Impact of Work Therapy on Health Status Among Homeless, Substance-Dependent Veterans:  A Randomized Controlled Trial FREE

MENTAL HEALTH professionals often refer clients to clinical programs where the primary goal is competitive employment. These programs of vocational rehabilitation, sheltered workshops, job clubs, and transitional and supported employment have shown varying degrees of success in helping the mentally ill attain employment.^1- 5 However, the relationship between structured work and nonvocational health outcomes remains inconclusive.

In contrast, economists consider productive work as an outcome of a healthy labor force.⁶ Epidemiologists emphasize work-related illnesses,⁷ hazards,^8- 9 stresses,^10- 11 and adverse health behaviors.¹² Public health research finds retirement often leading to poorer physical functioning,^13- 15 somatic complaints,¹⁶ and fatal heart attacks,¹⁷ although it is unclear whether retirement resulted in or from poor health. What is not known is whether structured work can change health behaviors and, thus, health status.

The current study focuses on the Department of Veterans Affairs (VA) compensated work therapy (CWT) program, a clinician-supervised, abstinence-contingent, performance-based work program for homeless persons. The CWT program offers pay rewards but requires random drug testing and adherence to a schedule of outpatient addiction services determined to be clinically appropriate for each client. The program's clinical setting ensures that the link between work performance (productivity and presentation) and health behaviors (sobriety and use of addiction services) to job rewards (wages, hours, and responsibilities) are not undermined by the vicissitudes of labor markets. The study focused on homeless persons who lack regular employment and structured environments. Study outcomes included substance dependence behaviors (substance consumption and use of addictions treatment services), health outcomes (addiction-related physical symptoms, psychiatric symptoms, and health functioning), and other aspects of quality of life (rates of incarceration and homelessness). The study was set at VA medical centers where all clients had access to comprehensive health services, including addictions treatment and vocational rehabilitation. Access to care did not depend on employment status.

Subjects were recruited from homeless centers and addiction units at 4 VA medical centers located in Bedford, Mass; Northampton, Mass; Topeka, Kan; and St Cloud, Minn. Qualifying subjects (1) were eligible for VA care,(2) were homeless at intake, (3) satisfied DSM-III-R criteria for substance dependence or abuse,¹⁸ and (4) were willing to enter a therapeutic work program. For this study, homeless persons have "a primary nighttime residence that is a supervised publicly or privately operated shelter designed to provide temporary living accommodations, an institution that provides a temporary residence for individuals intended to be institutionalized, or a public or private place not designed for, or ordinarily used as, a regular sleeping accommodation for human beings"(McKinney Act, 42 USC 11302). Excluded patients were those (1) institutionalized at intake, (2) mentally or physically unable to work in the opinion of the CWT admitting clinician, (3) diagnosed with a terminal illness, (4) planning to leave the study service area within 1 year, or (5) unresponsive at the intake interview.

First introduced in 1957 (38 USC 1718), the VA CWT programs¹⁹ teach work discipline with abstinent-contingent, performance-based work for pay in staff-supervised, structured settings. As administered at the 4 study sites between July 1, 1994, and March 31, 1997, CWT offered work opportunities (continued employment, higher wages, hours, promotion, and responsibility) based on client work performance (productivity, reliability, presentation, and punctuality) and health behavior (sobriety and use of recommended addiction services) as judged by CWT clinicians through client observation, random drug screenings, and chart reviews. The CWT combined elements of supported employment(nontrivial wages paid from revenues earned from private sector contracts) and stepwise programs (clinician supervision and VA workshops). Consequences for inappropriate behaviors (positive drug screenings) varied from a reprimand to program dismissal and loss of CWT employment. Unfettered by informal organizations, appellate bureaucracies, and labor regulations, clinician managers were free to protect the integrity of work incentives and make judgments in the client's best interest. Control patients were not enrolled in CWT for at least the 1-year follow-up, although both subjects who received CWT and control subjects had full access to VA comprehensive medical, mental, addictions, and vocational rehabilitation services, including job-seeking skills and support, job documentation and interviewing techniques, and career counseling and assessment.

