Suicide is the 10th leading cause of death in the United States, and national suicide rates have risen 35% since 1999 (
1). Increased attention has been devoted to identifying subpopulations at high risk of suicide, with emerging evidence that homeless adults represent an understudied high-risk group that shares risk factors such as poverty and serious psychiatric and substance use disorders (
2–
6).
Recent findings on suicidal behavior among homeless adults have come from several studies of U.S. veterans, because both suicide and homelessness are high-priority areas for preventive intervention by the Veterans Health Administration (VHA) (
2,
7). A survey of a nationally representative sample of U.S. veterans found that those with a lifetime history of homelessness had five times (6.9% versus 1.2%) the rate of suicide attempts in the past 2 years, compared with other veterans (
7). A study of VHA patients showed completed suicide to be twice as frequent among formerly homeless veterans, compared with other veterans (
8). A study of younger veterans found a hazard ratio for death by suicide of 2.7 among those who had ever been homeless, compared with their nonhomeless counterparts (
9), and an examination of suicide attempts among both homeless veterans and homeless nonveterans found elevated rates in both groups (
2).
Although these studies focused on veterans with any history of homelessness, a systematic review suggested that among currently homeless adults, including both veterans and nonveterans, recent and lifetime suicide attempts were elevated, at 9.2% and 28.8%, respectively (
10), thus linking suicide attempts to recent or current homelessness. In a large sample of 10,000 homeless veterans, 3% reported a suicide attempt in the previous month (
11), consistent with another recent review (
12).
It has thus been well demonstrated that homelessness is associated with suicidal behavior. However, the temporal relationship between homelessness and suicide attempts is not well understood. Current homelessness may lead to increased risk of suicide attempts during or after the episode of homelessness, or vice versa. One recent study using VHA administrative records reported that documentation of suicidality appeared to peak immediately prior to the first indication of homelessness (
13). As interest grows in both suicide prevention and homeless services programs, understanding this temporal association may have important implications for informing timely and focused interventions. We sought to replicate the findings of this VHA study by using nationally representative self-report data from both veterans and nonveterans.
In this study, we used national survey data from the National Epidemiologic Survey on Alcohol and Related Conditions–III (NESARC-III) (
14) to compare sociodemographic and clinical characteristics associated with past and recent suicide attempts among adults with past-year homelessness, homelessness prior to the past year, and no history of homelessness. We first identified differences in lifetime rates of suicide attempts among those with histories of homelessness across the entire sample. We then compared the timing and frequency of suicide attempts as well as other sociodemographic and clinical characteristics among those with past attempts across the three homelessness groups. We focused on the timing of their most recent attempt (and especially attempts in the past year), the timing of their first lifetime attempt, and the total number of lifetime reported attempts. We thus aimed to expand the available information on the association of homelessness and suicide attempts.
Methods
Data Source and Study Sample
Data came from the restricted version of the NESARC-III, sponsored by the National Institute on Alcohol Abuse and Alcoholism (
14), a nationally representative cross-sectional survey of noninstitutionalized civilian adults ages 18 and older, conducted between April 2012 and June 2013 (
15). Respondents were selected through multistage probability sampling, with Blacks, Asians, and Hispanics oversampled. In-person structured interviews excluded individuals who were institutionalized (e.g., nursing homes, prisons, hospitals, or shelters). The overall response rate was 60.1%, and the total original sample of NESARC-III included 36,309 U.S. adults (
15). Data were adjusted for oversampling and nonresponse, then weighted to represent the U.S. civilian population based on the 2012 American Community Survey (
16). The analytic sample included respondents who answered questions about past homelessness and lifetime suicide attempts, for a total sample of 36,127 (99.5% of the original sample, representing 235.4 million adults). Informed consent was electronically recorded, and respondents received $90 for participation. Institutional review boards (IRBs) at the U.S. National Institutes of Health and at Westat approved the study protocol. This study was approved by the IRBs of the Department of Veterans Affairs Connecticut Healthcare System and Yale School of Medicine.
Measures
Homelessness.
