Clinical and Demographic Factors Associated With Homelessness and Incarceration Among VA Patients With Bipolar Disorder

Abstract

Objectives. We assessed the association between homelessness and incarceration in Veterans Affairs patients with bipolar disorder.

Methods. We used logistic regression to model each participant's risk of incarceration or homelessness after we controlled for known risk factors.

Results. Of 435 participants, 12% reported recent homelessness (within the past month), and 55% reported lifetime homelessness. Recent and lifetime incarceration rates were 2% and 55%, respectively. In multivariate models, current medication adherence (based on a 5-point scale) was independently associated with a lower risk of lifetime homelessness (odds ratio [OR] = 0.80 per point, range 0–4; 95% confidence interval [CI] = 0.66, 0.96), and lifetime incarceration increased the risk of lifetime homelessness (OR = 4.4; 95% CI = 2.8, 6.9). Recent homelessness was associated with recent incarceration (OR = 26.4; 95% CI = 5.2, 133.4). Lifetime incarceration was associated with current substance use (OR = 2.6; 95% CI = 2.7, 6.7) after control for lifetime homelessness (OR = 4.2; 95% CI = 2.7, 6.7).

Conclusions. Recent and lifetime incarceration and homelessness were strongly associated with each other. Potentially avoidable or treatable correlates included current medication nonadherence and substance use. Programs that better coordinate psychiatric and drug treatment with housing programs may reduce the cycle of incarceration, homelessness, and treatment disruption within this vulnerable patient population.

Bipolar disorder, characterized by alternating manic and depressive episodes (manic depression), is a chronic mental illness associated with significant functional and social impairments as well as poor overall health outcomes. The World Health Organization has ranked bipolar disorder among the top 10 conditions associated with quality-adjusted life-year decrements in the Global Burden of Disease report.¹ Up to 5.5% of the US population has bipolar disorder.²

Persons with bipolar disorder exhibit unique symptoms that can worsen public health outcomes, including homelessness and incarceration.³ Intermittent manic episodes can lead to medication nonadherence, risky behaviors (e.g., substance use to sustain highs experienced during manic episodes), social consequences, and interaction with the legal system. Some risk factors for homelessness occur disproportionately in persons with bipolar disorder, such as substance-use disorders,⁴ which affect up to 72% of those with this condition. Individuals with bipolar disorder are more likely to be incarcerated than those with other mental disorders.⁵ They also experience high residential instability⁶ and are likely to leave supported housing earlier than persons with schizophrenia, schizoaffective disorder, or depression, which further exacerbates problems with nonadherence and substance use.⁷ Moreover, persons with bipolar disorder are better educated than individuals with other chronic mental illnesses, which leads to greater social, occupational, and economic losses when their illness is not adequately managed.⁸

Although mental illness and substance-use disorders generally are associated with increased risk of homelessness,^9,10 the interplay between homelessness and incarceration among persons with mental illness has not been fully explored, particularly for those with bipolar disorder. Incarceration is strongly and bidirectionally associated with homelessness in persons with mental disorders.^11,12 Loss of a stable home can interfere with treatment retention⁷ and is associated with poor medication adherence,¹³ which can facilitate another cycle of incarceration, homelessness, and symptom exacerbation. Incarceration is also associated with job loss, housing instability, and lost health care benefits.¹⁴ This jail–homelessness cycle was described by Hopper et al.,¹⁵ who referred to an “institutional circuit” in which socially excluded populations were shifted from place to place and where potentially manageable clinical aspects of mental illness were complicated by social determinants of the patients' situations (e.g., poverty).

Studies have pointed to the need to examine treatable contributors to the jail–homelessness cycle, such as symptoms, adherence, and substance use,¹⁶ but these studies have primarily focused on administrative data and have used little behavioral, clinical, or socioeconomic information. There is also little information on the risk factors associated with homelessness and incarceration among veterans. Approximately 40% of homeless individuals in the United States are veterans,^17–19 although only 8% of the overall US population are veterans. In fiscal year 2005, 1% of the users of the Veterans Administration (VA) health care system were homeless, but 7% of VA mental health care patients were homeless, per our queries of VA databases. Among homeless veterans, 41% in one study and 70% in another reported using VA health care services,^20,21 which makes the VA health care system a promising contact point for interventions targeting homeless veterans.

