Key result
Machine learning model identifies candidates for HBPM with ~0.84 AUC.
Why the study?
The frequency and factors associated with cardiologists' recommendations for home blood pressure monitoring in outpatients with hypertension were not well characterized, and a predictive model for such recommendations was lacking.
Does a machine learning predictive model based on electronic medical records accurately predict the recommendation of home blood pressure monitoring by cardiologists in outpatients with arterial hypertension?
Observational (n=44,358)
No
Does a machine learning predictive model based on electronic medical records accurately predict the recommendation of home blood pressure monitoring by cardiologists in outpatients with arterial hypertension?
Effect estimate: AUC-ROC 0.840
A machine learning model incorporating both structured data and free-text diagnosis descriptions can accurately identify outpatients with hypertension who receive recommendations for home blood pressure monitoring.
Background . Home blood pressure monitoring (HBPM) is recommended both for confirming the diagnosis of hypertension (HTN) and for monitoring treatment efficacy. The recommendation of the treating physician plays a key role in promoting active and correct HBPM. Objective . To assess the frequency of HBPM recommendations by cardiologists for outpatients with HTN, identify factors associated with HBPM recommendation, and develop a machine learning–based predictive model of HBPM recommendation using electronic medical record (EMR) data. Materials and methods . A retrospective analysis of EMR data from 44,358 outpatients with HTN seen by cardiologists in 2018–2024 was conducted. A predictive model was developed using the LightGBM algorithm based on structured clinical and demographic parameters and text features from the “Diagnosis Description” field (TF-IDF). Model performance was evaluated on a test set (70 % of 2018–2023 data) and an out-of-time validation set (OOT, 2024 data). Topic modeling (LDA) was applied to analyze recommendation structure. Results . HBPM recommendation was present in 40,180 patients (90.6 %). The machine learning model demonstrated high performance: test set AUC-ROC = 0.809, accuracy = 0.954, sensitivity = 0.810; OOT set AUC-ROC = 0.840, accuracy = 0.979, sensitivity = 0.837. Text features from diagnosis descriptions made the highest contribution (gain = 11,315.1), followed by age (3,461.8) and ICD code (3,138.4). Excluding text reduced AUC-ROC to 0.655–0.673. Higher probability of HBPM prescription was associated with younger age, advanced HTN stage, and uncontrolled blood pressure. LDA identified 6 topics; HBPM was mentioned in most. Conclusion . Cardiologists recommend HBPM to most patients with HTN. The decision to recommend HBPM is associated with objective indicators of disease severity and age. The developed predictive model accurately identifies patients for whom HBPM is indicated and can serve as a foundation for clinical decision support systems.
No takes yet. Share an insight, caveat, or question.
Ionov et al. (2026) conducted an observational in Arterial hypertension (n=44,358). Machine learning predictive model using electronic medical record data was evaluated on Model performance (AUC-ROC) for predicting home blood pressure monitoring recommendation on out-of-time validation set (AUC-ROC 0.840). A machine learning predictive model accurately identified patients for whom home blood pressure monitoring is indicated, achieving an AUC-ROC of 0.840 on the out-of-time validation set.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: