Regression Calibration in Epidemiology
Author Information
Author(s): Ingvild Dalen, John P. Buonaccorsi, Petter Laake, Anette Hjartåker, Magne Thoresen
Primary Institution: University of Oslo
Hypothesis
Is regression calibration appropriate for correcting measurement error in categorical exposure analysis?
Conclusion
Regression calibration is not suitable for measurement error correction when exposure is analyzed on a percentile scale, but relating back to the original scale resolves the issue.
Supporting Evidence
- Regression calibration does not maintain important qualities of the true exposure distribution when using categorized exposure.
- Using the median values of quintile groups improves the performance of regression calibration.
- The naive approach and regression calibration yield similar results when exposure is analyzed categorically.
Takeaway
This study looks at how to fix mistakes in measuring things in health studies, especially when we group people into categories. It finds that using the original measurements is better than just using the groups.
Methodology
The study used semi-analytical calculations and simulations to evaluate regression calibration methods against naive approaches in epidemiological data.
Potential Biases
The categorization of corrected exposure retains misclassification similar to naive approaches, which can bias effect estimates.
Limitations
The study's findings may not apply to all regression models and are limited to the specific context of measurement error in categorical exposure.
Participant Demographics
The study involved 898 individuals from the Norwegian Women and Cancer study, with a prevalence of depression at 19.7%.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website