analysis of compliance; such persons were asked if they had been able to take the medication as prescribed. If they reported no, they were classified as having been unable to comply with prescription medication.

Bivariate relationships between independent variables and the 5 outcome variables were assessed using unadjusted logistic regression models. Multivariable logistic models were chosen based on the importance of bivariate relationships and prior hypotheses. All regression analyses used weights provided by the Census Bureau to make parameter estimates nationally representative. These weights are based on the inverse of the probability of being included in the sample and incorporate information about the complexities of the sampling procedure. They therefore approximate the number of persons in the US population in 1996 that each survey respondent can be taken to represent.²⁴ Because stratum and primary sampling unit identifiers were withheld in the public use dataset to protect participant confidentiality, we used the approximation recommended by the Census Bureau to account for variance inflation resulting from the multistage sampling design. Specifically, the SEs for all estimates (both unadjusted and adjusted) are inflated by the square root of the mean design effect. The design effect estimate provided by the Census Bureau was 3.0.²⁴ In turn, confidence intervals (CIs) and P values were calculated using the inflated SEs.

Finally, because homeless persons with medical comborbidities or prescription drug use may be at greatest risk for barriers to care, we performed a subgroup analysis for the outcome of inability to obtain needed care. Also, to examine whether those who were eligible for insurance were receiving it, we explored the insurance patterns of veterans, who because of their low incomes, were entitled to VA health benefits.²⁵

Of the 4207 respondents to the NSHAPC, 2974 were currently homeless, 688 were formerly homeless, and 545 never homeless. Of the currently homeless subjects, 2806 (94.4%) initially approached for participation completed the interview: 168 of the subjects either refused participation (n = 51), were unable to complete the interview secondary to intoxication (n = 2), were mentally or physically incapable (n = 9), were unavailable (n = 69), or were ineligible for participation (n = 37); replacement subjects were selected. Of the respondents, 50 (1.1%) had a friend or family member assist them with their responses.

We restricted the analyses to the 2974 respondents who were homeless at the time of the survey. The currently homeless population overall is predominantly young (72.6% between 25 and 49 years) and male (67.6%) (Table 1). More than 90% were single, but 42.8% of respondents were married previously. At the time of the survey, 40.3% of women and 3.4% of men were living with a child. More than two thirds of the respondents were located in urban areas. Overall, 22.4% of the respondents were veterans: 32.6% of the men and 1.1% of the women. Of the veterans, 89% reported having received an honorable discharge. The respondents were very poor; less than one fourth of respondents noted income, from all sources, of greater than $600 per month.

The median duration of homelessness for the sample was 1 year; 15.1% were homeless for more than 5 years. Of the respondents, 44.8% were categorized as literally homeless, and the remainder were categorized as marginally housed.

The cohort exhibited high rates of chronic medical health conditions, chronic mental health conditions, and drug and alcohol abuse. Almost three fourths had a mental health condition or drug or alcohol abuse; many reported having more than 1 condition. More than half (55.6%) of the respondents reported being uninsured, and more than one fourth (26.7%) reported no contact with nurses, physicians, clinics, and hospitals in the past year. Among veterans, the uninsured rate was 48.4%; only 26.8% had VA insurance.

Ambulatory Care. Overall, 62.8% of currently homeless respondents reported receiving medical care at an outpatient site, excluding the emergency department, during the past year (Table 2). The most common sites included hospital outpatient clinics (27.9%), VA clinics (5.6%), homeless shelters (with physician or nurse on-site) (10.4%), health care for the homeless clinics (8.3%), community health clinics (22.0%), or a private physician's office (19.4%). In the multivariate analysis, those with 3 or more medical comorbid illnesses and those with medical insurance were more likely to have received ambulatory care in the prior year than those who had less than 3 comorbid illnesses or who were uninsured. There also was a trend toward greater ambulatory care use by women (76.5%) vs men (56.4%) and by those who were categorized as marginally housed (72.7%) vs those who were categorized as literally homeless (51.5%).

Emergency Department Use. Overall, 32.2% of respondents received medical care in an emergency department in the prior year (Table 2). In the multivariate analysis, respondents with mental illness (odds ratio [OR], 1.99; 95% CI, 1.04-3.81) and those with 2 medical comorbidities (OR, 2.59; 95% CI, 1.03-6.51) were significantly more likely to use the emergency department than those without mental illness or comorbidity. Insurance was not associated with emergency department use.

Hospitalizations. Almost one fourth of respondents (23.3%) had a hospitalization in the prior year (Table 2). In the multivariate model, insurance remained the only covariate significantly associated with an increased likelihood of hospitalization (OR, 2.60; 95% CI, 1.16-5.81). Veterans and those with 3 or more medical comorbidities were at an increased risk (OR, 2.02; 95% CI, 0.92-4.44 and OR, 2.59; 95% CI, 0.91-7.37, respectively). African Americans were statistically significantly less likely (OR, 0.47; 95% CI, 0.23-0.94) than non-Latino whites to have had a hospitalization in the prior year.

Ability to Obtain Medical Care When Needed. One fourth of respondents (24.6%) reported that at some point in the past year they had needed medical care that they had not been able to receive (Table 3). After adjustment, those with medical comorbidities were more likely to report an inability to receive needed care than those without comorbidities. Respondents who had insurance were less likely to report poor access. Because those with a medical comorbidity or the need for a prescription medication may be at greater risk for barriers to care, we specifically analyzed the correlates of barriers to care for this subgroup. There were no significant differences between this subgroup and the cohort as a whole.