A three-level homelessness variable was created that identified adults with homelessness in the past year, adults with homelessness prior to the past year, and adults who reported no history of homelessness. Lifetime homelessness was assessed with one question: “Since you were 15, did you have a time that lasted at least 1 month when you had no regular place to live—like living on the street or in a car?” A separate question asked, “In the last 12 months, have you at any time been homeless?”
Suicide.
Lifetime suicide attempts were assessed with the question, “In your entire life, did you ever attempt suicide?” followed by questions addressing age at first attempt, age at the most recent attempt, and the total number of lifetime attempts. Years since the first and last suicide attempt were calculated by subtracting the respondent’s reported age at the time of these attempts from his or her current age. A dichotomous variable represented any suicide attempt in the past year.
Sociodemographic characteristics.
Sociodemographic information included age, gender, race-ethnicity, marital status, annual income, education, rurality of residence, military service history, and health insurance coverage.
Clinical characteristics.
With the Alcohol Use Disorder and Associated Disabilities Interview Schedule–5, which is based on criteria from the
DSM-5 (
17), information was collected on lifetime and past-year psychiatric disorders, including major depressive disorder, dysthymia, bipolar I disorder, generalized anxiety disorder, posttraumatic stress disorder (PTSD), and panic disorder. Using these variables, we constructed a dichotomous variable representing any past-year psychiatric disorder and a second measure of the total number of psychiatric disorders (reflecting behavioral multimorbidity). Information on past-year substance use disorders (i.e., alcohol use disorder; tobacco use disorder; and drug use disorders, including sedative, cannabis, cocaine, stimulant, and heroin use disorders) was also assessed by using
DSM-5 criteria and a measure of the total number of such disorders.
Data Analysis
Data analysis proceeded in the following steps. First, with data from the sample with all relevant data (N=36,127), weighted frequencies and cross-tabulations were computed to examine differences in the proportion of adults in each of the three homeless groups who had ever made a suicide attempt in order to evaluate previous findings of greater proportions with suicide attempts among adults with past homelessness.
Subsequent analyses included only those who had reported past suicide attempts. Weighted chi-square tests and analysis of variance were used to compare adults who had ever made a suicide attempt across the three groups on sociodemographic and clinical characteristics, including years since the most recent suicide attempt, years since the first suicide attempt, and number of lifetime suicide attempts.
Because current age could be an important confounder of these comparisons, we further compared the three homelessness groups on the mean number of years since first and most recent suicide attempts, controlling for current age by using analyses of covariance.
To further evaluate the potential confounding effect of age, we conducted a logistic regression analysis in which past-year suicide attempt was the dependent variable and dichotomous variables representing those with past-year and prior to past-year homelessness were independent variables (with the never homeless group as the reference condition) and current age as a covariate. Because of the focus of previous research on veterans, we repeated this analysis, limiting the sample to veterans with lifetime suicide attempts.
In the final analysis to identify independent correlates of past-year suicide attempts in addition to homelessness, multivariable-adjusted logistic regression analysis included dichotomous variables representing those with past-year and lifetime homelessness, as well as sociodemographic and clinical variables that were significantly different between the homeless groups in our initial comparison.
The surveymeans, surveyfreq, and surveylogistic procedures in SAS, version 9.3, were used to compute variances that accurately reflected complex sample design and estimation procedures. For all analyses, poststratification weights were applied, and p<0.05 was used as the test of statistical significance.
Results
Sample Characteristics
In the entire NESARC-III sample (N=36,127 unweighted, representing 235.4 million adults nationwide), suicide attempts were reported by 24.3% (N=160) of those with past-year homelessness and 21.2% (N=316) of those with homelessness prior to the past year, substantially greater proportions than the 4.3% (N=1,516) of those who had never been homeless (p<0.001). Our analytic subsample included the 1,992 respondents (5.5% of the entire sample) who reported a lifetime suicide attempt. This sample had a mean±SD age of 42.3±13.89 years, and most were female (N=1,325, 66.5%), white (N=1,410, 70.8%), and never married, separated, or divorced (N=1,243, 62.4%). In this sample, 7.8% (N=155) were veterans.
