Cognitive,Emotional,and Physical Functioning as Predictors of Paid Employment in People With Stroke,Traumatic Brain Injury,and Spinal Cord Injury

Participants and Procedures

As part of a larger project, 480 community-dwelling adults with neurological disorders (SCI, n = 176; TBI, n = 155; stroke, n = 149) participated in a comprehensive 2-day assessment at three medical institutions in the midwestern United States. We recruited participants via patient registries, outpatient clinics, and affiliated hospital referrals. Eligibility criteria were community residence, at least 1 yr since the most recent neurological injury, ages 18–62 yr (i.e., of typical working age), and the ability to speak English. We chose an upper age of 62 yr because that is the earliest age for Social Security retirement benefits. Although we chose an upper age limit, we recognize that continuing paid employment has psychosocial and financial benefits for many workers who are older than the eligibility age for full Social Security retirement benefits (Fraser et al., 2009).

We reviewed medical records to verify eligibility. TBI eligibility criteria included a diagnosis of complicated mild, moderate, or severe TBI (Traumatic Brain Injury Model Systems National Data Center, 2006). We used emergency department Glasgow Coma Scale (GCS; Teasdale & Jennett, 1974) scores and positive neuroimaging results to confirm TBI severity; we defined severity as the lowest GCS score within the first 24 hr of injury not due to intubation, sedation, or intoxication. Fifty-one participants (11%) had mild TBI (GCS score = 13–15), 15 (3%) had moderate injury (GCS score = 9–12), 88 (18%) had severe injury (GCS score ≤ 8), and 1 (<1%) had unknown severity. SCI eligibility criteria were an acute traumatic lesion of neural elements in the spinal canal resulting in either temporary or permanent sensory or motor disabilities. We characterized SCI as paraplegia or tetraplegia and as complete or incomplete according to the International Standards for Neurological Classification of SCI (Kirshblum et al., 2011). Forty-eight participants (10%) had complete paraplegia, 36 (8%) had incomplete paraplegia, 41 (9%) had complete tetraplegia, 50 (10%) had incomplete tetraplegia, and 1 (<1%) had unknown classification.

Stroke eligibility criteria were medically documented, rapidly developing clinical signs of focal or global disturbance of cerebral function with symptoms lasting more than 24 hr and with apparent cause of vascular origin (Hatano, 1976). We used the Modified Rankin Scale (van Swieten et al., 1988) to classify stroke severity. Forty-one participants (9%) had mild strokes (score = 0–2), 41 (9%) had moderate strokes (score = 3), 66 (14%) had severe strokes (score = 4), and 1 (<1%) had unknown severity. We coded stroke etiology as either ischemic or hemorrhagic. Forty-three participants (9%) had a hemorrhagic stroke, and 101 (21%) had an ischemic stroke. To ensure sufficient vision, literacy, and communication skills to complete assessments, we added the following inclusion criteria: a Lighthouse Near-Visual Activity Test (Bailey & Lovie, 1976) score of less than 20/100, the ability to read the first 10 words of the Wide Range Achievement Test (Wilkinson & Robertson, 2006), and the ability to communicate some of the main ideas of a story in the Frenchay Aphasia Screening Test (Enderby & Crow, 1996).

Participants completed all measures under the guidance of a trained and certified tester with a graduate degree in psychology or occupational therapy who provided reasonable accommodations as needed. All testers underwent rigorous training procedures to ensure that testers at all sites followed a standard test administration protocol (Tulsky & Heinemann, 2017). We developed a training manual, provided in-person training for administering and scoring tests, and certified testers through direct observation of administration of the test battery to a practice participant. Annually, we maintained quality control through site visits, recertification, and double scoring of protocols by study investigators to ensure adherence and minimize drift from the protocol.

This study received ethics approval from the institutional review boards at each of the three collaborating sites, and all participants provided informed, written consent. Participants received an honorarium to acknowledge their research contribution.

Outcome Measure

We administered an occupational outcome questionnaire including items about return to work, time to return to work, job demands, satisfaction, earnings, and pace at work. It can be downloaded at https://wustl.box.com/s/3t6rlufhwd23u6q33hfiprrlcjp2pkuc. This measure was validated to determine work outcomes of people after a stroke (O’Brien & Wolf, 2010; O’Neill & Wolf, 2010). We used two items to operationalize paid employment: (1) “Did you return to work after an injury?” and (2) “Are you currently employed?” Participants who answered no for the first question and chose “employed in a volunteer position” for the second question were defined as not employed.

