Abstract: FR-PO563
Derivation and Validation of an Uplift Model to Personalize Blood Pressure Treatment Strategy
Session Information
- Hypertension: Clinical and Translational
November 03, 2017 | Location: Hall H, Morial Convention Center
Abstract Time: 10:00 AM - 10:00 AM
Category: Hypertension
- 1105 Hypertension: Clinical and Translational - Genetics and Epigenetics
Authors
- Wilson, Francis Perry, Yale School of Medicine, New Haven, Connecticut, United States
- Biswas, Aditya, Yale University, New Haven, Connecticut, United States
- Parikh, Chirag R., Yale University and VAMC, New Haven, Connecticut, United States
Background
The Systolic Blood Pressure Intervention Trial (SPRINT) demonstrated that, for non-diabetic patients with cardiovascular risk factors, a more intensive systolic blood pressure lowering strategy (<120 mm) was superior to a standard blood pressure lowering strategy to prevent cardiovascular outcomes. However, absolute risk reduction was low suggesting that the vast majority of patients exposed to lower systolic BP will not directly benefit. Uplift modeling is a novel strategy to predict the marginal benefit of an intervention at the level of an individual allowing for personalized targeting of interventions.
Methods
We divided the SPRINT cohort into a 70% training and 30% validation set. Using deep-phenotyping via auto-encoders and a random forest uplift model, we predicted the marginal benefit of the intensive blood pressure compared to standard strategy. We then dichotomized this metric into two groups - "likely to benefit" and "unlikely to benefit". We compared baseline factors in these two groups to examine factors strongly predictive of benefitting from intensive blood pressure control.
Results
In the test set, we identified 1,696 participants who would be likely to benefit from intensive blood pressure control and 1,112 who would be unlikely to benefit. The most common phenotypes of those likely to benefit included older white males with risk factors associated with the metabolic syndrome (such as higher serum glucose and microalbuminuria) but preserved eGFR. The hazard ratio (HR) of the intervention on the primary outcome in the “likely to benefit” group was 0.56 (95% CI 0.38 – 0.83, p=0.004) compared to 1.18 (0.71 – 1.98, p=0.53) among those unlikely to benefit (p-for-interaction 0.04).[Figure]
Conclusion
Use of an uplift-targeting approach in clinical practice would increase the efficacy of intensive blood pressure treatment, improve absolute risk reduction while simultaneously increasing the total number of cardiovascular events avoided at a population level.
Funding
- NIDDK Support