A better understanding of the associations among treatment and behavioral factors, homelessness, and incarceration could inform the development and implementation of interventions that could interrupt the jail–homelessness cycle and improve patients' outcomes. To that end, we conducted this study to: (1) estimate current and lifetime incidence of homelessness and incarceration among VA patients with bipolar disorder, (2) explore factors related to homelessness and incarceration, and (3) determine the variance in each condition that was accounted for by the other condition. We hypothesized that the risks of homelessness and incarceration would be increased among those with substance use, poor adherence, and low therapeutic alliance, after controlling for sociodemographic and clinical factors. We chose to focus on these potentially treatable factors in light of studies that have suggested that improving adherence,^22,23 reducing substance use,²⁴ and strengthening therapeutic alliances²³ led to improved remission rates and outcomes in patients with bipolar disorder and other mental disorders.

METHODS

Participants were veterans with bipolar disorder who were recruited into the Continuous Improvement for Veterans in Care–Mood Disorders (CIVIC-MD) study.²⁵ The purpose of CIVIC-MD was to identify modifiable individual and treatment factors associated with adverse outcomes among patients with bipolar disorder. CIVIC-MD aimed to enroll a more generalizable cohort of patients than is available from most randomized, controlled trials, which tend to rely on more restrictive inclusion and exclusion criteria. Patients receiving inpatient or outpatient care for bipolar disorder were sequentially recruited from a large, urban VA mental health facility in the northeastern United States from July 2004 to July 2006. Patients gave informed consent and completed a survey of current symptoms, treatment preferences, and health behaviors.

Inclusion criteria were current diagnosis of bipolar disorder, cyclothymia, or schizoaffective disorder–bipolar subtype, based on chart review and confirmatory diagnosis from their providers. Patients were excluded if they had an unstable medical condition or a significant cognitive impairment preventing provision of informed consent or completion of the baseline survey. Of 720 eligible patients approached for the study, 435 completed the survey. Exclusions occurred for the following reasons: provider determined that the patient was unable to give consent (n = 104), the patient refused (n = 148), and the patient enrolled in study but did not complete survey (n = 33).

Measures

Surveys assessed homelessness, incarceration, sociodemographic characteristics (age, race, gender, education, marital status, employment status), health behaviors (treatment adherence, substance use), treatment factors (beliefs, therapeutic alliance), and current bipolar episode (manic, mixed, depressed).

Surveys assessed 2 aspects of homelessness: (1) being without a permanent home and (2) staying overnight in a shelter, park, or abandoned building, or on the street. For each aspect of homelessness, respondents were asked whether the situation had occurred (1) in the past 4 weeks and (2) ever. For each time frame, the aspects of homelessness were combined, creating 2 indicators of homelessness (homeless in the past 4 weeks, ever homeless). Similarly, 2 items assessed experiences of being in jail or prison, also during the past 4 weeks or ever.

Treatment adherence was assessed with the Morisky index of 4 questions asking about forgetting to take medications, being careless about medications, stopping medications when feeling better, and stopping medications when feeling worse.²⁶ These items were reversed and then summed, with higher scores indicating better adherence.

On the basis of self-reported alcohol use, we defined hazardous drinking as usually consuming 5 or more drinks per single occasion. This criterion is derived from a question in the Alcohol Use Disorders Identification Test.²⁷ This question on usual quantity consumed correlates strongly with hazardous drinking as defined by the full Alcohol Use Disorders Identification Test²⁸ and has long been used in studies to track binge drinking.^29,30 We assessed substance use by asking whether respondents had used any of the following drugs in the previous year: marijuana, cocaine, stimulants, or opioids.

