Abstract 11487: Individualizing the Cardiovascular Benefits of Intensive Blood Pressure Reduction Using Computational Phenomaps of the SPRINT and ACCORD Trials

Abstract

Introduction: Intensive systolic blood pressure (SBP) control reduces major adverse cardiovascular events (MACE) in patients without type 2 diabetes mellitus (T2DM). However, these benefits are less clear in T2DM. We evaluated an application of machine learning to clinical trials of patients without and with T2DM in assessing a personalized cardiovascular benefit of intensive SBP control. Methods: In SPRINT, a trial of intensive (n=4678, SBP<120 mmHg) versus standard (n=4683, SBP<140 mmHg) SBP control in 9361 patients without T2DM, we created a topological representation of the trial patients using 59 baseline variables (trial phenomaps). Within each patient’s 5% topological neighborhood, we calculated hazard ratios (HR) for recurrent MACE (cardiovascular death, acute coronary syndrome, stroke, acute decompensated heart failure). We trained an extreme gradient boosting algorithm to predict the personalized effects of intensive SBP control using features linked to topological benefit. We then tested this machine learning tool in the ACCORD BP trial of patients with T2DM (n=2362 & 2371 in the intensive and standard arms respectively). Results: In SPRINT (age 68±9 years, 36% women) there were a total of 1046 recurrent MACE endpoints. The median individual patient HR was 0.58 [IQR, 0.38-0.82] ( A ). We developed a 10-variable tool in SPRINT ( B ) and subsequently tested in ACCORD BP (age 63±7 years, 48% women) where it identified individual patients with a higher benefit with intensive vs standard SBP control (adj. HR for time-to-MACE 0.77 [95% CI 0.61-0.98] in individuals with above median predicted benefit [high responders] ( C ) vs 0.97 [95% CI 0.77-1.23] ( D ) for below median predicted benefit [low responders]). Conclusions: We present a clinical trial-based, machine learning tool that identifies an individual’s personalized benefit from intensive versus standard SBP goals in patients with and without T2DM and may be used to guide clinical decision-making.