Ability to Comply With Prescribed Medications. Overall, 40.9% (1201/2936) of respondents stated that they were supposed to be taking a prescription medicine, and 32.6% (392/1201) reported they were unable to comply with these medications. When weighted to reflect the US population, 32.1% noted that they were unable to comply with prescription medications (Table 3). In the multivariate model, respondents aged 50 years and older and respondents with insurance were significantly less likely to report difficulty with medication compliance than younger respondents and the uninsured (P≤.05).

This study confirms on a national scale what previous research found at the local level; homeless persons reported high rates of acute hospital-based care, low rates of ambulatory care, and difficulty accessing health care. Our finding of an association between insurance status and increased use of ambulatory care and decreased barriers to care reflects previous research for housed and homeless persons alike: medical insurance enables the use of ambulatory care and decreases barriers to care.

Despite evidence of poorer health, homeless persons in our study were less likely than the overall US population to report an ambulatory care visit in the previous year. Three fourths of persons with US homes had at least 1 ambulatory visit in the prior year, compared with 63% of homeless persons.²⁶ Consistent with prior research, our study demonstrated that homeless persons are more likely to use emergency departments than the population as a whole.^{27 - 29} Hospitalization rates of homeless persons in this study were 4 times that of the US norms; these results parallel those in other studies.^{8 ,30} In the United States, only persons older than 65 years had similar rates of hospitalization as those in our study, whose median age was 37 years.²⁶

Differences in insurance status may explain part of the disparity in ambulatory care use. In our study, having insurance was associated with increased likelihood of having had an ambulatory care visit. While 17% of the US population was uninsured in 1997,²⁶ 56% of homeless persons were; this is approximately 50% higher than the US poverty population.³¹ Prior studies have found uninsured rates among homeless persons ranging between 30.1% and 71.6%.^{16 ,32 - 33} Among persons with homes, having insurance enables receipt of nonurgent discretionary health care.^{34 - 36} Prior research has suggested that insurance status is correlated with homeless persons' ability to seek nonurgent medical care,^{32 ,37} have a regular site of care that is not the emergency department,¹⁵ and receive inpatient treatment for alcohol and drug abuse disorders and mental illness.³³ In our study, insurance status was not correlated with emergency department use, thus suggesting that use of the emergency department may not be sensitive to insurance status.

We included veterans, who by virtue of their poverty qualify for VA benefits, to analyze whether homeless persons were receiving insurance benefits to which they were entitled. Almost 90% of veterans in our study report having received an honorable discharge, making them theoretically eligible for these benefits. Despite this, almost half of homeless veterans were uninsured; only one fourth had VA insurance. These data support prior research that suggests that a substantial portion of homeless persons remain uninsured despite meeting eligibility criteria.³³

More stable sheltering arrangement, which previously has been found to influence health care utilization,³² was associated with a trend toward greater use of ambulatory care in our study. The association with ambulatory care, reflecting that seen in other studies,¹⁵ suggests that stable housing may improve homeless persons' ability to seek nonurgent care.

Mental illness is associated with health service utilization in both homeless and nonhomeless populations.^{38 - 40} In our study, mental health conditions were correlated with emergency department use but not ambulatory care. Alcohol and drug abuse also are common among homeless persons.^{41 - 42} We found rates of substance abuse comparable with previous studies. In this sample, neither alcohol nor drug abuse was associated with health service use in multivariate models. This may be because our study was unable to look at repeated episodes of health service use: we were unable to ascertain if alcohol and drug abuse place homeless persons at the risk of becoming repetitive users of medical services.

Our study had several limitations. Because subjects were sampled from those using homeless assistance services, they do not represent those who do not use services or those in communities that have few or no homeless assistance services. However, a study of homeless persons in San Francisco found that 88% of those considered literally homeless had used either a homeless shelter or a food line in the previous 30 days.⁴³ The NSHAPC captured a far more extensive range of homeless service providers, suggesting that their methods were likely to capture the vast majority of homeless persons in communities where there were services provided.

The survey did not acquire information on the frequency of use of medical services. For example, we could not distinguish people with 1 hospitalization from those with several. This limited our ability to ascertain whether certain subgroups were at risk of becoming frequent users of medical care; prior research has suggested that those with substance use or mental health conditions are likely to fall in this category.²⁹

All data were by respondent self-report without independent confirmation. Research on self-report of homeless persons has suggested that while they are reasonably accurate reporters of the use of health care services in the past year, they are not accurate reporters of visit frequency.⁴⁴ Prior research that validated homeless persons' self-report of health care utilization found that they were only slightly less accurate in their reporting than the population at large.^{32 ,44 - 45}

While many studies document that insurance facilitates the use of health care services, it is possible that the associations observed in these data are not causal. Clinics and hospitals may facilitate health coverage reimbursement among eligible patients under their care. We were unable to distinguish whether the association between insurance and compliance with medications reflected insured persons' improved ability to purchase medication, differences in the patient-physician encounter, or unmeasured differences in homeless persons with and without insurance.

Finally, since our study was descriptive, multiple comparisons were made, thus increasing the potential for type I errors. Accounting for the effects of multistage sampling using an estimate of the mean design effect reduces the reliability of CIs for logistic model ORs. Since actual design effects may vary by both independent and dependent variables, use of the mean in all situations could have liberal or conservative effects. This estimate, however, is the most accurate available.

Our study provides a national perspective on the health care needs and utilization of homeless persons. Having insurance was consistently associated with indicators of improved access to care; however, the majority of respondents lacked insurance. Nearly half of veterans were uninsured, despite meeting criteria for VA health benefits. These data suggest that improving insurance rates among homeless persons will require not only expansion of eligibility but also improvements in identifying and enrolling those who qualify. The provision of medical insurance may favorably impact homeless persons' ability to obtain ambulatory health care and decrease the tendency among homeless persons to delay needed care. This, in turn, may contribute to decreasing high rates of emergency department use and inpatient hospitalizations and may improve morbidity among homeless persons.