Risk factors for homelessness among schizophrenic men: a case-control study

Risk Factors for Homelessness among

Schizophrenic Men: A Case-ControlStudy

Carol L. M. Caton, PhD, Patrick E. Shrout, PhD, Paula F. Eagle, MD,Lewis A. Opler, MD, PhD, Alan Fel"x MD, and Boanerges Dominguez, MS

IntroductionA recent governmental report esti-

mates that 1 in 20 of the severely mentallyill experience homelessness.' Using dif-ferent definitions of homelessness orhousing instability, empirical studies ofstate mental hospital constituents havefound that from 9% to 28% have beenhomeless.-4 The crisis nature of home-lessness, both for the individual and forsociety, demands a clearer understandingof why some of the severely mentally illbecome homeless while others do not.

Previous studies in which differencesbetween the homeless and the residen-tially stable have been explored suggestthat the homeless have more psychiatricdisorders meeting standard diagnostic cri-teria5'6 and higher rates of hospitalizationand arrest.5,7 Studies focused on definedpsychiatric populations have found thatthe homeless are more likely to abuse al-cohol3'8 and/or drugs,3'4'6 have highersymptom levels,3'7'8 and are less likely tocomplywith prescribed treatments.3,8 Theuse of different definitions of homeless-ness, the employment of varied ap-proaches to the assessment of psychiatricdisorder, and the selection of differentsubgroups of the homeless population inthese investigations limit their generaliz-ability.

Attempts to identify areas of differ-ences between the homeless and the neverhomeless outside of the illness domain aremore limited. Studies of childhood ante-cedents of adult homelessness have re-vealed that homelessness (or its duration)is associated with placement in fostercare9 '0 or group homes,10 physical abuse,9and runaway episodes.'0 Although thebreakdown offamily ties has been thoughtto play a role in the genesis of homeless-ness, perceptions of the kinship bond in

adulthood have not been found to be asimportant as patient characteristics in dis-tinguishing the homeless from the nonho-meless.

In spite of the discourse on deinsti-tutionalization ofmental health services asa possible factor in homelessness amongthe severely mentally ill,11,12 the serviceuse histories of homeless and never-homeless persons have not been com-pared with regard to access to and utiliza-tion of community-based treatment andsupport services. Although the literaturesuggests that multiple factors may sepa-rate the homeless from the never home-less, there has been no prior attempt toincorporate multiple risk factors into a sin-gle study.

We recently completed a case-con-trol study ofindigent schizophrenic men inNew York City designed to test hypothe-ses about differences between the home-less and never-homeless mentally ill. Tofacilitate an adequate test of hypothesesabout differences between the homelessand never homeless, we elected to focusour study on a homogeneous group of in-digent adults suffering from schizophre-nia. Schizophrenia is the most commondiagnosis leading to chronic mental ill-ness13 and is widely prevalent amongthose living in shelters.1,14,15 Because fourof five homeless adults are male,16 it is

Carol L. M. Caton, Paula F. Eagle, Lewis A.Opler, Alan Felix, and Boanerges Dominguezare with the Department of Psychiatry andSchool of Public Health, College of Physiciansand Surgeons, Columbia University, NewYork,NY. PatrickE. Shrout iswith the Department ofPsychology, New York University.

Requests for reprints should be sent toCarol L. M. Caton, PhD, Department of Psy-chiatry, Columbia University, 722 W 168th St,Box 114, New York, NY 10032.

This paperwas accepted August 23, 1993.

American Journal of Public Health 265

.......................... ZI-11,

Caton et aL

possible that pathways to homelessnessare gender-specific. Therefore, our sam-plewas stratified bygender. Findings fromour study ofwomen will be reported later.

We probed three domains of influ-ence on homelessness: severity of illness,family background, and prior mentalhealth service use. We hypothesized thatschizophrenic menwho became homelesswould have poorer pre-illness social func-tioning, higher levels of the positive andnegative symptoms of schizophrenia,greater abuse of drugs and alcohol, andmore antisocial personality disorder. Inaddition, we hypothesized that the home-less would have experienced greater fam-ily disorganization in childhood, wouldhave imposed a greater caretaker burdenafter becoming mentally ill, and wouldhave less current family support. Finally,we hypothesized that the homeless wouldbe less adherent to prescnbed treatmentand would have less help from the mentalhealth system in finding adequate housingand procuring long-term follow-up care.