Physical Function

For mobility, participants indicated whether they used a mobility device and then specified the device they used most often. An occupational therapy practitioner with Assistive Technology Professional certification confirmed their classification as either wheelchair users or ambulators. We found this dichotomous classification to be useful for predicting participation outcomes of people with stroke, TBI, and SCI (Magasi et al., 2018). For upper extremity functioning, participants completed the Quality of Life in Neurological Disorders—Upper Extremity item bank (Cella et al., 2012), which was computer administered via the National Institutes of Health (NIH) Assessment Center^® (https://www.assessmentcenter.net/). The Upper Extremity item bank consists of 20 items. We chose it because it assesses one’s ability to carry out various activities involving digital, manual, and reach-related functions ranging from fine motor (e.g., “Are you able to turn a key in a lock?”) to ADL (e.g., “Are you able to button your shirt?”) functioning. Higher scores indicate better functioning. Excellent test–retest reliability (interclass correlation coefficient [ICC] = .88) and internal consistency (Cronbach’s α = .82) of this measure for adults with stroke had been established.

Emotional Function

Participants completed the NIH Toolbox (NIHTB) Emotion Battery (Salsman et al., 2013) administered via computer. We administered 15 measures of negative affect (i.e., anger–affect, anger–hostility, anger–physical aggression, fear–affect, fear–somatic, perceived stress, sadness), social satisfaction (i.e., emotional support, friendship, instrumental support, loneliness, perceived rejection), and psychological well-being (i.e., meaning, life satisfaction, positive affect). Higher scores indicate more of the trait.

Cognitive Function

On the basis of an earlier study (Baum et al., 2017), we characterized cognition in three domains: functional, fluid, and crystallized. Participants completed five fluid cognition tests from the NIHTB Cognition Battery (Weintraub et al., 2013): the Dimensional Change Card Sort Test, to measure cognitive flexibility; the Flanker Inhibitory Control and Attention Test, to measure inhibitory control; the Picture Sequence Memory Test, to measure episodic memory; the List Sorting Working Memory Test, to measure working memory; and the Pattern Comparison Processing Speed Test, to measure processing speed. They also completed two tests of crystallized cognition from the NIHTB Cognition Battery, including the Picture Vocabulary Test to measure receptive vocabulary and the Oral Reading Recognition Test to measure decoding skills. For all NIHTB Cognition Battery measures, uncorrected standard scores (mean [M] = 100, standard deviation = 15) were used, with higher scores representing better performance. We chose the NIHTB because it provides a brief, comprehensive assessment of motor, cognitive, emotional, and sensory functions and was normed on a nationally representative sample (Gershon et al., 2010). We reported evidence to support the clinical utility and construct validity of this tool in people with stroke, TBI, and SCI (Carlozzi et al., 2017).

The Executive Function Performance Test (EFPT; Baum et al., 2008) is a performance assessment designed to measure cognitive integration (or functional cognition). It is used to assess which executive function components are impaired as a person completes the following four tasks: simple cooking, telephone use, medication management, and bill payment. We rated the cues that participants needed to complete the tasks successfully while initiating, organizing, sequencing, using judgment and safety, and completing the tasks. Adequate to excellent interrater reliability (ICC range = .79–.94) and adequate internal consistency (Cronbach’s α range = .77–.88) were established for survivors with chronic stroke. Moderate correlations between EFPT and NIHTB Cognition Battery scores were found in TBI (rs = −.42, −.50), suggesting that the EFPT measures similar but distinct cognitive constructs from the NIHTB Cognition Battery (Baum et al., 2017). Higher scores on the EFPT indicate lower levels of executive function.

