A WARNING ABOUT USING PREDICTED VALUES TO ESTIMATE DESCRIPTIVE MEASURES

Article Mathematical & Computational Biology

Bias due to Berkson error: issues when using predicted values in place of observed covariates

Gregory Haber et al.

Summary: This study investigated the impact of unmeasured confounding on linear regression outcome models, finding that unmeasured confounding can lead to different inferences based on Berkson error measurements and observed covariates, which limits the practical application of substituting predicted values for observed covariates.

BIOSTATISTICS (2021)

Add to Collection

Article Public, Environmental & Occupational Health

A Warning About Using Predicted Values From Regression Models for Epidemiologic Inquiry

Elizabeth L. Ogburn et al.

Summary: Researchers may use regression-based estimates when direct access to data is not available, which can result in uninterpretable analyses. Simulations and observational data show that predicting outcomes of interest from an auxiliary model can lead to bias in any direction under both null and alternative hypotheses.

AMERICAN JOURNAL OF EPIDEMIOLOGY (2021)

Add to Collection

Article Public, Environmental & Occupational Health

On the Use of Regression Calibration in a Complex Sampling Design With Application to the Hispanic Community Health Study/Study of Latinos

Pedro L. Baldoni et al.

Summary: The study introduces regression calibration as a method for adjusting regression parameter estimates for covariate measurement error and proposes two variance estimators for a complex sampling design. Through extensive simulation using HCHS/SOL data, the relative performance of these variance estimation strategies is assessed, and the application of regression models with regression-calibrated exposures is demonstrated. Practical considerations for implementing these variance estimators in a large multicenter study are also discussed.

AMERICAN JOURNAL OF EPIDEMIOLOGY (2021)

Add to Collection

Article Public, Environmental & Occupational Health

On the Use of Regression Calibration in a Complex Sampling Design With Application to the Hispanic Community Health Study/Study of Latinos

Pedro L. Baldoni et al.

Summary: This study investigates variance estimation strategies for regression calibration methods in the context of complex sampling designs, proposing two methods for multistage probability-based sampling designs. The relative performance of these methods is evaluated through simulation studies and analysis of real data from the HCHS/SOL cohort.

AMERICAN JOURNAL OF EPIDEMIOLOGY (2021)

Add to Collection

Review Public, Environmental & Occupational Health