Predicting the Risk of & Time to Impairment for ALS patients: Report for the Lab on Intelligent Disease Progression Prediction at CLEF 2022
Résumé
This report details our participation at the Intelligent Disease Progression Prediction (iDPP) track at the Conference & Labs of the Evaluation Forum (CLEF) 2022. This task focuses on the progression of Amyotrophic Lateral Sclerosis (ALS), a progressive neurodegenerative disease that affects nerve cells in the brain and spinal cord. The goal of this work is to use patient demographic data & certain medical history details along with collections of records of responses to an ALS diagnostic questionnaire to calculate risk scores corresponding to the likelihood that a patient will suffer an adverse event, and to predict the time window in which that event will occur. We present an approach based on ensemble learning, in which gradient-boosted regression trees are used to separately predict risk scores and estimate survival times. By normalising & thresholding the risk scores, we generate event predictions which are combined with the time-to-event predictions to produce time-interval predictions. While some aspects of the results seem encouraging, especially given the amount of training data available, it is clear that more sophisticated and specialised solutions are required in order for techniques like these to become a reliable part of clinical decision-making.
Origine : Fichiers éditeurs autorisés sur une archive ouverte