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

This prospective cohort study demonstrates that incorporating visit-to-visit blood pressure variability into cardiovascular disease risk prediction models significantly enhances their discriminatory accuracy for patients aged 55 and older compared to models relying solely on a single blood pressure measurement.

The Gist

The "Weather Forecast" of Your Heart: Why One Snapshot Isn't Enough

Imagine you are trying to predict the weather for next week. If you step outside for just one second and see the sun, you might confidently say, "It's going to be a perfect week!" But what if, five minutes later, a storm hits? Or what if the temperature swings wildly from freezing to boiling every hour? A single snapshot tells you very little about the pattern of the weather.

This is exactly what a new study by researchers in Australia discovered about blood pressure and heart disease.

The Old Way: The Single Snapshot

For decades, doctors have predicted a patient's risk of heart disease (like heart attacks or strokes) by looking at their blood pressure at just one specific moment during a clinic visit.

Think of this like judging a whole movie by looking at a single frozen frame. If that frame shows a calm scene, you assume the whole movie is peaceful. But if the character is actually screaming in the next frame, your prediction is wrong.

The study found that relying on this single "frozen frame" of blood pressure isn't very good at predicting who will get heart disease. In fact, the computer models using this method were only about 72% accurate (like guessing the outcome of a coin flip, but slightly better).

The New Way: The "Rollercoaster" Tracker

The researchers asked a simple question: What if we don't just look at the blood pressure once, but watch how it bounces around over time?

They looked at patients who had their blood pressure checked at least five times over several years. They didn't just care about the average number; they cared about the variability.

• The Smooth Ride: A patient whose blood pressure stays steady at 120/80 every visit.
• The Rollercoaster: A patient whose blood pressure jumps from 110 to 150 and back down to 115 between visits.

The study found that the "Rollercoaster" patients were at a much higher risk of heart trouble, even if their average blood pressure looked normal.

The Results: A Massive Upgrade

When the researchers swapped the "single snapshot" for the "rollercoaster tracker" in their prediction models, the results were dramatic:

• Accuracy Jump: The prediction accuracy soared from 72% to 83%.
• The Metaphor: It's like upgrading from a blurry, black-and-white security camera to a high-definition, color camera with night vision. Suddenly, you can see the dangers that were previously hidden in the shadows.

The study showed that the more your blood pressure bounces around between doctor visits, the higher your risk of a heart event, regardless of whether your average pressure is high or low.

Why Does This Happen?

Think of your arteries (blood vessels) like rubber bands.

• If your blood pressure is steady, the rubber band stretches gently and consistently.
• If your blood pressure swings wildly (high variability), the rubber band is constantly being snapped tight and then released. This "whiplash" damages the inside of the vessel, causing inflammation and wear and tear that leads to blockages and heart attacks.

What Does This Mean for You?

1. One Visit Isn't Enough: If you only get your blood pressure checked once a year, you might be missing the warning signs. The study suggests that regular, repeated checks are crucial.
2. It's Not Just About the Number: Even if your average blood pressure is "okay," if it fluctuates wildly, you might still be at risk.
3. Better Tools for Doctors: The researchers built a new "calculator" for doctors. Instead of just plugging in one number, they can now plug in the history of your blood pressure. This helps doctors spot high-risk patients earlier and treat them sooner.

The Bottom Line

This study is a wake-up call. We can't just look at a single moment to understand our heart health. We need to look at the story of our blood pressure over time. By tracking the "rollercoaster" instead of just the "snapshot," we can predict heart disease much better and keep more people safe.

In short: Don't just check your blood pressure once and forget it. The journey of your numbers matters just as much as the destination.

Technical Summary

1. Problem Statement

Current cardiovascular disease (CVD) risk prediction models (e.g., Framingham Risk Score, QRISK3, Australian CVD risk model) predominantly rely on a single blood pressure (BP) measurement taken at a specific point in time. However, BP is a dynamic physiological parameter that fluctuates significantly between clinical visits.

• The Gap: While Visit-to-Visit Blood Pressure Variability (VVV BPV) is recognized as a risk factor, it is underutilized in predictive modeling. Existing attempts to incorporate BPV (e.g., adding Standard Deviation to the most recent SBP in QRISK3) have yielded conflicting results, with some studies showing no improvement in model discrimination.
• The Objective: This study aims to develop and validate a CVD risk prediction model that replaces the traditional single BP measurement with VVV BPV metrics (Standard Deviation and Coefficient of Variation) derived from longitudinal Electronic Health Record (EHR) data, specifically for individuals aged 18–55.

2. Methodology

Study Design & Data Source

• Type: Prospective cohort study using a landmark age approach.
• Data Source: The electronic Practice-Based Research Network (ePBRN) in Southwestern Sydney, Australia (2006–2019). This dataset links General Practice (GP) and secondary/tertiary hospital records.
• Cohort Selection:
  • Inclusion: Patients aged 18–55 with at least five BP readings across separate clinical encounters (minimum 2-week interval) prior to age 55. No prior atherosclerotic CVD diagnosis at the landmark age (55).
  • Exclusion: Missing gender/year of birth, no follow-up after age 55, or outcome data recorded before age 55.
  • Final Sample: 3,065 patients (54.6% male, 45.4% female).
• Landmark Approach: Risk prediction time-origin was set at age 55. All BP variability metrics were calculated using data collected prior to this age.

