Employment outcome prediction after traumatic brain injury

Traumatic brain injury (TBI) leads to long-term effects on individuals' daily functioning. This can be manifested in a wide range of cognitive, physical, sensory, emotional and behavioral changes. These functional changes can complicate reintegration into society after initial care and rehabilitation. Return to work is one aspect of this reintegration that is particularly important given that a large share of individuals with TBI is in the active working population. Moreover, return to work plays a central role for the individual in financial stability, daily activities, self-esteem, social integration and quality of life. For society, employment outcomes are also important because of the large social cost associated with long-term disability and loss of human potential.
Because of the large heterogeneity in clinical manifestation inherent to TBI, it is also difficult for clinicians to form an accurate prognosis about the likelihood that a person will return to work. In practice, however, such predictions could help to set realistic expectations among patients and those around them, as well as appropriate goals for rehabilitation. At societal level, individualized predictions about the probability of returning to work could provide the basis for identifying people who need greater access to intervention programs and support services.
The application of statistical tools and algorithms to large datasets on the target population can lead to accurate clinical prediction models. This PhD examines the feasibility of individualized predictions about employment outcomes after TBI with existing methods and datasets and the modalities of such a prediction model. At the methodological level, we try to keep a good middle ground between optimizing the predictive performance of the models on the one hand as well as their interpretability and transparency on the other. Several datasets are used, each with their own characteristics and different advantages and disadvantages: a Belgian administrative dataset (N=5718), a European research dataset with a focus on acute care (N=4509) and an American research dataset with a focus on rehabilitation (N=19477). For these studies, employment outcomes were always considered at a one-year time horizon. The models created were examined based on the predictive performance obtained and the variables most influential to the predictions made.
At the beginning of this research project, a systematic literature review was conducted on the known predictors of return to work after TBI. This revealed that person-related factors and injury factors have already been studied frequently. However, the situation is completely different for functional factors, for which little evidence is available, and even less so for environmental factors.
The prediction model created using administrative data was found to contain insufficient data on the patient's functional status to make good predictions. The European prediction model achieved a good balance between positive and negative predictive value, which is important in terms of correctly informing the patient. The American model provided a better balance between correctly identifying a large proportion of the group that will not return to work, without unfairly labeling too many people that way.
Regarding the predictors of return to work, age and factors related to employment before TBI were systematically the most important person-related factors. Duration of post-traumatic amnesia and duration of acute hospitalization were found to be the most predictive indicators of injury severity. Not unexpectedly, assessments for general functional independence were of great importance in predicting employment outcomes. The U.S. model identified the payer source for the received care as the only relevant environmental factor in this study.
For each of the models, a significant portion of the variation in employment outcomes still remains unexplained. While further advancements in sophisticated modeling techniques are possible, it seems more likely that the explanation primarily lies in variables that were not accounted for in the datasets we used. This could include factors focused on the individual's experiences after TBI, as well as factors related to workplace organization. In addition, caregivers may also play a role in the likelihood of reemployment. Finally, the potential influence of the healthcare system should not be underestimated in relation to the patient's and employer's incentives for successful return to work. An exploratory focus group study was conducted as part of this PhD to gain a better understanding of the latter factors.
This PhD concludes with some concrete recommendations for future research in this field. The proposed approach first seeks to gain more insight into employment outcomes after TBI, before embarking on the new data collections that will be necessary for more accurate predictions. In addition, practical recommendations are also made for other social actors to improve positive and sustainable employment outcomes after TBI.