Medication history, not clinical profiles, predicts antihypertensive treatment persistence across health systems

A parsimonious antihypertensive medication-history score predicts treatment non-persistence across CHARLS, SHARE, and HRS, achieving AUC up to 0.85 and outperforming cross-sectional profiles.

• Why it matters: Hypertension control is undermined by patients stopping medication, yet existing ML persistence models are often single-country and lack external validation against simpler clinical information.
• What they did: Using adults aged ≥50 with doctor-diagnosed hypertension, the study harmonized three prospective cohorts and predicted next-wave non-use using baseline history features and a 3-feature medication-history score, then tested direct transport without retraining.
• The result: Medication-history models raised AUC from 0.59–0.67 to 0.78–0.85 within cohorts and retained AUC 0.73–0.76 in SHARE and 0.83–0.85 in HRS, while cross-sectional models fell to 0.55–0.57; across 18 SHARE countries, discrimination was 0.74–0.76 with history models and calibration worsened as non-persistence burden increased.

AI summary of the abstract · may contain errors · not medical advice · read the paper.