MethodsWe employed a case-control design

with 100 men in the (sheltered) homelessgroup and 100 men in the never-homelessgroup. All 200 subjects had experienced atleast one psychiatric hospitalization andwere between the ages of 18 and 44 years.To be eligible for inclusion, the men wererequired to meet DSM-llI-R criteria forschizophrenia or schizoaffective disorder,determined through a research diagnosticinterview, the Structured Clinical Inter-view for DSM-HI-R (SCID).17 Voluntaryinformed consent was elicited from eachsubject after a treating clinician attested tothe subject's ability to fully comprehendthe implications of participation in our re-search.

Case subjects met criteria for literalhomelessness, meaning that they had nofixed abode andwere forced to sleep in thestreet or in shelters.18 Theywere recruitedby a mental health day treatment programlocated on-site at a municipal shelter forhomeless men in Upper Manhattan.

Control subjects had no lifetime his-tory of literal homelessness, as determinedby a screening instrument.4 They were dis-charge-ready inpatients or outpatients re-cruited from Columbia University-affiliatedpsychiatric services in Upper Manhattan.More than four of five control subjects re-sided in a family setting, usually with a par-ent or sibling.

Because Blacks have been found tobe overrepresented among the sheltered

homeless in New York City,19 we madeevery effort to match case and control sub-jects on ethnicity (Black, White, Hispan-ic). When we encountered difficulty infinding an adequate number of Black sub-jects who had never been homeless, weabandoned our matching strategy. Rea-sons for the paucity of Black never-home-less subjects constitute a topic in need offurther study.

A total of 248 men were asked to par-ticipate in the study. Eleven (4.4%) re-fused, and 22 (8.9%) dropped out beforecompleting the interview battery. Four offive dropouts were homeless men whoprecipitously left the shelter without let-ting the mental health staff know how toreach them. Fifteen (6.1%) of the home-less subjects with completed interviewswere eliminated from the study because ofinconsistent or poor-quality data thatcould not be improved with a reinterview.In such cases the subject was either toodelusional or too disorganized to functionas a reliable informant. It is possible thatthe more disturbed subjects were not in-cluded in the final sample.

Subjects were interviewed by mentalhealth clinicians specially trained to ad-minister the assessment battery. Clinicaland anamnestic data were verified withinformation contained in the clinical caserecord. A family member was also inter-viewed for 57% of the homeless and 68%of the never homeless. The most commonreason for the lack of a family interviewwas the subject's estrangement from kin.

InstrunentsThe following research instruments

were used to assess key study variables:Pre-illness social functioning was

rated with the Los Angeles Social Attain-ment Scale, a seven-item instrument fo-cused on peer relationships and social par-ticipation in late adolescence.20 The alphareliability coefficient for this scale is .80.

The positive (active psychotic symp-toms such as delusions, hallucinatory be-havior, and conceptual disorganization)and negative (deficit symptoms such asblunted affect, emotional withdrawal, anddifficulty with abstract thinking) dimen-sions of schizophrenia were assessed withthe Positive and Negative SyndromeScale, a 30-item rating scale evaluatingsymptoms present during the 7-day periodprior to the interview.21 The alpha reliabil-ity coefficient was .79 for the positivescale, .84 for the negative scale, and .83for the general psychopathology scale.

Current and lifetime alcohol and drugabuse or dependence were evaluated with

the SCID, which yields information on theextent to which heavy use of alcohol orseven classes of drugs meets criteria for adiagnosis based on standards commonlyapplied in psychiatry.17

Antisocial personality disorder wasevaluated with the SCID-II, the segmentof the SCID dealing with personality dis-orders.17 This instrument probes the pres-ence of conduct problems-such as run-ning away from home, being truant fromschool, initiating physical fights, or belong-ing to a gang-occurring before the age of15, as well as patterns of irresponsible, de-structive, or illegal activity carrying overinto adulthood. For a subject to meet cri-teria for a diagnosis of antisocial personal-ity disorder, a conduct disorder must havebeen present in early adolescence.

