Modifiable Predictors of Supported Employment Outcomes Among People With Severe Mental Illness

The study was registered as a clinical trial (NCT00895258) and data were collected from June 2008 to February 2014. Study procedures were approved by the University of California, San Diego, Institutional Review Board, and all participants provided written informed consent. One hundred fifty-three unemployed outpatients with severe mental illness (schizophrenia or schizoaffective disorder, N=58; bipolar disorder, N=37; and major depressive disorder, N=58) enrolled in the trial. Inclusion criteria were age 18 and older; literate and fluent in English; DSM-IV diagnosis of schizophrenia, schizoaffective disorder, bipolar disorder, or major depressive disorder confirmed via Structured Clinical Interview for DSM-IV (24) or Mini International Neuropsychiatric Interview (25); and being unemployed for at least 30 days and being interested in working. All participants received IPS for up to 2 years, based on their preference, and were randomly assigned to also receive either compensatory cognitive training (N=77) or additional supported employment sessions (enhanced supported employment; N=76) for the first 12 weeks of the trial (see Twamley et al. [26] for further details). Each group had its own employment specialist. Fidelity to supported employment was rated as “fair” during the study period. Because work outcomes associated with compensatory cognitive training and enhanced supported employment did not differ (26), the groups were collapsed for all analyses. Table 1 provides the demographic and clinical characteristics of the sample.

Baseline neuropsychological, clinical, and functional assessments were used in the current analyses. Data from these participants have been used in prior publications (26–30); however, the analyses presented in this article have not been published previously.

The following assessments were administered at baseline, prior to randomization. All raters were trained to a high degree of interrater reliability (intraclass correlation coefficient ≥.90).

Premorbid intellectual ability was estimated with the reading subtest of the Wide Range Achievement Test-III (31). The cognitive subtests of the Measurement and Treatment Research to Improve Cognition in Schizophrenia Consensus Cognitive Battery (32) assessed neuropsychological functioning in the domains of processing speed (Trail Making Test, Part A [TMT-A], Brief Assessment of Cognition in Schizophrenia Symbol-Coding [BACS-SC], and category fluency), sustained attention (Continuous Performance Test–Identical Pairs), working memory (Wechsler Memory Scale-III Spatial Span and University of Maryland Letter-Number Span [LNS]), verbal learning (Hopkins Verbal Learning Test–Revised [HVLT-R]), visual learning (Brief Visual Memory Test–Revised [BVMT-R]), and executive functioning (Neuropsychological Assessment Battery [NAB] Mazes). All t test scores were corrected for age and education. Additional tests of executive functioning measured set shifting (Trail Making Test, Part B [TMT-B] [33]), letter fluency using the letters F, A, and S (FAS) (33), and reasoning and set shifting (Wisconsin Card Sorting Test–64-card version [WCST-64] [34] total errors). Additionally, prospective memory ability was measured by using the Memory for Intentions Screening Test (35). We calculated a global deficit score (GDS) by transforming individual neuropsychological test scores (i.e., t test scores) to deficit scores ranging from 0, no impairment, to 5, severe impairment, which were then averaged across all tests (36).

The University of California San Diego Performance-Based Skills Assessment–Brief (37) assesses performance-based functional capacity in the domains of financial management and communication. The Social Skills Performance Assessment (38) measures social skills relevant to neutral and adversarial situations. Positive and negative symptom severity and general psychopathology were measured with the Positive and Negative Syndrome Scale (39), and depressive symptom severity was measured by using the Hamilton Depression Rating Scale (40).

Work outcomes (competitive job attainment, total weeks of competitive employment, and wages earned) were gathered weekly during the 2-year study by the employment specialist (if engaged in weekly contact with the participant) or a research assistant; work participation and earnings were corroborated with paystubs. Competitive work was defined as employment paying at least minimum wage and not set aside for a person with a disability. Participants who dropped out of the study prior to obtaining a job were assumed not to have worked.

One-way analysis of variance and chi-square tests were conducted to examine differences in characteristics between diagnostic groups. Prior to analyses, model assumptions were checked, including screening for outliers and evaluating for multicollinearity, with tolerance values of <.40 and variance inflation factor values of >2.5 suggestive of multicollinearity (41). Job attainment during the 2-year study period (0=no; 1=yes) was analyzed by using a logistic regression model. Examination of variable distributions for competitive weeks worked and wages earned showed positively skewed distributions, with excess zeroes. Thus these variables were log-transformed before being included in all analyses and modeled by using a zero-altered count regression approach, known as a hurdle model. Hurdle models are two-part models, in which all the zeroes are modeled with a probit regression, and nonzero counts are modeled by a truncated count regression (i.e., truncated because it does not include zero) (42). Our hurdle models reflected the two-stage process resulting in the observed distributions of competitive weeks worked and wages earned. That is, participants first had to attain a job (i.e., pass the “hurdle”) to report weeks of employment and wages earned. Thus the probit regression component of the hurdle model examined the predictors of job attainment, whereas the count regression in the hurdle models examined weeks worked and wages earned for those who attained a job (N=72). Analyses were conducted by using SPSS. version 24.0, except for analyses of hurdle models, which were conducted by using STATA/IC, version 15.0.

Bivariate Pearson and point-biserial correlations between individual tests of neuropsychological functioning (i.e., neuropsychological measures, including premorbid intellectual ability), GDS, psychiatric symptom severity, performance-based functional capacity and social skills, demographic variables, and work outcomes, were conducted. Bivariate-significant correlates of job attainment (p<0.05) were entered as predictors of job attainment in the logistic regression model as well as the probit component of the hurdle models. Similarly, for the subset of participants who attained a job, bivariate-significant variables were entered as predictors of weeks of competitive work and wages earned for the count regression models. (Although work history was not significantly associated with weeks worked, it was entered in the model because of computational requirements of hurdle model, i.e., algorithm limitations. Including work history did not affect model estimates. Our model excluding work history converged using an alternative linear maximum likelihood model [i.e., zero-inflated negative binomial model], and had similar results, thereby bolstering the robustness of the hurdle models.) There were significant demographic differences by diagnostic group in years of education and premorbid IQ estimate (p<0.05), which were controlled for in subsequent analyses.