We hypothesized that the mandatory addiction treatment and drug screening protocols in CWT, among other factors, would have an immediate impact on addiction status as assessed by patient use of drugs and alcohol, and secondarily on patient use of addiction services and prevalence of substance use–related symptoms. We also hypothesized that subjects in the CWT group would gradually experience better psychiatric and medical status during the 1-year follow-up after CWT began.

Research assistants at each of the 4 study sites admitted patients, checked eligibility, administered baseline questionnaires, and then called a national coordinator (A.W.) in Dallas, Tex, to receive a patient identification number. Once the subject was assigned to a number, the coordinator revealed treatment assignment by opening, in sequence, sealed envelopes that had been addressed to both site and patient identification number. The enclosed assignments were based on random numbers generated by an SPSS program (SPSS Inc, Chicago, Ill). To attract subjects, a sampling ratio of 1 (control) to 4 (subject in CWT) was adopted. Subjects randomized to CWT were offered employment as soon as a CWT-sponsored job became available, usually within 6 days. After the baseline interview, neither patient nor assessors were blinded to study assignments.

Data came from face-to-face interviews conducted at baseline and at 3-month intervals for 1 year by trained interviewers under supervision of master's- or doctorate-level prepared social workers and psychologists. Information was verified by VA computerized file extracts and patient chart reviews. To ensure reliability of data collection, interviewers received group training, participated in routine conference calls, and were tested by rating prerecorded taped interviews.

As the primary measure, addiction status was determined by the Addiction Severity Index (ASI),^20- 22 summarizing drug and alcohol use problems into composite scores ranging from 0 to 1, with higher scores indicating more problems. Administered under confidential circumstances, these measures have been shown to be valid.²³ As confirmatory measures of outcome, addiction care–seeking behavior was represented by a dichotomous variable that assumed a value of 1 if the patient had obtained at least 1 outpatient addiction visit during a 3-month period before each follow-up interview, and 0 otherwise. Substance use–related physi cal problems were measured as a simple count of 11 symptoms that included the presence or absence of seizures, delirium tremens, or memory trouble; seeing or hearing things not there after cutting down on substance use; liver disease; hepatitis or yellow jaundice; vomited blood or other stomach troubles; tingling or numbness in feet; inflammation of pancreas or pancreatitis; drug overdose; subacute bacterial endocarditis; cellulitis or other infections due to intravenous drug use; and continued use when aware that consumption may exacerbate serious physical illness.

Exploratory outcomes included psychiatric status assessed by a normalized global Brief Symptom Inventory,^24- 25 the ASI Psychiatric Status, and the Mental and Social Functioning components of the 36-Item Short Form Health Survey from the Medical Outcomes Study (SF-36).^26- 27 Medical status was assessed with the Physical Functioning component of the SF-36 and the Medical Status component of the ASI. Other measures included presence or absence of episodes of criminal justice incarceration and homeless nights during the 3 months before each interview.

The cost of inpatient (outpatient) care was calculated in 3-month intervals by multiplying the number of days (visits) by a cost per day (per visit) and summed over all respective bed sections (clinic stops). Costs were calculated for medicine, psychiatry, and addiction and rehabilitation services. Unit costs were based on median cost (1997 US dollars) for VA medical centers. Data came from VA's cost files that include costs for professional time, supplies, indirect administrative support and building maintenance, and depreciation for equipment and buildings.²⁸ Separate estimates were calculated for medical, psychiatric, and addiction and rehabilitation services.

Two-tailed χ² (discrete) and t(continuous) tests compared group differences at baseline. Longitudinal outcome data were analyzed by hierarchical regressions with the use of HLM/3L software²⁹ to compute outcome differences between CWT and control groups as a percentage of baseline scores (continuous measures) and an odds ratio (dichotomous measures).

Tracking homeless patients necessitated the use of parametric statistics to account for missing observations and variable intervals between follow-up interviews, discussed in clinical trials generally³⁰ and mental health specifically.^31- 32 With these approaches, time of observation enters as an explanatory variable to determine whether measured outcomes change with time.^33- 34 Clinical programs are evaluated by detecting whether differences between treated and control subjects "grow" over time,^35- 37 although estimates are unbiased only to the extent that the occurrence of missing values and length of intervals between follow-ups are uncorrelated with treatment assignment. The approach permits one to adjust for patient use of other services as time-dependent covariates.

