Estimating uncertainty when providing individual cardiovascular risk predictions: a Bayesian survival analysis

Area of Science:

• Cardiology
• Biostatistics
• Medical Informatics

Background:

• Cardiovascular disease (CVD) risk scores traditionally provide point estimates without quantifying individual risk uncertainty.
• Accurate risk assessment is crucial for effective cardiovascular disease (CVD) prevention and management.

Purpose of the Study:

• To demonstrate the feasibility of using Bayesian methods to calculate uncertainty in individual CVD risk predictions.
• To explore the clinical utility of these uncertainty measures in patient care.

Main Methods:

• A Bayesian Weibull model was developed using data from 8,355 individuals with established atherosclerotic CVD from the Utrecht Cardiovascular Cohort-SMART study.
• The model predicted 10-year recurrent CVD risk, incorporating 95% credible intervals (CIs) for individual risk estimates.

Main Results:

• Bayesian model predictions were similar to traditional models, but crucially provided 95% credible intervals (CIs) for individual risk.
• A significant portion of the population (17%) exhibited a 95% CI width of 10% or greater, highlighting substantial individual risk uncertainty.
• Uncertainty measures decreased with larger sample sizes used in model derivation.

Conclusions:

• Calculating uncertainty in individual CVD risk predictions using Bayesian methods is feasible and clinically relevant.
• These uncertainty measures can aid in comparing treatment options and determining the probability of risk falling below treatment thresholds.
• Physicians require training to interpret these uncertainty measures accurately, as they reflect sampling error rather than prediction biases.