Regression analysis with categorized regression calibrated exposure: some interesting findings

Area of Science:

• Epidemiologic research
• Biostatistics
• Public health

Background:

• Measurement error is a common issue in epidemiologic studies.
• Standard regression calibration methods are not directly applicable to categorical exposure data.
• Categorical exposure analysis is frequently used in epidemiology.

Purpose of the Study:

• To evaluate the performance of regression calibration for categorical exposures.
• To compare the proposed approach with the naive method (no error correction).
• To assess the impact of incorporating original scale information into categorical variables.

Main Methods:

• Semi-analytical calculations and simulations were employed.
• The study compared approaches using quintile scales and original scale incorporation.
• Real-world data from the Norwegian Women and Cancer study (NOWAC) were analyzed.

Main Results:

• Regression calibration without considering the original scale can lead to biased variance and percentile estimates.
• The analyzed approach often retains significant misclassification from the observed exposure.
• Corrected estimates using categorical scales can remain biased, sometimes equaling naive estimates.
• Regression calibration is superior to the naive method when using category medians.

Conclusions:

• Standard regression calibration is inappropriate for measurement error correction in percentile-scaled exposures.
• Re-establishing the original exposure scale resolves the bias issues.
• These conclusions apply to all regression models used in epidemiologic research.