Family disorganization in childhoodwas evaluated with 4-point rating scalescontained in the Community Care Sched-ule.22 These ratings are based on carefullydefined anchor points for the assessmentof nurturing constancy, residential stabil-ity, adequacy of income, dependence onpublic assistance, family violence, paren-tal criminality, parental mental illness, andparental substance abuse-the compo-nents of an index of family disorganiza-tion. Information for this instrument waselicited from both the subject and a familymember (whenever possible) to obtain themost complete picture offamilylife duringthe subject's childhood. The alpha reli-ability coefficient for the seven-item indexof family disorganization is .69.

Current family support was rated ona scale of adequacy, based on the degreeof support and assistance available fromfamily members regarding money, shel-ter, food, clothing, advice, and compan-ionship.2 Information for this rating wasobtained from the subject and corrobo-rated by a relative when possible.

Caretaker burden was assessed witha semistructured interview that yielded a3-point rating of overall burden-23 Unlikeother family variables, family burden re-quired an assessment from a family mem-berwith firsthand knowledge of the familyexperience of livingwith the patient. Fam-ily burden data were missing for one thirdof never-homeless and nearly one half ofhomeless case subjects.

Prior service use was explored withitems contained in the Community CareSchedule22 that were designed to elicit de-tailed information on medication adher-ence patterns, long-term follow-up care,and housing placement at hospital dis-charge. Medication adherence was ratedon a 4-point scale on the basis of the sub-

266 American Joumal of Public Health Febnk-uy 1994, Vol. 84, No. 2

Schzohfi Men

ject's self-report. Long-term follow-upcarewas defined in terms ofthe number ofmonths in outpatient treatment with thesame therapist.

Statistical Methods ofAnalysisThe case subjects (homeless) were

comparedwith the control subjects (neverhomeless) with respect to three classes ofkeyvariables. In the illness domain, therewere six variables: premorbid social at-tainment scale, positive symptom score,negative symptom score, and binary indi-cators of alcohol abuse, drug abuse andantisocial personality disorder. In the fam-ily domain, there were three variables:family disorganization index, adequacy offamily support, and assessment ofburden.In the service use domain, there were twovariables: medication adherence and ex-istence of a long-term therapist.

First we compared the case and con-trol subjects without taking into accountsampling characteristics of the groups,and then we compared them after makingstatistical adjustments for demographicvariables that could possibly account forobserved differences. Statistical adjust-ments were made by means of logistic re-gression, with the binary case/controlvariable treated as the outcome and therisk variables and possible confounderstreated as explanatory variables. Possibleconfounders were chosen from those de-mographicvariables shown in Table 1 thatwere at all related to the case/control vari-able (at theP < .15 level). These includedplace of birth, New York City residencehistory, ethnicity, and veteran status. Wedid not adjust for current income vari-ables, because they can be expected to berelated to homelessness structurallyrather than incidentally.

To facilitate the comparison ofadjusted and unadjusted associations be-tween the risk variables and the case/control distinction, we presented both us-ing results from logistic regressionanalyses. The test statistic we present isthe likelihood ratio chi-square from the lo-gistic analyses.2 For the unadjusted case/control comparisons the pattern of signif-icance results fromthe likelihood ratio chi-square is the same as would be obtainedby means of t tests or Pearson chi-squarestatistics.

Rems

Demographic ProfileThe homeless (case) group and nev-

er-homeless (control) group had many

traits in common, including median age(32 years); marital status (four fifths hadnever married or were currently single);median level of education (11 years); andemployment status (more than four fifthswere unemployed). Moreover, most (89%of the homeless and 76% of the neverhomeless) were members of ethnic minor-ities. However, there were more Blacksamong the homeless (66% vs 22%) andmore Hispanics among the never home-less (54% vs 23%). Ethnic differenceswere reflected in differences in religiousbackground (Catholics predominatedamong the never homeless, whereas Cath-olics and Protestants were nearly equallyrepresented among the homeless) andplace ofbirth. Nearly twice asmanyneverhomeless as homeless (39% vs 21%) wereforeign-born.

