ADDRESSING CLINICAL AND RESEARCH INERTIA USING MACHINE LEARNING APPROACHES: THE HOMED-BP STUDY

Abstract

Objective: The failure or reluctance to adequately adjust antihypertensive therapy despite uncontrolled blood pressure (BP) in outpatients—clinical inertia—remains a significant barrier to effective BP management. Based on the clinical trial dataset and related surveys of participating doctors, we employed machine learning (ML) approaches to identify latent variables that interfere with achieving target BP levels, aiming to enhance understanding of treatment barriers and improve clinical decision-making. Design and method: Data were derived from the Hypertension Objective treatment based on Measurement by Electrical Devices of Blood Pressure (HOMED-BP) study, a Japanese nationwide, randomized clinical trial. Although the study successfully demonstrated the clinical usefulness of self-measured home BP, we suffered from the low achievement rate of home BP target levels during the follow-up period. To ensure a comprehensive screening of variables while minimizing bias and reducing the risk of overfitting, we employed five feature selection methods—Lasso, Ridge Regression, ElasticNet, Ordinary Least Squares (OLS), and Backward Elimination—to identify distinctive features (characteristics) potentially relevant to BP management. Model performance was evaluated based on predictive accuracy and interpretability, including clinical relevance. Results: We successfully identified 30 key characteristics impacting the achievement of target home BP levels while mitigating overfitting risks. Notable features include an initial systolic BP, concerns that the clinic BP has fallen too low, and patient disagreement with the target level. These variables were utilized to develop predictive models, i.e., Random Forest, Logistic Regression, and RuleFit. The accuracy of the three models ranged from 0.65 to 0.69. Conclusions: This study demonstrated the utility of ML in uncovering factors that hinder effective hypertension treatment. Appropriate actions for these latent factors that may contribute to clinical/research inertia can refine treatment strategies and advance personalized approaches to the long-term treatment of hypertension.

Ähnliche Arbeiten