Lesion Volume as a Predictor of Return to Work After Stroke

Abstract

Introduction: Facilitating a successful return to work plays a pivotal role in reintegrating stroke survivors into the community. This study aims to explore the significance of lesion volume in the acute phase of stroke as a predictor for returning to work. Methods: The study population consisted of patients <65 years undergoing thrombectomy at Rikshospitalet and included in the ongoing ''The Oslo Acute Reperfusion Stroke Study'', Oslo University Hospital from January 1, 2017 to May 2019. Clinical examinations and MRI scans of the brain were conducted at admission, and employment status was obtained at the start and end of follow-up. Stroke lesion volumes were segmented using ITK Snap and DWI images from the MRI scan. A logistic regression model was conducted using R, to evaluate the impact of lesion volume on work- status after 3 years, adjusted for age, sex, vascular territory and education. We also explored an explainable machine learning model, Extreme Gradient Boosting, using Python and scikit-learn to investigate the impact of clinical and imaging variables on the outcome. Results: Of the 108 patients included in the study, 71 (66%) completed the three year follow-up. Of these, 57 were occupationally active at the stroke onset, with MRI available for stroke lesion segmentation, and are included in this analysis. Median age at study inclusion was 52 years (IQR 46-58), 51% were female, and median stroke volume on admission was 16 ml (IQR 7-32). Three years after the stroke, 34 participants (60%) had successfully returned to work. Risk of not returning to work increased by 4% for every 1 ml increase in lesion volume (p-value<0.01).Notably, the most influential features in the ML model for predicting return to work were lesion volume and reperfusion grade (eTICI), with feature importance 20% and 14% respectively. Discussion: This study explores the association between lesion volume and returning to work after stroke. ML models including measures of lesion volume may be helpful in prediction of post-stroke outcomes, and as guiding for intervention aiming to regain pre-stroke activity levels.