Cook et al 2007 — Effects of Co-occurring Disorders on Employment Outcomes in a Multisite Randomized Study of Supported Employment

Summary

Uses the full Employment Intervention Demonstration Program (EIDP) sample ($n = 1{,}273$, 7 sites, 24-month follow-up) to examine how co-occurring medical, physical, and cognitive conditions moderate four vocational outcomes: total earnings, total hours worked, competitive employment, and workplace disclosure of psychiatric disability. Co-occurring conditions affected outcomes substantially but not uniformly: physical comorbidities were most damaging, substance abuse had null effects, and supported employment (SE) remained effective across comorbidity groups on all outcomes except disclosure.

Key Claims

• Comorbidity prevalence: $59.9\%$ of participants had at least one co-occurring condition; $25\%$ had two or more; $35.4\%$ had substance abuse/dependence; $12.3\%$ intellectual disability; $11.2\%$ head injury; $10.5\%$ visual impairment
• Any comorbidity: lower total earnings ($\beta = -0.06^{*}$) and lower odds of competitive employment (odds ratio (OR) $= 0.76^{*}$) vs. no comorbidity
• Physical comorbidities only: lower earnings ($\beta = -0.06^{*}$), fewer hours ($\beta = -0.06^{*}$), significantly lower odds of competitive employment (OR $= 0.48^{\dagger}$, $p < 0.01$)
• Specific condition effects on competitive employment: intellectual disability OR $= 0.62^{*}$, visual impairment OR $= 0.53^{*}$, HIV/AIDS OR $= 0.33^{*}$; chronic/acute medical condition OR $= 2.23^{*}$ (unexpectedly higher — authors note this may reflect selection)
• Substance abuse/dependence: null effect on all four vocational outcomes — consistent with multiple prior studies; substance use should not be an exclusion criterion for SE programs
• Disclosure: individuals with both physical and cognitive comorbidities were $2\times$+ more likely to disclose psychiatric disability at work; learning disability associated with $3.5\times$ disclosure odds — disclosure is more a workplace environment measure than a performance measure
• Experimental condition: significant advantage in 3 of 4 models (earnings, hours, competitive employment); disclosure was the exception — high in both conditions ($71\%$ control, $75\%$ experimental), suggesting disclosure is common regardless of SE model
• Number of comorbidities (count): not significantly associated with any outcome — presence/absence and type matter more than count

Concepts Introduced or Extended

• Supported Employment

Entities Mentioned

• Judith A. Cook
• Employment Intervention Demonstration Program

Quotes

"Individuals with severe mental illness and co-occurring disorders have poorer employment outcomes than those without comorbidities, and their psychiatric status is more likely to be disclosed to others in the workplace. Even though EBP supported employment services are effective for this population, knowledge of these relative disadvantages can be used for the enhancement of services for people with comorbidities by the tailoring of existing models."

"Our findings regarding the absence of effects of substance abuse/dependence on vocational outcomes were not surprising, given that prior studies of individuals receiving return-to-work services reached similar conclusions."

My Take

The null substance abuse finding is the most counterintuitive and best-replicated result, with multiple prior studies reaching the same conclusion. This has direct policy implications: requiring sobriety before SE participation (common in some programs) is unsupported by evidence. The physical comorbidity result (OR $= 0.48$) is clinically important — psychiatric rehabilitation programs are rarely equipped for physical disability — but the study cannot distinguish whether the effect operates through functional limitation, stigma, time constraints, or unmeasured severity. The "chronic/acute medical condition → higher competitive employment" anomaly likely reflects selection: workers actively managing a treatable condition (arthritis, diabetes) may be more motivated or higher-functioning than the base comparison group. The ordinary least squares (OLS)/logit design is appropriate given the experimental randomization, though comorbidity coding from chart review at baseline may undercount conditions.