Traditional parametric approaches determine treatment effects by detecting whether differences in outcomes between treated and control patients grow with time. These growth strategies may not describe programs that focus on behavior (eg, addictions). Here, program benefits may mature immediately after treatment begins, only to diminish with time because (1) efficacy "wears off,"(2) efficacious alternatives help control clients "catch up," (3) the intervention only enhances the speed of recovery, or (4) the intervention postpones inevitable outcomes of an intractable disorder. Thus, we estimated immediate and growth-rate effects by means of a declining-effects model (T.M.K., Thomas J. Carmody, PhD, Trish Suppes, MD, PhD, A. John Rush, MD, M. Lynn Crismon, PharmD, Alexander L. Miller, MD, Marchia Toprac, PhD, and Madhukar Trivedi, MD, unpublished data, August 2001):

y_sit = β₀+ β₁I_si+ β₂₀Z_t + β₂₁Z_tI_si + β₃₀t + β₃₁tI_si + β₄c_sit + α_s + u_si + v_sit,

where y_sit is the outcome assessed for patient i at site s (s = 1, 2, . . . 4) at t quarters (3 months) past baseline (t = 0). To evaluate CWT, the "intent-to-treat" variable I_si assigns a value of 1 for patients randomized to CWT and 0 if assigned to the control group. Z_t is a dummy variable that assumes a value of 1 for t>0, and 0 otherwise. The variable c_sit is use of care during 3 months before t, including psychiatric care (psychiatric status outcomes), medical care (medical outcomes), and addiction and rehabilitation services (addiction status). The use of care covariate was excluded when use of addiction services was the dependent variable. The u_si and v_sit are random variates representing each patient and each assessment, assumed to be independent and normally distributed with 0 mean and constant variance during the 1-year follow-up.

As with traditional models, the growth-rate effect (β₃₁)describes how outcome differences between CWT and controls grow, or decline, with time. The immediate effect (β₂₁) describes how a lump-sum difference in outcomes between CWT and control subjects may occur initially after active CWT. Thus, outcome difference between CWT and control subjects for t>0 equals β₂₁ + β₃₁t. Other parameters are the average starting value for control subjects (β₀); difference between CWT and control groups at baseline (t = 0) (β₁); initial "lump-sum" change in outcomes for control subjects (β₂₀); growth rate in outcomes for control subjects (β₃₀);use of care main effect (β₄); and fixed effects reflecting site-specific differences in average outcomes (α_s), with α_s = 0 for s = 1, the reference site.

The equation simplifies to a growth model when β₂₀ =0 and β₂₁ = 0 and can be modified to a logistic regression for dichotomous outcome variables.

Data collection began July 19, 1994, with recruitment ending March 22, 1996, and follow-up assessments ending April 1, 1997. Figure 1 describes the progress through the trial of the original 1090 subjects who had been discharged from VA addiction programs and whose names were referred for study. The large number of no-shows and patients unable to be located reflects difficulties inherent in studying homeless populations. The final sample (N = 142) was evenly distributed among the 4 sites (37%, 22%, 21%, and 20%) and included 22 (15%) who completed 2 follow-ups; 21 (15%), 3 follow-ups; and 99 (70%), 4 follow-ups.

The final sample did not differ from the 201 excluded eligible subjects(181 nonconsenting and 20 lost to follow-up) with respect to demographic characteristics(mean age, 43.1 vs 41.9 years, t = 1.44, P<.15; African American, 26% vs 25%, t= 0.22, P<.83; years of education, 12.7 vs 12.6, t = 0.21, P<.84), cost of VA care during the year before baseline ($26 199 vs $25 133, t = 0.41, P<.68), and health status (ASI for alcohol problems, 0.36 vs 0.36, t= 0.03, P<.98; ASI for drug problems, 0.11 vs 0.12, t = 1.22, P<.23; physical symptoms relating to substance use, 3.52 vs 3.53, t = 0.04, P<.97; ASI psychiatric status, 0.39 vs 0.38, t = 0.25, P<.81; Brief Symptom Inventory global status, 0.002 vs 0.001, t = 0.16, P<.87; SF-36 Mental Functioning, 50.2 vs 50.6, t = 0.16, P<.87; SF-36 Social Functioning, 58.9 vs 57.8, t = 0.30, P<.77; ASI edical Status, 0.35 vs 0.29, t = 1.39, P<.17; and SF-36 Physical Functioning, 85.9 vs 84.2, t = 0.77, P<.44).

