PROGNOSTIC SIGNIFICANCE OF BLOOD PRESSURE PARAMETERS FOR PREDICTING CARDIOVASCULAR DEATH

Abstract

Objective: The nycterohemeral rhythm of blood pressure (BP) and heart rate has been associated with cardiovascular outcomes. However, long-term prognostic significance of BP at different levels of heart rates has rarely been studied. We therefore conducted the present analysis to investigate the prognostic values for cardiovascular outcome by the ambulatory BP parameters during different time windows Design and method: We enrolled 5 large cohorts from whole world, including CARDIA, IDACO, JHS, SPRINT to find out independent prognostic factors for cardiovascular disease mortality, feature engineering and feature transformation were used. Several statistical and machine learning methods including the stepwise procedure, lasso penalty, and random survival forest with bootstrap technique were used to screen important candidates of risk factors. Models building are based on systolic BP replacement into the established Framingham risk score. Directly one to one replacement and indirectly two stage replacement methods are considered with variable diversity. C-statistics, NRI are used for model performance evaluation. Results: In model building procedure, nighttime BP combined with pulse pressure, morning-evening difference of SBP and weighted SBP consistently had a higher C-statistics than had office systolic BP. In validation datasets, equations incorporating mean BP and pulse pressure had best performance in C-index (0.754). In contrast, the C-index (higher is better), of the usual Framingham risk score with office systolic BP are 0.744. Furthermore, the best model had acceptable predicted accuracy based on model calibration statistics (p > 0.05). Conclusions: Our study demonstrated that mean BP at night combined with pulse pressure, morning-evening difference of SBP and weighted SBP, throughout 24 hours had the largest predictive power of 20-year cardiovascular death in population with ABPM recordings. The transportability of the proposed cardiovascular risk function should be evaluated in future studies.