An instrument adapted from Meredith et al.^31,32 posed 4 questions to assess belief in the likelihood that medications would be effective in achieving treatment goals. Higher scores indicated stronger belief in the benefits of treatment. Patients rated the likelihood that medication would (1) relieve manic or depressive episodes; (2) improve social, family, or job functioning; (3) minimize risk of treatment side effects; or (4) prevent recurrence of manic or depressive episodes. Each item was rated on a scale of 0 to 4 (“very unlikely” to “very likely”). Items were summed to generate scores ranging from 0 to 16. This measure correlated positively with medication adherence among bipolar patients.³²

Therapeutic alliance was assessed by means of the Health Care Climate Questionnaire, which was developed for patients with bipolar disorder to assess their comfort with mental health treatment. Sample items include “I feel understood by my mental health team” and “I am encouraged to ask questions about my treatment.” Items were rated on a 7-point Likert scale (strongly disagree to strongly agree) and summed. The Health Care Climate Questionnaire correlated with patient satisfaction (r = 0.63; P < .001) and with the structured clinical interview for The Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition³³ in the development study³⁴ and has been demonstrated to be associated with medication adherence among bipolar patients.³⁵

Current bipolar disorder episode was derived from the Internal State Scale, a 15-item measure of bipolar disorder symptoms.³⁶ Rated on a 100-mm long visual analog scale (with anchors designated “Not at all” and “Very much of the time”), sample items include “Today I feel great inside” and “Today I feel depressed.” Items formed 4 subscales: well-being (3 items), activation (5 items), perceived conflict (5 items), and depression (2 items). Scoring high on both activation (≥ 155) and well-being (≥ 125) indicated active mania, scoring low on both activation (< 155) and well-being (< 125) indicated current depression, and scoring high on activation (≥ 155) and low on well-being (< 125) denoted current mixed state. These measures have been validated against clinical diagnoses of manic, depressive, mixed, and euthymic mood by psychiatrists expert in the diagnosis and treatment of bipolar disorder.³⁷

Analysis

Descriptive data are presented for sociodemographic characteristics, behaviors, treatment factors, current bipolar episode, homelessness (past 4 weeks, ever), and incarceration (past 4 weeks, ever). We conducted bivariate analyses that used logistic regression to determine the unadjusted association between each patient factor and outcomes for homelessness and incarceration. We used multivariate logistic regression to examine factors predicting incarceration (ever), first without homelessness (ever) as an independent variable and then in a full model including homelessness. The small number of persons experiencing recent incarceration precluded multivariate analysis of that outcome; VA privacy rules prohibit reporting on groups of fewer than 11 persons.

We used a similar approach to assess the impact of incarceration on models of homelessness (separately for recent and lifetime homelessness). We assessed model fit by means of the C statistic, a measure of the predictive ability of the model in which 0.50 is no different from chance. We used the Bonferroni method to adjust criterion α for significance to 0.0125. We used SAS version 9.1 (SAS Institute Inc, Cary, NC) to complete analyses. There were very few missing data (11 participants deleted in multivariate models), so no data were imputed.

RESULTS

Patients (n = 435) averaged 49 years of age (SD = 11; range 21–78); 14% were women and 13% were African American. Nationally, VA patients with bipolar disorder have characteristics comparable to those of this population (mean age = 52; SD = 12; 13% women, 10% African American).³⁸ Most patients in the sample were male (86%), White (77%), unmarried (70%), and impoverished (60% reported an annual income of less than $20 000), although they were not uneducated; 17% were college graduates, and another 49% had some college education (Table 1). Substance use (28%) was somewhat more common than binge drinking (21%). The prevalences of recent and lifetime homelessness were 12% and 55%, respectively. About 2% of respondents had been incarcerated in the past 4 weeks, and 55% of respondents reported having been incarcerated at some time in their lives.

TABLE 1.

