Integrating administrative health data and machine learning to predict ALS onset

Area of Science:

• Medical Informatics
• Computational Biology
• Public Health

Background:

• Amyotrophic Lateral Sclerosis (ALS) diagnosis presents challenges, necessitating novel predictive approaches.
• Administrative healthcare data offers a rich resource for identifying disease patterns.
• Early detection of ALS is crucial for timely intervention and patient management.

Purpose of the Study:

• To develop and evaluate a Machine Learning (ML) model for predicting the initial diagnosis of Amyotrophic Lateral Sclerosis (ALS).
• To identify key clinical and demographic factors associated with early ALS diagnosis.

Main Methods:

• A stacked ensemble model was developed, integrating logistic regression, decision tree, random forest, and extreme gradient boosting algorithms.
• The model utilized comprehensive healthcare administrative data from 2,924,590 elderly individuals in Catalonia (2014-2021).
• Data linkage included socioeconomic factors and medication records to enhance predictive capabilities.

Main Results:

• The stacked ML model achieved high predictive performance with an AUC of 0.86, accuracy of 0.86, specificity of 0.88, and sensitivity of 0.84.
• Key predictors included immunization encounters, South American origin, general and special examinations, hypertensive heart disease, and counseling.
• Other significant features comprised sciatica, heart failure, liver metastases, healthcare utilization, and chronic conditions like hypertension and kidney disease.

Conclusions:

• Stacked Machine Learning models demonstrate significant potential for predicting ALS diagnoses from administrative health data.
• Identified predictors highlight the importance of early clinical symptoms and healthcare-seeking behaviors in ALS detection.
• Further research is warranted to refine these models and facilitate their integration into clinical practice for improved ALS diagnostic strategies.