Classification of Visit-to-Visit Blood Pressure Variability: A Machine Learning Approach for Data Clustering on Systolic Blood Pressure Intervention Trial (SPRINT)

Background: Blood pressure variability (BPV) is associated with the cardiovascular disease. However, there is no standard risk stratification method to evaluate BPV. Our study aims to cluster BPV into three levels, namely, low, medium and high levels, by a machine learning approach. Methods: The Systolic Blood Pressure Intervention Trial (SPRINT) dataset, which includes patients with hypertension or at risk of cardiovascular diseases, was obtained from a clinical data sharing platform. In the clinical trial, participants with systolic blood pressure (SBP) of at least 130 mmHg and an increased cardiovascular risk were randomized to receive intensive treatment (targeting SBP below 120 mmHg) or standard treatment (targeting SBP below 140 mmHg), and blood pressure (BP) were measured and recorded during the follow-up periods. Visit-to-visit BPV was measured by the deviation between the observed records and the personalized BP trends, and two-dimensional clustering on SBP and diastolic BP were applied. Different curve fitting techniques (linear regression and cubic regression) and clustering methods (K-means and Agglomerative Clustering) were attempted and compared with each other. Results: With 8,092 participants and a median follow-up of 3.26 years, linear regression was a simple and reliable method to capture the BP trend. K-means model showed stable data clustering results. Intensive treatment showed to be effective for participants with a high level of BPV. Conclusion: Machine learning can be used for data clustering on BPV.