Estimating predictive values for blood pressure measurements from multivariate regression models with covariates

Area of Science:

• Biostatistics
• Epidemiology
• Health Sciences

Background:

• Predictive values estimate true measurements, excluding error and variability.
• Previous applications focused on blood pressure, estimating current or future hypertension risk.
• Existing methods lack robust incorporation of covariates for predictive modeling.

Purpose of the Study:

• To extend predictive value estimation to include covariates.
• To compare multivariate regression models with ordinary and logistic regression for predictive values.
• To assess the impact of covariates on the accuracy and reliability of predictive estimates.

Main Methods:

• Utilized multivariate regression models to predict future blood pressure levels based on current measurements and covariates.
• Compared predictive value estimates from multivariate models against ordinary linear and logistic regression.
• Employed childhood blood pressure data from East Boston, MA for analysis.

Main Results:

• Multivariate regression models yielded preferable predictive value estimates, showing reduced bias and variability compared to traditional methods.
• The inclusion of covariates significantly improved the accuracy of predictive estimates.
• Differences between multivariate and ordinary regression estimates were linked to the conditional reliability of future blood pressure levels.

Conclusions:

• Multivariate regression models offer a superior approach for estimating predictive values, particularly when covariates are considered.
• These models provide more reliable predictions of true underlying measurements, crucial for understanding health trajectories.
• The findings have implications for epidemiological studies and clinical risk assessment, especially in cardiovascular health.