Related Experiment Video
Updated: May 28, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Development and Validation of a Machine Learning Model for Predicting Medication Adherence Among Home-Dwelling
Yujie Zhang1,2, Yongli Han2, Xuemei Yin3
1Department of Pharmacy, Beijing Tiantan Hospital, Capital Medical University, Beijing, People's Republic of China.
Background:
Elderly chronic disease management is often complicated by multimorbidity and the need for lifelong polypharmacy, yet poor adherence severely impedes progress. Existing studies mainly focus on exploring influencing factors of medication adherence, while most predictive models lack model interpretability and rarely integrate psychological and social support factors for community elderly populations. Predictive models identifying risk of adherence could enable proactive intervention.
Objective:
To develop an interpretable machine learning prediction model that fills the above research gap to predict the medication adherence of elderly patients with chronic diseases in China.
Methods:
From January to December 2024, data were collected from chronic disease patients aged 60 years and older receiving home-based medication therapy through face-to-face interviews conducted by pharmacists. Variables included demographic information, comorbidities, chronic diseases and medications information, medication adherence, self-efficacy in rational drug use, medication beliefs, social support, and medication literacy. The dataset was randomly divided into a training set and a test set at a 7:3 ratio. Multivariate logistic regression analysis was performed on all data, and predictors were selected from the training set via the Least Absolute Shrinkage and Selection Operator (LASSO). Six machine learning algorithms were applied in R software to develop predictive models using the training set, and their performance was compared on the test set. The Shapley Additive Explanations (SHAP) approach was used to interpret the optimal model.
Results:
A total of 1722 patients were included in the statistical analysis. The gradient boosting machine (GBM) exhibited the best predictive performance among the six models (AUC = 0.811, 95% CI 0.774-0.840), with its core predictors being self-efficacy in rational drug use, medication practice, concern beliefs, and availability of social support. Through SHAP analysis, the interpretability of the model was significantly enhanced, providing a clear decision-making basis for clinicians.
Conclusion:
We constructed a prediction model for home medication adherence in elderly patients with chronic diseases, which incorporates important social and psychological factors affecting patients' adherence and provides robust evidence for developing targeted interventions.
