Interpretable Machine Learning in Healthcare through Generalized Additive Model with Pairwise Interactions (GA2M): Predicting Severe Retinopathy of Prematurity

Abstract

We have investigated the risk factors that lead to severe retinopathy of prematurity using statistical analysis and logistic regression as a form of generalized additive model (GAM) with pairwise interaction terms (GA2M). In this process, we discuss the trade-off between accuracy and interpretability of these machine learning techniques on clinical data. We also confirm the intuition of expert neonatologists on a few risk factors, such as gender, that were previously deemed as clinically not significant in RoP prediction.