Predicting Non-invasive Ventilation in ALS Patients Using Stratified Disease Progression Groups

Abstract

Amyotrophic Lateral Sclerosis (ALS) is a neurode-generative disease highly known for its rapid progression, leading to death usually within a few years. Respiratory failure is the most common cause of death. Therefore, efforts must be taken to prevent respiratory insufficiency. Preventive administration of non-invasive ventilation (NIV) has proven to improve survival in ALS patients. Using disease progression groups revealed to be of great importance to ALS studies, since the heterogeneous nature of disease presentation and progression presents challenges to the learn of predictive models that work for all patients. In this context, we propose an approach to stratify patients in three progression groups (Slow, Neutral and Fast) enabling the creation of specialized learning models that predict the need of NIV within a time window of 90, 180 or 365 days of their current medical appointment. The models are built using a collection of classifiers and 5x10-fold cross validation. We also test the use of a Feature Selection Ensemble to test which features are more relevant to predict this outcome. Our specialized predictive models showed promising results, proving the utility of patient stratification when predicting NIV in ALS patients.