Characteristics of VA Patients With Bipolar Disorder (N = 435): Continuous Improvement for Veterans in Care–Mood Disorders Study, 2004-2006

Patient Characteristics	Sample size, No.	Mean (SD) or %
Age, mean, y	435	49.4 (10.6)
Medication adherence score, (range 0–4)	431	2.43 (1.35)
Therapeutic alliance score, (range 0–60)	430	39.4 (15.0)
Ethnicity
White	336	77.3
African American	58	13.3
Other	41	9.4
Education
College graduate	75	17.3
Some college	214	49.4
Less than college	144	33.3
Income
< $10 000	134	31.6
$10 000–$19 999	120	28.3
$20 000–$29 999	74	17.5
$30 000–$39 999	52	12.3
≥ $40 000	44	10.4
Married	131	30.2
Living alone	154	35.4
Current manic episode	126	29.1
Any affective disorder	240	55.2
Current alcohol, tobacco, and substance use
Binge drinking	93	21.4
Substance use	123	28.3
Smoking	267	61.4
Homelessness and incarceration
Homeless, previous 4 weeks	53	12.2
Homeless, ever	237	54.6
Incarcerated, previous 4 weeks	< 11	∼2
Incarcerated, ever	239	55.1

Homelessness and incarceration were associated with each other for the recent measures (odds ratio [OR] = 16.1; 95% confidence interval [CI] = 3.9, 66.5) and the lifetime measures (OR = 4.7; 95% CI = 3.1, 7.1). A number of potentially treatable conditions displayed bivariate associations with lifetime homelessness, including binge drinking and substance use, and medication adherence displayed an inverse association (all P < .001). Recent homelessness was associated with binge drinking (P < .001), substance use (P < .001), African American race (P < .001), and unemployment (P < .01), and it was inversely associated with adherence (P < .01). Binge drinking, drug use, and being unmarried were significantly associated with ever being incarcerated (all P < .001) and were inversely associated with adherence (P < .01). In bivariate models, recent incarceration was significantly related only to binge drinking (P < .05); there were insufficient cases of recent incarceration for multivariate analysis.

In the first multivariate logistic regression, the relative odds of lifetime homelessness were greater for unmarried individuals (adjusted OR = 2.3; 95% CI = 1.5, 3.6; P < .001), and the odds decreased by about 20% per point increase (i.e., with improvement) in Morisky adherence score (OR = 0.78; 95% CI = 0.66, 0.92; P < .01). When lifetime incarceration status was included in the model, the impacts of unmarried status and adherence changed little (unmarried OR = 2.0, 95% CI = 1.3, 3.2; P < .01; adherence OR = 0.80; 95% CI = 0.66, 0.96; P = .013), and lifetime incarceration was significant (OR = 4.4; 95% CI = 2.7, 6.9; P < .001). Binge drinking and substance use were not significant factors. The fit of the model was better when incarceration was included (C = 0.77 vs C = 0.70; Table 2).

TABLE 2.

Risk Factors for Lifetime Homelessness Among VA Patients With Bipolar Disorder (N = 435): Continuous Improvement for Veterans in Care–Mood Disorders Study, 2004–2006