Both groups were heavily dependenton public assistance. Nearly four fifths ofthe never homeless and three fifths of thehomeless were receiving income from en-titlements, averaging about $430 permonth. Thirteen percent of the neverhomeless and 19% of the homeless earnedan average of $100 per month at unskilledjobs. Thirty-three percent of the neverhomeless and 8% of the homeless also re-ceived some financial support from theirfamilies. For subjects who had everworked, the longest duration of steadyemploymentwas an average of 12 months.Unemployed subjects had been out ofwork for an average of 4 to 5 years. Ninepercent of the never homeless and 22% ofthe homeless had served in the armedforces.

Family History and Ilbless OnsetThirty-five percent of the never

homeless and 23% of the homeless hadlived with both biological parents frombirth to 18 years of age. The majority inboth groups experienced one or morechanges in the primary nurturing adult inchildhood and adolescence. One fourth ofthe never homeless and more than onethird ofthe homeless had experienced twoor more such changes. One never-home-less subject and six homeless subjects ex-perienced foster care placement; fourhomeless subjects had been placed in agroup home. Four never-homeless sub-jects and 20 homeless subjects had a fa-ther, mother, or sibling with a history ofhomelessness.

The median age of onset of psychiat-ric disorder was 21 years for the neverhomeless and 20 years for the homeless.Never-homeless subjects experiencedtheir first psychiatric hospitalization at a

median age of 22 years; for homeless sub-jects it was 21 years. Most subjects hadextensive hospitalization histories. Fifty-eight percent of the never homeless and56% of the homeless had been hospital-ized in the 12-month period preceding the

American Journal of Public Health 267February 1994, Vol. 84, No. 2

Caton et aL

interview. The median number of hospitaldays in this time period was 60 for thenever homeless and 43 for the homelesssubjects. At the termination of the mostrecent hospitalization episode, 11 home-less subjects and 1 never-homeless subjecthad been discharged against medical ad-vice. Ninety-six percent of the neverhomeless and 83% of the homeless hadbeen given a recent prescription for neu-roleptic medication (the class of psycho-tropic drugs often used to treat schizo-phrenia).

Subjects with substance abuse prob-lems had initiated alcohol and drug use inmidadolescence. The median age of onsetof drug abuse (any drugs) was 18 years fornever-homeless subjects (44%) and 16years for homeless subjects (77%). Themedian age of onset for alcohol abuse was18 years for both groups (44% of the neverhomeless and 49% of the homeless). Fif-teen percent of the never homeless and34% of the homeless had received someform of substance abuse treatment in the12-month period preceding the interview.

Thirteen percent of the never home-less and 46% of the homeless had had se-vere conduct problems before the age of15 years. Six percent of the never home-less and28% ofthe homeless reported thatthey had run away from home at leastonce during early adolescence. Nine per-cent of the never homeless and 42% of thehomeless were given a research diagnosisof adult antisocial personality disorder,and25% ofthe never homeless and 72% ofthe homeless had jail or prison histories.

Distinctions between the Homelessand the Never Homeless

Table 2 summarizes tests ofstudy hy-potheses about the differences betweennever-homeless and homeless schizo-phrenic men. The tests are presented intwo forms, one that is unadjusted for pos-sible case-control differences that mayconfound the group comparisons and onethat adjusts for demographic factors thatmay be related to selection into the study.

Within the illness domain, there wereno significant differences between the

homeless and the never homeless on theLos Angeles Social Attainment Scalescores in either unadjusted or adjustedtests. Average ratings on peer socializa-tion and social participation in late adoles-cence were at the "poor premorbid" levelin both groups. Homeless subjects didshow significantly higher levels ofpositivesymptoms than did the never-homelesssubjects (P < .01, unadjusted; P < .05adjusted). However, there were no majordifferences in negative symptom levels.There were no meaningful differences inthe proportion of subjects in each groupwith a concurrent diagnosis of alcoholabuse, but a significantly larger number ofhomeless subjects had a concurrent diag-nosis of drug abuse (P < .01 for both un-adjusted and adjusted tests). Similarly, asignificantly larger number of homelesssubjects had a concurrent diagnosis of an-tisocial personality disorder (P < .01 forboth unadjusted and adjusted tests).

Within the family domain, the meanscore on the index of family disorganiza-tion was higher, indicating greater impair-ment, for homeless subjects (P < .01, un-adjusted; P < .05, adjusted). Familysupport was less adequate for the home-less (P < .01 for both tests). However,there were no significant differences in theoverall levels of caretaker burden.

