Incorporation of Visit-to-Visit Blood Pressure Variability into Cardiovascular Disease Risk Prediction

Abstract BACKGROUND Visit-to-visit blood pressure variability (VVV BPV) is an important yet underutilised risk factor for cardiovascular disease (CVD) risk prediction. Incorporating VVV BPV in the model predicting CVD could improve its performance. This study aims to incorporate VVV BPV into a CVD risk prediction model and to evaluate its performance by comparing the discrimination and calibration of models using a single BP measurement versus those incorporating VVV BPV METHODS This prospective cohort study included data from the electronic practice-based research network (ePBRN) in Southwestern Sydney, focusing on patients aged 18-55 years with at least five BP readings, excluding those with incomplete data or no follow-up after 55. VVV BPV measured by standard deviation (SD) and coefficient of variation (CV). The main outcome is the first occurrence of CVD. We developed the models using Cox proportional hazards regression with 10-fold cross-validation on all imputed datasets. Model performance was evaluated for discrimination and calibration. Discrimination was assessed using Harrell’s C-index and time-varying AUC for five-year CVD prediction. Calibration was assessed using calibration slopes and Brier scores, which were also evaluated annually. RESULTS The study involved 3,065 patients, with 45.41% women. Incorporating VVV BPV improved the prediction of CVD risk in people aged 55 years. The model with a single systolic blood pressure (SBP) measurement had a Harrel C-Index of 0.716 (95% CI: 0.658 - 0.775), while those using SD and CV scored higher at 0.833 (95% CI: 0.804 - 0.862) and 0.837 (95% CI: 0.810 - 0.864), respectively. Five years’ AUC for SBP was 0.852 (95% CI: 0.820 – 0.885) for SD and 0.856 (95% CI: 0.824 – 0.888) for CV. In contrast, the single SBP model had a lower AUC of 0.757 (95% CI: 0.700 – 0.815). No significant difference was observed in calibration slopes and Brier scores between the model using single BP and VVV BPV. CONCLUSIONS This study developed a model for CVD risk estimation using VVV BPV instead of a single blood pressure measurement. Replacing a single BP measure with VVV BPV significantly enhanced the model’s predictive accuracy.