Despite a 12% loss to follow-up, Table 1 shows no statistically significant difference between the CWT(n = 111) and control (n = 31) groups for demographic characteristics and outcome measures assessed at baseline. During the year after baseline, both CWT and control groups had extensive use of health services, including psychiatric and rehabilitation care, although differences reached statistical significance only in inpatient medicine days and outpatient addiction visits.

Subjects assigned to CWT received a job within a median 6 days after baseline and worked a mean of 393 hours (SD, 474 hours) at an average wage of $4.04/h (SD, $1.80/h; range, $0.50/h to $10.75/h). Of hours worked, 58% were unskilled, 22% semiskilled, 16% skilled labor, and 4% clerical, sales, or technical related. Working subjects were screened for substance use an average 8.6 times (SD, 10.7), or approximately once every 20 work days. Among those tested, 44% had at least 1 positive screening, with 15% of screenings testing positive.

Utilization covariates (measured in dollars) were significant predictors of patient outcomes for ASI drug (β₄ = 0.003, t₃₉₃ = 2.25, P<.02) and alcohol(β₄ = 0.008, t₃₉₃ = 2.31, P<.02) use problems, psychiatric status (β₄ = 0.011, t₃₉₃ = 2.18, P<.03), number of substance use– related physical symptoms(β₄ = 0.18, t₃₉₃ = 3.85, P<.001), and SF-36 Mental Functioning (β₄ = −1.07, t₃₉₃ = 1.97, P<.049) and Social Functioning (β₄ = −1.88, t₃₉₃ = 2.53, P<.01).

The CWT immediate and growth-rate effects are presented in Table 2 for each of the 12 outcome measures as a percentage of average baseline scores (continuous) or as odds ratios(dichotomous). Consistent with expectations, CWT clients were initially 2.7 times more likely to use addiction treatment than their control counterparts. After adjustment for greater use of addiction treatment with ±SE, patients who received CWT experienced immediate reductions in drug (−44.7% ± 12.8%) and alcohol (−45.4% ± 9.4%) use problems, and the number of substance use–related physical symptoms (−64.4% ± 8.0%). These immediate differences tended to decline with time, although such a "catch-up" pattern was statistically significant for only the number of substance use–related physical symptoms. Here, group differences declined by 17.0% ± 5.1% per quarter as control subjects, unlike their CWT counterparts, experienced a reduction in symptoms during follow-up by an average of −19.2% ± 4.4% per quarter.

There were no significant difference in psychiatric status between the CWT and control groups on any of the 4 outcome measures. On the other hand, differences between CWT and control groups tended to grow during follow-up by −24.4% ± 8.7% per quarter on the basis of ASI Medical Status and 6.9% ± 2.0% per quarter on the basis of SF-36 Physical Functioning. However, these gains were due primarily to declining function among control subjects (ASI Medical Status, 23.9% ± 7.6% per quarter; SF-36 Physical Functioning, −6.2% ± 1.8% per quarter) rather than from improved functioning among subjects in the CWT group (ASI Medical Status, −0.4% ± 11.5% per quarter; SF-36 Physical Functioning, 0.6% ± 2.7% per quarter). Finally, patients in the CWT group were initially 30% and 10% as likely to report an episode of incarceration and homelessness, respectively, as their control counterparts.

In this randomized, controlled trial, we evaluated a clinician-supervised, performance-based work program that tied paid work opportunities (hours, wages, and responsibilities) to performance (productivity, punctuality, and reliability) and health behaviors (sobriety and use of addiction services). Among homeless, substance-dependent persons with expressed desires to work, CWT was associated with greater use of addiction services. Even after adjustment for use of health and rehabilitative services, these data suggest that CWT clients experienced fewer problems associated with substance use and fewer episodes of homelessness and incarceration. The CWT clients also did not report the same decline in medical status and physical functioning exhibited by control subjects. However, no work effect on psychiatric status was detected. These differences are remarkable given that both groups had extensive use of medical, psychiatric, addictions, and rehabilitation services from local VA medical centers. However, patients in the CWT group reported generally more inpatient days and outpatient visits, suggesting that these outcomes may come at the price of higher health care costs.

