October 1, 1990American Journal of Epidemiology
Correction of Logistic Regression Relative Risk Estimates and Confidence Intervals for Measurement Error: The Case of Multiple Covariates Measured With Error
Authors
Discussion
Loading...
Member takes
Implication
Prospective cohort analysis reveals measurement error adjustments alter dietary risk estimates for breast cancer, suggesting saturated fat associations remain null despite misclassification.
Key Points
- To establish a statistical method that corrects both point and interval relative risk estimates in multiple logistic regression models when multiple continuous covariates contain measurement error.
- Developed a correction procedure using a separate validation study to model relationships between surrogate variables and true risk factors via multivariate linear regression.
- Incorporated parameter estimation variability from the validation study directly into the corrected relative risk confidence intervals.
- Applied the method to prospective cohort data assessing dietary fat, calories, and alcohol intake in relation to breast cancer, using a questionnaire validation study.
- For a 25 g increment in alcohol intake, the age-adjusted relative risk rose from 1.33 (95% CI, 1.14–1.55) before correction to 1.62 (95% CI, 1.23–2.12) after measurement error correction.
- For a 10 g increment in saturated fat intake, the age-adjusted relative risk was 0.94 (95% CI, 0.83–1.06) prior to correction and 0.84 (95% CI, 0.59–1.20) following correction.
- Measurement error across nutritional covariates did not explain the lack of a substantial positive association between saturated fat intake and breast cancer risk.
Cite This Study
Rosner et al. (1990) studied this question.