Related Experiment Video
Updated: Jul 10, 2026
12:18
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Machine learning-based identification of medication adherence predictors in two Spanish primary care cohorts
Rodrigo Martín Gómez Del Moral Herranz1, Miguel Rujas1, Peña Arroyo-Gallego1
1Life Supporting Technologies Research Group, Universidad Politécnica de Madrid, Avda Complutense 30, 28040, Madrid, Spain.
Scientific Reports
|July 8, 2026
Summary
Identifying medication adherence factors is crucial for patient outcomes. Artificial intelligence identified key predictors in Spanish primary care, enabling personalized adherence strategies.
Area of Science:
- Pharmacoeconomics and Health Services Research
- Artificial Intelligence in Healthcare
- Clinical Pharmacy and Practice
Background:
- Medication non-adherence is a significant global issue impacting patient outcomes and increasing healthcare costs, especially in Spain with its high chronic disease rates.
- Identifying factors influencing adherence is essential for developing effective, targeted interventions to improve patient care and reduce economic burden.
Purpose of the Study:
- To analyze large-scale real-world primary care data from Madrid and Catalonia to identify key factors influencing medication adherence.
- To evaluate the performance of various machine learning models and feature selection methods in predicting medication adherence.
Main Methods:
- Utilized four feature selection methods and three machine learning classifiers on two primary care databases.
- Employed Recursive Feature Elimination with Cross-Validation (RFECV) and Extreme Gradient Boosting (XGBoost) for optimal performance.
- Conducted threshold optimization, calibration analysis, bootstrap confidence intervals, and importance analyses for robust results.
Main Results:
- Identified 48 structural factors influencing medication adherence across both cohorts (19 in Madrid, 29 in Catalonia).
- Achieved consistent validation and test performance with AUROC values of 0.6953/0.6952 (Madrid) and 0.7775/0.7788 (Catalonia).
- Key factors varied by region: number of medications and chronic disease burden in Madrid; smoking, rural/urban context, and prescription timing in Catalonia.
Conclusions:
- Artificial intelligence effectively identifies patterns in medication adherence, supporting the development of patient-centered strategies.
- Findings highlight the need for context-specific interventions, considering regional differences in adherence predictors.
- Future research should address data limitations, including social determinants and external validation, to enhance generalizability.