These findings are consistent with those of the Milby et al study⁴⁰ of substance-abusing homeless persons, in which abstinent contingent work therapy was found to be associated with 36% fewer positive cocaine toxicologic tests within 2 months of beginning treatment. These results are also consistent with the findings of Bell et al^41- 42 that paid vs unpaid work at VA medical facilities was associated with symptom reductions for patients with schizophrenia or schizoaffective disorders, and with Rosenheck and Seibyl's⁴³ observational study that showed that receiving care at a residential support facility offering therapeutic work, but requiring responsible behavior, was associated with substance abuse, but not psychiatric symptoms. These results contrast with findings from assertive community treatment programs for homeless substance abusers that failed to find high rates of abstinence,⁴⁴ and from supported employment in competitive work environments that failed to find symptom reductions.^33,45 However, these latter studies focused on severe mental illness.

It is unknown which CWT elements may have contributed to program outcomes: structured work, pay incentives for mandatory substance screenings and adherence to addiction treatment schedules, clinician supervision, and improved access to care. Comparing other studies, Ames and Janes⁴⁶ found that boring, stressful, and isolating work can in fact contribute to substance use behaviors, while alcoholic subjects who returned to work were likely to remain sober, but only when the job discouraged drinking behaviors and offered structured supervision.^47- 48 A recent econometric analysis (2-stage least-squares adjusting for self-selection) of national survey data (1991 and 1993 National Health Interview Surveys) suggested that workplace bans were associated with a 5–percentage point reduction in smoking prevalence and a 10% reduction in daily consumption among smokers, and accounted for all of the explained variance in smoking reduction among workers relative to nonworkers.⁴⁹ Additional study is needed to untangle how program elements contribute to outcomes. With the success of supported employment programs, however, future studies should consider how outcomes vary between clinical and competitive employment settings.

There are study limitations. Findings were limited to a VA setting, 142 subjects, 1 year of follow-up, a low consent rate (47%), and a 1:4 sampling ratio that further limited statistical power. Outcomes measures were based on patient self-reports, although between-group biases may be small. First, CWT drug screenings averaged once every 20 days, so that hiding information from independent research teams may have had little perceived consequence on work opportunities. Second, control subjects wanting to enter CWT after completing the study had incentives to also underreport symptoms. Third, reduced substance consumption was consistent with patterns observed for other outcomes, including use of addiction services, substance use–related symptoms, medical status, and other measures.

Biases associated with nonrandomly occurring missing data and variable intervals between repeated measures may also be small. Compared with control subjects, CWT clients had a comparable number of observations (3.6 vs 3.8; t = 1.44, P<.15) and time between observations (1.5 vs 1.4 quarters; t = 1.26, P<.21). Regression to the mean bias either works against study findings (symptom counts) or is small compared with estimated effect size (drug and alcohol use problems).

In summary, the VA's CWT program was associated with better addiction status and with fewer episodes of homelessness and incarceration, and appears to prevent further deterioration in medical status and physical functioning. More research is needed to assess long-term impacts and to determine the active program ingredients contributing to outcomes.

Submitted for publication December 3, 1999; final revision received December 14, 2001; accepted December 20, 2001.

This study was supported in part by the Department of Veterans Affairs, Health Services Research and Development Service grant IIR 92-043 and Research Career Scientist Award 92-403, Washington, DC, and by grant 5-R24-MH53799 from the National Institute of Mental Health, Bethesda, Md.

We thank Mary Kashner, Pauline Salyer, Perry Lambird, MD, Mona Sue Lambird, JD, George Moore, Robert D. Fowler, MD, Stephen Berman, MSW, and A. John Rush, MD, for their special contributions in advancing this study.

The opinions expressed herein are those of the authors and do not necessarily reflect the views of supporting agencies and sponsors.

Corresponding author and reprints: T. Michael Kashner, PhD, JD, Department of Psychiatry, The University of Texas Southwestern Medical Center at Dallas, 5323 Harry Hines Blvd, Dallas, TX 75390-9086.