Data Analysis

We computed descriptive statistics to characterize the total sample (Table 1) and examined any differences in sociodemographic characteristics, physical function, emotion, and cognition between employed and unemployed groups. We used χ² tests for categorical variables, independent t tests for variables with a normal distribution, and Mann–Whitney U tests on variables with skewed distributions to test for group differences. To correct for the chance of Type I errors due to our use of multiple t tests, we adjusted the α level for each of these analyses from the standard p < .05 to p < .01. We also reported characteristics of participants who obtained postinjury employment to understand any job characteristics associated with taking full- versus part-time employment and working with same versus new employers postinjury (Table 2). Before regression analyses, we assessed the presence of any outlier by running boxplots. An outlier is considered when any value is larger than 3 times the interquartile range. We assessed the extent of collinearity by examining the variance inflation factor (VIF) to ensure that variables met the assumption of regression analyses. Collinearity is considered high when the VIF is greater than 10 (Kutner et al., 2004).

To account for unique variances in employment explained by physical, emotional, and cognitive domains, we performed a series of logistic regression models by domains in which we forced all sociodemographic variables as covariates in Step 1, and we added predictor variables within each different domain in Step 2. Multivariate odds ratios (ORs) in Table 3 show information on the independent domain (i.e., single-domain models). Recognizing the likely multifactorial nature of employment in people with neurological injuries, we entered variables in all domains into a full model, and we still entered all sociodemographic variables as covariates to account for variance in employment explained by all domains. Because variables in physical, emotional, and cognitive domains could be highly correlated, we chose forward selection because it allowed us to keep covariates that had stronger associations with the employment outcomes retained in the final model. We used IBM SPSS Statistics (Version 20; IBM Corp., Armonk, NY) for statistical analyses and set α at p < .05.

Characteristics of Participants Who Were Employed Postinjury

Among 480 participants, 122 participants (25%) returned to work after injury; 18% returned to work with their previous employer, including 10% who worked full time and 8% who worked part time with their previous employer. Most participants who were employed within 3 mo of injury returned to their previous employer (see Table 2): Fifty-five percent worked full time, and 38% worked part time with their previous employer, whereas 7% worked with a new employer. Participants who resumed work more than 1 yr after injury worked mainly with a new employer: 33% worked full time, and 48% worked part time, whereas 19% worked with their previous employer (p < .001). No statistical differences were found in other characteristics, such as ability to meet job demands, satisfaction with job performance, earnings, or the pace of work.

Function-Specific Domains Affecting Employment

In the baseline sociodemographic multivariate model (see Table 3), greater likelihood of employment was predicted by White race, being married or partnered, and having a college education, but it was not predicted by type of neurological injury, age, or number of years since injury. This baseline model explained 11% of the variance in employment. When demographic variables were entered as covariates, better upper extremity function predicted employment; these variables explained an additional 6% (total 17%) of the variance. For emotion domains, individual variables in the negative affect and social satisfaction domains did not predict employment; however, meaning and positive affect within the psychological well-being domain explained an additional 2% (total 13%) of the variance when demographic variables were entered as covariates.

Next, we evaluated three domains of cognition independent of physical and emotional function by entering demographic variables as covariates. Fluid cognition explained the greatest variance in employment (total R ² = 21%), followed by functional (total R ² = 17%) and crystallized (total R ² = 13%) cognition when demographic variables were entered as covariates. Organization and sequencing scores measured by the EFPT and inhibitory control measured by the NIHTB Flanker Inhibitory Control and Attention Test were significant predictors of employment.

Multifactorial Model of Employment

Table 4 shows results for the full model. Variables did not violate the assumption of collinearity because VIFs were less than 1.8. Greater likelihood of employment was predicted by being married or having a partner (OR = 2.21, p = .018) and having a college degree (OR = 2.64, p = .018). Cognition remained statistically significant, even after multifactorial adjustment. Participants requiring less assistance on sequencing tasks (OR = 0.48, p = .05) and demonstrating better inhibitory control (OR = 1.10, p < .001) predicted a greater chance of employment, whereas crystallized cognition was not statistically significant; it explained 25% of the variance in employment.

To ensure that we achieved enough power to run the regression, we used G*Power (Version 3.1.9.2; Heinrich Heine University Düsseldorf, Düsseldorf, Germany) for the power calculation, including 39 variables in the final regression model (Faul et al., 2009). Given our sample size (N = 480) and a medium effect size (f ² = 0.15), the study has 99.9% power to detect predictors at α = .05, suggesting that the sample size provides sufficient power for the analysis.