Characteristics of Homeless Groups
The group that had been homeless in the past year, compared with the other two groups, was younger, more likely to be Black, and less likely to be married; reported lower income and less educational attainment; and had lower rates of being insured (
Table 1). Generally, the group with a lifetime history of homelessness was more similar in terms of sociodemographic characteristics to the group that had never been homeless than to the group with past-year homelessness, although those with a lifetime history were more likely to be male and Black and less likely to be married and reported less income and education than those who were never homeless.
The two homeless groups were generally similar to each other in terms of clinical characteristics (
Table 2). Compared with those who were never homeless, the two groups had greater proportions with any past-year psychiatric diagnoses and greater proportions with individual diagnoses. Compared with the lifetime homeless group, those reporting past-year homelessness had a greater mean number of psychiatric diagnoses and were more likely to report major depression, PTSD, and panic disorder. The proportions reporting past-year disorders with various substances were also larger in the group with past-year homelessness, compared with the lifetime homeless group, and those with past-year homelessness had poorer mental health–related quality of life.
Temporal Relationship to Suicide Attempts
Those who had been homeless in the past year reported that it had been 8.4 years, on average, since their most recent suicide attempt, more recent than for the other two groups (
Table 2). After the analysis controlled for age, the differences were smaller; however, the most recent suicide attempt reported by those with past-year homelessness was, on average, 3.8 years more recent, compared with those with homelessness prior to the past year, and 4.5 years more recent, compared with those who were never homeless (both p<0.001).
Similarly, the length of time since the first suicide attempt in the entire sample was almost two decades (19.9 years); however, in the unadjusted analysis, the time was significantly shorter among those who were homeless in the past year (
Table 2). After the analysis controlled for age, these differences were smaller and no longer significant (least-square mean of years: past-year homelessness, 19.5±0.8; lifetime homelessness, 21.4±0.6; never homeless, 19.9±0.2). Finally, in the unadjusted analysis, the two homeless groups reported a large and significantly greater number of lifetime suicide attempts, compared with the group that had never been homeless (
Table 2), and these differences remained significant after the analysis controlled for age (least-square means=7.3±1.0, 5.3±0.7, and 2.7±0.3, respectively; p<0.001).
Regarding past-year suicide attempts, a significantly greater proportion of individuals who were homeless in the past year reported a suicide attempt in the past year, compared with those with lifetime homelessness and those who were never homeless (21.0%, 5.8%, and 6.3%, respectively, p<0.001). In an unadjusted logistic regression analysis, those who were homeless in the past year were 3.9 times more likely than those who were never homeless to report a suicide attempt in the past year (95% confidence interval [CI]=2.27–6.87). After adjustment for age, the odds ratio (OR) decreased to 3.2, but remained significant (95% CI=1.81–5.70). Those with homelessness prior to the past year were no more likely than those who had never been homeless to report a past-year suicide attempt in the unadjusted analysis (OR=0.90, 95% CI=0.53–1.56) or in the age-adjusted analysis (OR=0.97, 95% CI=0.56–1.68).
After adjustment for other risk factors that were elevated among adults who had been homeless in the past year (identified from
Tables 1 and
2), homelessness in the past year was no longer significantly and independently associated with having made a suicide attempt in the past year, although there was a clear trend (OR=1.81, 95% CI=0.99–3.27) (
Table 3). The aggregate number of past-year psychiatric diagnoses and higher levels of pain were the only factors found to be significantly and independently associated with past-year suicide attempts (
Table 3).
In an unadjusted analysis that included only veterans with a history of suicide attempts, those who were homeless in the past year were 9.6 times more likely than veterans who were never homeless to report a past-year suicide attempt (95% CI=1.13–80.79). When adjusted for age, the OR decreased to 7.1 (95% CI=1.12–45.40).
