Regression calibration method for correcting measurement-error bias in nutritional epidemiology

Area of Science:

• Epidemiological statistics
• Biostatistics
• Nutritional epidemiology

Background:

• Measurement error in epidemiological studies can bias effect estimates from regression models.
• Previous regression calibration methods addressed logistic regression for odds ratios.
• Accurate assessment of nutrient-disease relationships is crucial for public health.

Purpose of the Study:

• To extend regression calibration for Cox proportional hazards and linear regression models.
• To adjust incidence rate ratios and regression slopes for measurement error.
• To improve the accuracy of estimating associations between dietary factors and health outcomes.

Main Methods:

• Applied regression calibration to Cox proportional hazards models for incidence rate ratios.
• Applied regression calibration to linear regression models for regression slopes.
• Utilized validation studies with gold standards or reliability studies with replicate measurements.

Main Results:

• Demonstrated correction of rate ratios for breast cancer incidence and dietary intakes (vitamin A, alcohol, total energy) in the Nurses' Health Study.
• Illustrated estimation of regression slopes for bone density and dietary intakes (caffeine, calcium, total energy) in the Massachusetts Women's Health Study.
• Developed SAS macros for implementing the extended regression calibration methods.

Conclusions:

• Regression calibration effectively adjusts for measurement error in Cox and linear regression models.
• The extended methods provide more accurate estimates of nutrient-disease and exposure-outcome relationships.
• This statistical approach enhances the reliability of findings in nutritional epidemiology and related fields.