Development of a machine learning-based model for predicting individual responses to antihypertensive treatments

Background and Aims Personalized antihypertensive drug selection is essential for optimizing hypertension management. The study aimed to develop a machine learning (ML) model to predict individual blood pressure (BP) responses to different antihypertensive medications. Methods and Results We used data from a pragmatic, cluster-randomized trial on hypertension management in China. Each patient's multiple visit records were included, and two consecutive visits were paired as the index and subsequent visits. The least absolute shrinkage and selection operator method was used to select index visit variables for predicting subsequent BP. The dataset was randomly divided into training and test sets in a 7:3 ratio. Model performance was evaluated using mean absolute error (MAE) and R-square in the test set. A total of 19013 hypertension management visit records (6282 patients) were included. The mean age of the study population was 63.9 years, and 2657 (42.3%) were females. A total of 12 phenotypical features (age, sex, smoking within seven days, body mass index, waist circumference, index visit systolic BP, diastolic BP, heart rate, comorbidities of diabetes, dyslipidemia, coronary heart disease, and stroke), together with currently taking any prescribed antihypertensive medication regimens and visits time interval were selected to build the model. The Extreme Gradient Boost model performed best among all candidate algorithms, with an MAE of 8.57 mmHg and an R2 = 0.28 in the test set. Conclusion The ML techniques exhibit significant potential for predicting individual responses to antihypertensive treatments, thereby aiding clinicians in achieving optimal BP control safely and efficiently. Trial Registration ClinicalTrials.gov, NCT03636334. Registered 3 July 2018, https://clinicaltrials.gov/study/NCT03636334.