Discussion and Conclusions
Using nationally representative survey data, this study found that 21.0% of U.S. adults who reported homelessness in the past year also reported a suicide attempt in the past year, more than three times the proportions in the other two groups. Recent homelessness was thus associated with a substantially elevated risk of recent suicide attempts, suggesting that homelessness and suicidality strongly co-occur and may be causally related. Suicidal behavior in all groups appeared to have started decades ago, and the most recent suicide attempt among those reporting past-year homelessness occurred an average of more than 8 years prior to the survey. Thus, even the most recent suicide attempt most likely occurred well before the current episode of homelessness, although possibly during an earlier episode of homelessness.
Furthermore, those reporting past-year homelessness had many sociodemographic characteristics and current mental health diagnoses associated with both homelessness and suicidality, making it impossible to identify a significant independent association of homelessness and suicide attempts excluding other risk factors. Thus, although recent homelessness is clearly associated with recent suicide attempts and may be used to identify an appropriate target population for suicide prevention, the long history of suicidal behavior and the presence of numerous other risk factors, many of which are broad social determinants of mental health (
18,
19), do not allow us to conclude that there is an independent causal relationship between homelessness and suicide attempts, although their association appears to be robust and clinically relevant.
Our findings are consistent with a number of prior studies demonstrating that a history of homelessness is associated with increased suicidal behavior (
7,
9) and that this association is strong among both veterans and nonveterans (
2,
8). Our findings are also consistent with other studies demonstrating a high co-occurrence of current or recent homelessness and recent suicide attempts (
10–
13).
Using self-report data from both veterans and nonveterans, our study replicated findings from recent studies based on administrative data demonstrating that recent homelessness and suicidality strongly co-occur (
13). Our study extends prior research by presenting comparative data on the temporal relationship between recent homelessness and past suicide attempts in a national sample. Additionally, previous studies that have found an increased risk of suicide among homeless populations did not provide precise information, as we did, on the time since past suicide attempts or comparisons with those who had either not experienced homelessness or no experience of homelessness in the past year. Although our findings showed a strong bivariate association between recent homelessness and suicide attempts, we identified a number of other shared risk factors for both attempted suicide and homelessness and thus could not conclusively determine that homelessness acts as an independent antecedent risk factor leading to suicide attempts, as other studies have suggested (
2,
13).
These findings should, nevertheless, encourage suicide prevention activities in homeless service programs. Although homelessness services and suicide prevention are two priority areas for the VHA and other public mental health care systems, these services have not typically been coordinated (
2). Recently, national conference calls for clinicians working in homelessness service programs have begun to take place to draw their attention to the risk of suicide in this population. Moreover, VHA has recently initiated a program of training and service coordination, called S.A.V.E. (signs, ask, validate, encourage and expedite), to educate clinicians to identify and respond appropriately to veterans who may be at risk of suicide (
20). Our findings, along with those of others, underscore the potential value of such efforts to provide training to clinicians in homelessness service programs. At the same time, however, treatment of suicidality and underlying conditions appears to be needed long before the onset of homelessness. Currently homeless adults are a relevant target population for suicide prevention, and individuals treated for suicidality deserve screening for homelessness risk or referral to homelessness service programs (
13).
This study had several limitations that are important to consider. First, the cross-sectional design did not allow for the determination of causality between homelessness and suicide attempts or other variables, although their association was robust and clinically relevant. Additionally, we could not determine the precise onset or duration of episodes of homelessness and thus whether suicide attempts occurred before, during, or after a homeless episode. Third, the target population of the NESARC-III comprised civilian residents of households and selected group quarters and thus excluded currently homeless adults, limiting the generalizability of these data to this important population. Survey respondents with past-year homelessness were all domiciled at the time of the survey. This limitation is part of a broader challenge for the field in assessing homeless individuals in nationally representative population-based studies.
Despite these limitations, this study demonstrated an elevated rate of recent suicide attempts among recently homeless adults. However, long histories of suicidal behavior were typical and may not have been precipitated by a recent episode of homelessness. These data, however, support integrated efforts to address both suicidality among homeless adults and risk of homelessness among adults who report recent suicide attempts.