Variables & Predictors

• Outcome: First occurrence of CVD (fatal/non-fatal coronary heart disease, cerebrovascular events, heart failure, peripheral artery disease) after age 55.
• Baseline Predictors (at age 55): Gender, smoking, fasting glucose, BMI, TC/HDL ratio, eGFR, family history, hypertension, diabetes.
• BP Metrics:
  • Model A (Control): Single SBP/DBP measurement.
  • Model B (Experimental): VVV BPV calculated as Standard Deviation (SD) and Coefficient of Variation (CV).
  • Categorization: SD and CV were categorized using empirically derived cut-offs from prior work (e.g., SBP SD >19 mmHg, SBP CV >14%).
• Data Handling: Missing predictor data were addressed using Multiple Imputation by Chained Equations (MICE) generating 20 datasets.

Statistical Analysis

• Modeling: Cox proportional hazards regression with 10-fold cross-validation.
• Performance Metrics:
  • Discrimination: Harrell's C-index and time-varying Area Under the Curve (AUC) for 1–5 year predictions.
  • Calibration: Calibration slopes and Brier scores (assessed annually).
• Software: R version 4.4.0 (packages: mice, survival).

3. Key Contributions

1. Methodological Innovation: Successfully integrated VVV BPV (SD and CV) into a CVD risk model using real-world longitudinal EHR data, replacing the static single-measurement approach.
2. Empirical Validation: Demonstrated that VVV BPV significantly outperforms single BP measurements in predicting CVD risk for the 18–55 age group, a demographic often underrepresented in CVD studies.
3. Metric Comparison: Provided a direct comparison between SD and CV, finding that while both improved models, CV (normalized to the mean) showed slightly superior performance, particularly for Diastolic BP (DBP).
4. Clinical Insight: Showed that combining single BP with VVV BPV did not further improve the model over using VVV BPV alone, suggesting VVV BPV captures the necessary risk information more comprehensively than a static snapshot.

4. Key Results

Model Performance (Discrimination)

• Harrell's C-Index (5-year prediction):
  • Single SBP Model: 0.716 (95% CI: 0.658–0.775)
  • SBP + SD Model: 0.833 (95% CI: 0.804–0.862)
  • SBP + CV Model: 0.837 (95% CI: 0.810–0.864)
  • Observation: Incorporating VVV BPV increased discrimination by ~0.12 points.
• Time-Varying AUC (5-year):
  • Single SBP: 0.757
  • SBP + SD: 0.852
  • SBP + CV: 0.856
• Diastolic BP (DBP): Models using DBP variability showed slightly higher AUCs than SBP-only models, consistent with literature suggesting DBP variability is a strong predictor of arterial stiffness and endothelial dysfunction.
• Combined Models: Adding single BP measurements to VVV BPV models did not significantly alter the AUC or C-index compared to VVV BPV models alone.

Calibration and Other Metrics

• Calibration: No significant difference in calibration slopes or Brier scores between single BP and VVV BPV models. All models were well-calibrated (slopes close to 1), indicating neither over- nor under-prediction of risk.
• Subgroup Analysis: Performance improvements were consistent across males and females. However, the single BP model tended to underpredict risk in males, a bias corrected by the VVV BPV models.

5. Significance and Implications

Clinical Impact

• Enhanced Risk Stratification: Replacing a single BP reading with VVV BPV significantly enhances the accuracy of CVD risk estimation for adults under 55. This allows for earlier identification of high-risk individuals who might be missed by static models.
• Dynamic Monitoring: The study supports the shift from "snapshot" BP monitoring to longitudinal tracking. It suggests that BP variability itself is a robust biomarker for vascular aging and hemodynamic stress.
• Medication Adherence Proxy: High VVV BPV is linked to poor medication adherence. Incorporating this metric into risk scores could indirectly flag adherence issues, allowing clinicians to intervene before a CVD event occurs.

Health Policy & EHR Integration

• EHR Automation: The study advocates for integrating automated VVV BPV calculations into GP Electronic Health Records (EHR). This would enable real-time, dynamic CVD risk scoring during routine visits without requiring manual calculation.
• Prevention Strategy: By identifying risk earlier (in the 18–55 age group), healthcare systems can shift focus toward primary prevention, potentially reducing the future burden of CVD.

Limitations

• Generalizability: The cohort was restricted to patients with sufficient BP records (indicating higher chronic disease burden), which may limit applicability to the general healthy population.
• Age Restriction: The model is validated for risk prediction starting at age 55; further research is needed for younger landmark ages.
• Data Quality: Reliance on EHR data necessitated multiple imputation for missing variables, though this is standard for real-world data studies.

Conclusion
The study concludes that Visit-to-Visit Blood Pressure Variability is a superior predictor of CVD risk compared to single BP measurements. Integrating VVV BPV (specifically using SD and CV) into risk prediction models significantly improves discrimination, offering a more personalized and accurate tool for CVD prevention in primary care settings.