An unsupervised machine learning approach to predict recovery from traumatic spinal cord injury

BackgroundNeurological and functional recovery after traumatic spinal cord injury (SCI) is highly heterogeneous, challenging outcome predictions in rehabilitation and clinical trials. We propose k-nearest neighbour (k-NN) matching as a data-driven, interpretable solution. MethodsThis study used acute-phase International Standards for Neurological Classification of SCI exams to forecast 6-month recovery motor function as primary evaluation endpoint. Secondary endpoints included severity grade improvement, independent walking, and self-care ability. Different similarity metrics were explored for NN matching within 1267 patients from the European Multicenter Study about Spinal Cord Injury before validation in 411 patients from the Sygen trial. ResultsWe obtained a population-wide root-mean-squared error (RMSE) in motor score sequence of 0.76(0.14, 2.77) and competitive functional score predictions (AUCwalker=0.92, AUCself-carer=0.83). The validation cohort showed comparable results (RMSE = 0.75(0.13, 2.57), AUCwalker=0.92). Prediction performance in AIS grade B and C patients ([~]30%) showed the largest deviations from true recovery scores, in line with large SCI heterogeneity. ConclusionsOur approach provides detailed predictions of neurological and functional recovery based on a highly interpretable unsupervised machine learning concept. The k-NN matching strategy further enables the integration of historical control data into the evaluation of clinical trials and provides a data-driven digital twin for recovery trajectory exploration.