In terms of service use issues, thehomeless initially appeared to adhere lessto prescribed medication regimens(P < .01, unadjusted), but that differencedid not withstand adjustment for potentialconfounders (P < .10). Fewer homelesssubjects had a long-term therapist, an in-dicator of continuity of care (P < .01,both adjusted and unadjusted).

At discharge from the most recentpsychiatric hospitalization, 50% of thehomeless (undomiciled before hospitaliza-tion) were discharged to a shelter or thestreets. The 95% confidence interval forthe proportionP is between 40% and 60%.

Table 2 shows how each of the indi-vidual risk variables relates to homeless-ness after demographic factors that mightbe related to selection are controlled. Wecarried out additional analyses to deter-mine the extent to which these variablesaccounted for the same variation in thehomelessness distinction. When all therisk variables in Table 2 were entered si-multaneously in a logistic model fitting logodds of homelessness (along with the de-mographic control variables), one variablein each domain remained either significantor a trend. Still significant by the Waldstatistic were drug abuse (from the illnessdomain) (Wald = 5.28, df = 1, P < .05)

268 American Journal of Public HealthF February 1994, Vol. 84, No. 2

and adequacy of family support (from thefamily domain) (Wald = 14.41, df = 1,P < .001), whereas having a long-termtherapist (from the service use domain)was reduced to a trend (Wald = 3.07,df = 1, P < .10). These three variablesseem to reflect different aspects of the dif-ferences between the homeless and never-homeless groups.

Two variables under the illness do-main that are signifcant in Table 2 are notsignificantwhen adjusted for the other riskvariables. Positive symptoms is no longera discriminating variable when other ill-ness domain variables are controlled, butantisocial personality disorder loses sig-nificance only when adequacy of currentfamily support is adjusted.

Adequacy of family support also ap-pears to be the variable that accounts forthe effects of family disorganization indexand having a long-term therapist. Thesevariables are significant in Table 2, butthey do not appear to differentiate thehomeless from the never homeless whenadequacy of family support is adjusted.

DiwiussionIn this study we have attempted to

identify factors that differentiate econom-ically disadvantaged schizophrenics wholive in private homes from those who arehomeless and who live in public shelters.Although not definitive in terms of causalpathways, a case-control design such asours is an efficient way to spot potentialrisk factors for homelessness. Some vari-ables we explored preceded the firsthomeless experience and may play acausal role. In addition to childhood an-tecedents, our data indicate that drugabuse and antisocial personality disorderhad their onset prior to homelessness. Thecross-sectional nature of our study pre-vents us from detemiining whether othervariables are a cause or a consequence ofhomelessness. Our data show an associa-tion between family disorganization inchildhood and poor family support inadulthood in assessing the risk of home-lessness. However, a longitudinal ap-proach is needed to establish that break-down of family relationships plays acausal role in homelessness.7 Higher cur-rent levels of positive symptoms and thelack of a long-term therapist could be ei-ther a cause or a consequence of home-lessness. Higher symptom levels can pre-cipitate behaviors that undermine housingstability, but the vicissitudes of street orshelter living can increase distress and dis-rupt usual patterns of mental health care.

Ourdataonhospital discharge practicesforhomeless subjects suggest that the pattemofdischargingpatients to shelters or streetsfunctions to perpetuate homelessness.

The sampling design ofour study for-mally limits us to generalizing our findingsonly to persons who have the same de-mographic and clinical characteristics asour study subjects andwho are enrolled inmental health treatment programs similarto those from whichwe drew our sample.However, the pattern of findings suggeststhat they may be more general, particu-larly among the severely mentally ill. Wefound that the homeless schizophrenicmen differed from their domiciled coun-terparts in all three domains we investi-gated; family background, nature of ill-ness, and service use history.