Effect	OR (95% CI)	P
Model without lifetime incarceration indicator (C statistic = 0.70)
Age	0.99 (0.97, 1.02)	.599
Binge drinking	1.40 (0.81, 2.44)	.233
Substance use	1.62 (0.98, 2.67)	.058
Current episode
Manic episode	1.37 (0.78, 2.40)	.275
Mixed state	1.38 (0.77, 2.48)	.282
Depressed	1.09 (0.55, 2.14)	.811
Unmarried	2.30 (1.46, 3.61)	< .001
African American	2.02 (1.02, 4.01)	.043
Female	0.73 (0.40, 1.33)	.298
Unemployed	1.29 (0.80, 2.10)	.303
Medication adherence score	0.78 (0.66, 0.92)	.004
Therapeutic alliance score	0.99 (0.98, 1.01)	.346
Meredith medication beliefs score	1.00 (0.94, 1.07)	.924
Model with lifetime incarceration indicator (C statistic = 0.77)
Age	0.99 (0.97, 1.01)	.309
Binge drinking	1.20 (0.67, 2.15)	.532
Substance use	1.17 (0.69, 1.98)	.559
Current episode
Manic episode	1.56 (0.86, 2.84)	.145
Mixed state	1.54 (0.83, 2.85)	.174
Depressed	1.36 (0.66, 2.81)	.401
Unmarried	2.01 (1.25, 3.23)	.004
African American	1.89 (0.93, 3.86)	.081
Female	1.16 (0.61, 2.22)	.651
Unemployed	1.20 (0.72, 2.00)	.488
Medication adherence score	0.80 (0.66, 0.96)	.013
Therapeutic alliance score	0.99 (0.98, 1.01)	.483
Meredith medication beliefs score	1.00 (0.93, 1.08)	.975
Ever incarcerated	4.36 (2.73, 6.94)	< .001

Note. OR = odds ratio; CI = confidence interval.

In the second multivariate model, African American patients were more likely to have been recently homeless (OR = 3.1; 95% CI = 1.4, 6.7; P < .01). Binge drinking (OR = 2.1; 95% CI = 1.0, 4.2) and being unemployed (OR = 3.0; 95% CI = 1.2, 7.8) had effects that were not significant per Bonferroni correction (P = .04 and P = .02, respectively). In the model including recent incarceration as a predictor, African American race remained associated with recent homelessness (OR = 3.7; 95% CI = 1.6, 8.2; P = .001; Table 3). Recent incarceration had a large effect on recent homelessness (OR = 26.4; 95% CI = 5.2, 133.4; P < .001), and adding this factor also improved model fit (from C = 0.76 to C = 0.80).

TABLE 3.

Risk Factors for Recent Homelessness Among VA Patients With Bipolar Disorder (N = 435): Continuous Improvement for Veterans in Care–Mood Disorders Study, 2004–2006

Effect	OR (95% CI)	P
Model without recent incarceration indicator (C statistic = 0.76)
Age	0.99 (0.96, 1.03)	.679
Binge drinking	2.06 (1.01, 4.22)	.047
Substance use	1.77 (0.87, 3.63)	.116
Current episode
Manic episode	1.06 (0.41, 2.71)	.906
Mixed state	1.76 (0.70, 4.45)	.233
Depressed	1.07 (0.32, 3.50)	.917
Unmarried	1.68 (0.75, 3.77)	.204
African American	3.05 (1.39, 6.70)	.006
Female	0.51 (0.18, 1.45)	.206
Unemployed	3.02 (1.17, 7.80)	.022
Medication adherence score	0.86 (0.66, 1.11)	.236
Therapeutic alliance score	0.99 (0.96, 1.01)	.169
Meredith medication beliefs score	1.07 (0.96, 1.20)	.201
Model with recent incarceration indicator (C statistic = 0.80)
Age	1.00 (0.96, 1.03)	.869
Binge drinking	1.73 (0.82, 3.66)	.152
Substance use	1.93 (0.92, 4.04)	.082
Current episode
Manic episode	1.02 (0.39, 2.69)	.969
Mixed state	1.72 (0.66, 4.54)	.270
Depressed	0.90 (0.25, 3.27)	.878
Unmarried	1.50 (0.66, 3.39)	.333
African American	3.67 (1.64, 8.21)	.002
Female	0.44 (0.15, 1.34)	.148
Unemployed	2.91 (1.09, 7.72)	.032
Medication adherence score	0.84 (0.65, 1.10)	.201
Therapeutic alliance score	0.98 (0.96, 1.00)	.088
Meredith medication beliefs score	1.06 (0.94, 1.19)	.337
Recent incarceration	26.41 (5.23, 133.4)	< .001

Note. OR = odds ratio; CI = confidence interval.

