Interpretable Predictive Modeling for Mucopolysaccharidoses Early Diagnosis

Mucopolysaccharidoses (MPS) are rare, chronic, and progressive metabolic disorders caused by deficiencies in lysosomal enzymes responsible for degrading glycosaminoglycans. These disorders are characterized by significant clinical heterogeneity and low prevalence, which often hinder timely and accurate diagnosis, an essential factor for initiating effective treatment and improving patient outcomes. While machine learning (ML) has shown promise in advancing disease diagnosis, many models lack interpretability, limiting their clinical adoption. This study addresses the challenge by developing an interpretable ML framework to predict MPS cases using real-world electronic health records (EHRs) from 106 subjects, including MPS patients and controls, provided by the SEHA Abu Dhabi Health Services network. The proposed framework explores and compares several rule-based interpretable ML models across multiple feature sets, selected using both automated feature selection techniques and a domain expert-driven selection approach. The best-performing Fast Interpretable Greedy-Tree Sums model achieved an Area Under Curve (AUC) of 0.964, F1-score of 0.911, and accuracy of 90.91%. Additionally, the model featured clinically relevant indicators that align with known clinical manifestations of MPS, such as "acute pharyngitis,""dental caries,"and "hepatomegaly". The proposed framework achieved high diagnostic performance and provides transparent clinical reasoning, supporting its potential in AI-driven rare disease diagnosis.