Schizophrenic men who came fromfamilies with high levels of family disor-ganization, characterized by lack of nur-turing constancy, residential instability,economic inadequacy, dependence onpublic assistance, family violence, mentalillness, substance abuse, or criminality ina parent, were more likely to be in thehomeless group than in the never-home-less group, even when demographic vari-ables such as place of birth and ethnicitywere controlled. Our findings broaden thescope of childhood antecedents of adulthomelessness to include family economicproblems and parental pathology, the lat-ter being strongly associated with out-of-home placement.25 However, the logisticregression analysis revealed that poorfamily support is a more important riskfactor for homelessness than childhoodantecedents. We thought it reasonable toask whether the overall burden experi-enced by the family in taking care of theschizophrenic relative might differentiatethe homeless from the never homeless.Our negative findings should be inter-preted cautiously because our ability tomeasure family burden was limited.

The menwho ended up in our home-less group do appear to have a strikinglydifferent profile ofpsychopathology. Theywere more likely to have positive symp-toms of psychosis and to have concurrentdrug abuse and antisocial personality dis-order. The logistic regression analysis re-vealed that drug abuse was the most im-portant risk factor in the illness domain.There were no significant differences be-tween the homeless and the never home-less in premorbid social adjustmnent, neg-ative vmnptoms, or alcohol abuse. Thus,the homeless in our study were not moreseverely ill in all dimensions of psychopa-thology.

Schzohri Men

The finding that 50% of the homeless(undomiciled before hospitalization) weredischarged to a shelter or the streets indi-cates that the homeless in our study hadless access to adequate care. Althoughthis findingmay reflect local mental healthpractices, similar findings have emergedfrom studies carried out in Chicago2- andrural Ohio.v The logistic regression anal-ysis revealed that the most important riskfactor in the service use domain was nothaving a long-term therapist. Future workwill have to determine whether lack of along-term therapist is an access problemor a consequence of resistance to treat-ment.

Clearer identification of risk factorsfor homelessness can lead to the develop-ment of preventive intervention strate-gies. For example, thorough assessmentof patterns of substance use, family sup-port, and prior outpatient service use inconcert with routine discharge planningcan identify those who require additionalservices and supportive housing. Patientswith co-disorders can be singled out forintensive case management to improvethe successful coordination of mentalhealth and substance abuse treatmentservices.23-31 Finally, although risk fac-tors aid in identifying those individualsmost vulnerable to homelessness, they donot obviate the need for policymakers atall levels of government to address thechronic shortage of affordable housing forindigent Americans. [

AcknowledgentsThe research on which this report is based wassupported by the National Institute of MentalHealth, grant MH44705.

Our thanks to our dedicated and talentedresearch support staff, including Luz Romero,BS (who also assisted in the preparation of themanuscript); Andrea Cassells, MPH; LouisCaraballo, BA; Diane Engel, CSW; Lee Futro-vsky, PhD; Barbara Holton, CSW; ElizabethMargoshes, PhD; and Mary Ellen Russell, MS.We are also grateful to Richard Jed Wyatt, MD,and Jerome K. Myers, PhD, for their criticalreading of an earlier draft.

References1. Federal Task Force on Homelessness and

Severe Mental Illness. Outcasts on MainStreet. Washington, DC: InteragencyCouncil on the Homeless; 1992:18.

2. Mowbray CT, Johnson US, Solar A.Homelessness in a state hospital popula-tion. Hosp Community Psychiatry. 1987;38:880-882.

3. Drake RE, Wallach MA, Hoffman JS.Housing instability and homelessnessamong aftercare patients of an urban statehospital. Hosp Community Psychiatry.1989;40:46-51.

4. Susser ES, Lin SP, Conover SA. Risk fac-

February 1994, Vol. 84, No. 2 Amencan Journal of Public Health 269

Cat.. et aL

tors for homelessness among patients ad-mitted to a state mental hospital.AmJPsy-chiatry. 1991;148:1659-1664.

5. Fischer PJ, Shapiro S, Breakey WR, An-thony JC, Kramer M. Mental health andsocial characteristics of the homeless: asurvey of mission users. Am J PublicHealth. 1986;76:519-524.

6. Koegel P, Burnam A, Farr RK. The prev-alence of specific psychiatric disordersamong homeless individuals in the innercity of Los Angeles.Arch Gen Psychiaby.1988;45:1085-1092.

7. Tessler RC, Gamache GM, Rossi PR, Leh-manAF, GoldmanHH. The kdndred bondsof mentally ill homeless persons. NEngIJPublc Poly. 1992;8:265-280.