In the third logistic regression model (Table 4), the relative odds of lifetime incarceration were greater for respondents reporting substance use (OR = 2.8; 95% CI = 1.7, 4.6; P < .001) or unmarried status (OR = 2.0; 95% CI = 1.3, 3.1; P < .01) and less for females (OR = 0.21; 95% CI = 0.11, 0.41; P < .001), with a C statistic of 0.73. When lifetime homelessness was added to the model, it was significant (OR = 4.3; 95% CI = 2.7, 6.8) and once again improved the fit of the model (C = 0.78).

TABLE 4.

Risk Factors for Lifetime Incarceration Among VA Patients With Bipolar Disorder (N = 435): Continuous Improvement for Veterans in Care–Mood Disorders Study, 2004-2006

Effect	OR (95% CI)	P
Model without homelessness indicator (C statistic = 0.73)
Binge drinking	1.56 (0.89, 2.76)	.121
Substance use	2.78 (1.66, 4.64)	< .001
Current episode
Manic episode	0.71 (0.40, 1.27)	.249
Mixed state	0.79 (0.43, 1.44)	.439
Depressed	0.54 (0.27, 1.08)	.082
Unmarried	1.97 (1.25, 3.11)	.004
African American	1.43 (0.73, 2.81)	.296
Female	0.21 (0.11, 0.41)	< .001
Unemployed	1.46 (0.90, 2.35)	.126
Medication adherence score	0.86 (0.73, 1.03)	.096
Therapeutic alliance score	0.99 (0.98, 1.01)	.450
Meredith medication beliefs score	1.00 (0.94, 1.07)	.963
Model with homelessness indicator (C statistic = 0.78)
Binge drinking	1.45 (0.81, 2.62)	.213
Substance use	2.60 (1.52, 4.44)	< .001
Current episode
Manic episode	0.61 (0.33, 1.12)	.110
Mixed state	0.67 (0.35, 1.27)	.217
Depressed	0.48 (0.23, 1.00)	.049
Unmarried	1.56 (0.95, 2.54)	.077
African American	1.17 (0.57, 2.37)	.673
Female	0.20 (0.10, 0.40)	< .001
Unemployed	1.39 (0.84, 2.29)	.204
Medication adherence score	0.92 (0.77, 1.11)	.384
Therapeutic alliance score	1.00 (0.98, 1.01)	.628
Meredith medication beliefs score	1.00 (0.93, 1.07)	.999
Ever homeless	4.24 (2.67, 6.71)	< .001

Note. OR = odds ratio; CI = confidence interval.

DISCUSSION

Homelessness and incarceration are common among VA patients with bipolar disorder and share many risk factors. Homelessness and incarceration have a bidirectional relation: homelessness can lead to loitering or criminal acts and then to arrest, and release from incarceration can turn someone out onto the street with nowhere to go. Among VA patients with bipolar disorder, we found that 12% reported having been homeless in the previous 4 weeks, which is comparable to the homelessness rates of 13% to 17% reported among persons with chronic mental illness.^16,38 Slightly more than half of our sample reported ever having been homeless (55%) or incarcerated (55%), suggesting that this group is at high risk of having unstable treatment courses and poor outcomes. We also found that lifetime experience of homelessness was associated with 4-fold increased odds of lifetime experience of incarceration, and that recent homelessness was strongly related to recent incarceration.

Homelessness can have devastating consequences for persons with mental illness by making transportation, self-care, access to health care, and nutrition more difficult. Although lifetime incarceration was one of the strongest correlates of lifetime homelessness, current medication adherence was also independently associated with a lower risk of lifetime history of homelessness, even after we controlled for incarceration history. Nonadherence is likely to occur repeatedly over a patient's lifetime; thus, past nonadherence may be the unmeasured predictor for both lifetime homelessness and current nonadherence. Nonadherence may lead to manic episodes and consequent risky or disruptive behaviors, causing loss of job or eviction that can result in homelessness. Although the cross-sectional nature of this study prevented us from examining manic episodes or incarceration events over time, this finding suggests that adherence interventions may reduce the risk of homelessness among veterans with bipolar disorder.

