Survival Prediction with Limited Features: a Top Performing Approach from the DREAM ALS Stratification Prize4Life Challenge

TL;DR

This paper presents a highly accurate survival prediction method for ALS patients using limited, engineered features derived from longitudinal data, demonstrating its clinical relevance and confirming key prognostic factors.

Contribution

The study introduces a novel feature engineering approach and a survival model that outperforms many existing methods in predicting ALS patient survival with limited data.

Findings

Engineered features significantly improve prediction accuracy.

Model ranks second in the DREAM ALS Challenge.

Key predictors include ALSFRS score, disease duration, onset site, and age.

Abstract

Survival prediction with small sets of features is a highly relevant topic for decision-making in clinical practice. I describe a method for predicting survival of amyotrophic lateral sclerosis (ALS) patients that was developed as a submission to the DREAM ALS Stratification Prize4Life Challenge held in summer 2015 to find the most accurate prediction of ALS progression and survival. ALS is a neurodegenerative disease with very heterogeneous survival times. Based on patient data from two national registries, solvers were asked to predict survival for three different time intervals, which was then evaluated on undisclosed information from additional data. I describe methods used to generate new features from existing ones from longitudinal data, selecting the most predictive features, and developing the best survival model. I show that easily obtainable engineered features can significantly improve prediction and could be incorporated into clinical practice. Furthermore, my prediction model confirms previous reports suggesting that past disease progression measured by the ALSFRS (ALS functional rating scale score), time since disease onset, onset site, and age are strong predictors for survival. Regarding prediction accuracy, this approach ranked second.