8. Drake RE, WallachMA, Teague GB, Free-man DH, Paskus TS, Clark TA. Housinginstability and homelessness among ruralschizophrenic patients. Am J Psychiaty.1991;148:330-336.

9. Winkleby MA, Rockhill B, Jatulis D, Fort-mann SP. The medical origins ofhomeless-ness. Am J Public Health. 1992;82:1394-1398.

10. Susser ES, Lin SP, Conover SA, StrueningEL. Childhood antecedents of homeless-ness in psychiatric patients.Am JPsychi-atry. 1991;148:1026-1030.

11. Lamb HR. Deinstitutionalization and thehomeless mentally ill. Hosp CommunityPsychiatry. 1984;35:899-907.

12. LambHR. Willwe save the homeless men-tally ill? Am J Psychiatry. 1990;147:649-651.

13. Goldman HH, ManderscheidRW. Chronic

mental disorder in the United States. In:Manderscheid RW, Barrett SA, eds. Men-talHealth, United States, 1987. Rockville,Md: US Dept of Health and Human Serv-ices, National Institute of Mental Health;1987:1-11.

14. Arce AA, Tadlock M, Vergare MJ, Sha-piro SH.A psychiatric profile of street peo-ple admitted to anemergency shelter.HospCommunity Psychiatry. 1983;34:812.

15. Bassuk EL, Rubin L, Lauriat A. Is home-lessness a mental health problem? Am JPsychiaty. 1984;141:1546-1549.

16. Burt MR, Cohen BE. Ame,ica's Home-less: Nwnbers, Characterisics, and Pro-grams That Serve Them Washington, DC:Urban Institute Press; 1989:36.

17. Spitzer RL, Williams JBW, Gibbon M,First MB. Stnucwral Clinical InteriewforDSM-III-R. Washington, DC: AmericanPsychiatric Press; 1990.

18. Rossi PH, WrightJD, Fisher GA, Willis G.The urban homeless: estimating composi-tionand size. Science. 1987;235:1336-1341.

19. Susser E, Struening EL, Conover S. Psy-chiatric problems in homeless men. ArchGen Psychiatry. 1989;46:845-859.

20. Goldstein MJ. Further data concerning therelation between premorbid adjustmentand paranoid symptomatology. SchizophrBuLL 1978;4:236.

21. Kay SR, Fiszbein A, Opler LA. The Pos-itive and Negative Syndrome Scale(PANSS) for schizophrenia. SchizophrBulL 1987;13:261-275.

22. Caton CLM. 7he C(mmity Care Sched-ule. New York, NY: NewYork State Psy-

chiatric Institute; January 1989.23. Pai S, Kapur RL. The burden on the family

of a psychiatric patient: development of aninterview schedule. BrJPsychiatry. 1981;138:332-335.

24. Hosmer DW, Lemeshow S. Applied Lo-gi;i Regression. New York, NY: JohnWiley & Sons Inc; 1989.

25. Caton CLM, Eagle P, Felix A, Shrout P.Famiy factors in the genesis of homeless-ness. Presented at the 120th Annual Meet-ing of the American Public Health Associ-ation; November 11, 1992; Washington,DC.

26. Sosin MR, Grossman S. The mental healthsystem and the etiology of homelessness: acomparison study. J Communiy Psychol-ogy. 1991;19:337-351.

27. Belcher JR. On becoming homeless: astudy of chronically mentally ill persons. JCommunity Psychology. 1989;17:173-185.

28. Osher FC, Kofoed LL. Treatment of pa-tients with psychiatric and psychoactivesubstance abuse disorders. Hosp Commu-nity Psychiatry. 1989;40:1025-1030.

29. Caton CLM, Wyatt RJ, Felix A, GrunbergJ. Follow-up of chronically homeless men-tally ill men. Am J Psychiatry. 1993;150:1639-1642.

30. Abram KM, Teplin LA. Co-o rng dis-orders among mentally ill jail detainees.Am Psycholgist. 1991;46:1036-1045.

31. Steadman HJ, McCarty DW, MorrisseyJP. 7he Mentally Ill in Jaik Planning ForEssenial Services. New York, NY: Guil-ford Press; 1989.

270 Amencan Journal of Public Health February 1994, Vol. 84, No. 2