Incarceration can also have severe negative effects on persons with mental illness by reducing employability, impeding personal relationships, and interrupting the continuity of health care. Mental health services are often minimal in correctional facilities, so there is a great need to intervene with patients at risk for incarceration. In this study, homelessness was the factor most strongly associated with incarceration among veterans with bipolar disorder. We also found that, among potentially treatable factors, current substance use was associated with lifetime incarceration history (after we controlled for other factors, including homelessness), which suggests that persons with incarceration histories have a special need for preventive services and proactive assessment for possible substance use. Failure to assess and treat comorbid substance use undercuts treatment of bipolar disorder.

In recognition of the many issues faced by veterans with mental illness, the VA has initiated a number of programs to assist this group.^39–42 In many communities, the VA is already working with local police to divert veterans into VA mental health care in lieu of incarceration, which may increase intervention opportunities. In addition, the VA is currently developing a transition assistance program for incarcerated veterans that is intended to prevent homelessness by giving information about health care, housing assistance, and employment services to veterans nearing release from jail or prison.⁴³ Although no single strategy can improve quality of life for all veterans with bipolar disorder, the many factors identified here and in other studies represent numerous opportunities to address a complex problem.

Few interventions have been specifically tailored to the unique risk factors associated with bipolar disorder. Persons with bipolar disorder are prone to adverse outcomes, particularly because of manic episodes, which lead to treatment nonadherence and self-medication through substance use. These behaviors subsequently increase risky behaviors leading to incarceration, and because incarceration may interrupt treatment course, reduce employment, and cause domicile loss, persons with bipolar disorder are then at risk for homelessness. The relatively few effective programs for this group have focused on improving treatment adherence⁴⁴ and reducing substance use.²⁴ Brief but effective psychoeducation (i.e., patient education) programs focused on treatment adherence, substance use reduction, and symptom management may be the most adaptable to existing programs for veterans that focus on jail diversion or housing.^44,45

Limitations

Although this cross-sectional study was unable to identify repeat offenders or temporal associations, and although self-report data may have distorted some objective experiences, we believe the associations we found are recursive among homelessness, substance use, nonadherence, and incarceration. Longitudinal data might permit causal relationships to be assessed, and we encourage future research into this challenging modeling problem.

Other limitations of this work include: (1) our northeastern–United States urban sample may not represent all VA patients adequately, although it was demographically similar to the population of VA patients with bipolar disorder; (2) these findings should be generalized to non-VA patient groups cautiously, keeping in mind the sharp difference in gender distribution between VA and non-VA populations; (3) the data may not reflect experiences of community-based samples of persons with bipolar disorder (vs clinic-based samples); (4) incarceration was associated with reduced likelihood of using VA services in a previous well-constructed study,⁴⁶ so our sample may be biased downward in terms of representing VA patients with a history of incarceration; and (5) illness course could have an important impact on these study variables, but we only assessed current psychiatric status.

Conclusions

Homelessness and incarceration are public health problems that can disrupt treatment course and lead to poor outcomes. In our study, incarceration and homelessness were strongly and bidirectionally associated among veterans with bipolar disorder. Factors associated with incarceration and homelessness, such as current medication nonadherence and substance use, may be addressed through treatment efforts, although it would be preferable to reduce this association by preventing homelessness and incarceration. Additional significant correlates were unmarried status (associated with lifetime homelessness), African American race (associated with recent homelessness), and male gender (associated with lifetime incarceration).

These demographics sketch out a highly at-risk population, and a demographic well represented in the VA healthcare system: VA patients are more likely to be non-White, and are poorer and less healthy, than the general US population.⁴⁷ Interventions that focus on psychiatric treatment, substance use treatment, and housing stability may be required to interrupt the jail–homelessness cycle among veterans with bipolar disorder.