Integrating administrative health data and machine learning to predict ALS onset.

BACKGROUND: This study aims to develop a Machine Learning (ML) model to predict the initial diagnosis of Amyotrophic Lateral Sclerosis (ALS). METHODS: To predict ALS, a stacked model combining four ML algorithms-logistic Regression, Decision Tree, Random Forest, and Extreme Gradient Boosting-was implemented. The analysis utilized healthcare administrative data from Catalonia, encompassing 2,924,590 elderly individuals from 2014 to 2021, which were linked to socioeconomic factors and medication records. RESULTS: The stacked model successfully predicted first-time ALS diagnoses, achieving an AUC of 0.86, with an accuracy of 0.86, specificity of 0.88, and sensitivity of 0.84. The most influential predictors included immunization encounters, South American origin, general medical and special examinations, hypertensive heart disease, and counseling. Other relevant features were sciatica, heart failure, liver metastases, healthcare use patterns, and chronic conditions such as hypertension, kidney disease, and hypercholesterolemia. These features reflect early clinical symptoms and healthcare usage patterns relevant to ALS detection. CONCLUSIONS: Machine Learning models, particularly stacked approaches, show promising results in predicting ALS diagnoses using administrative health data. Continued research is necessary to improve detection strategies and support their integration into healthcare systems.