Related Experiment Videos
Linear regression calibration: theoretical framework and empirical results in EPIC, Germany
Gisela Kynast-Wolf1, Nikolaus Becker, Anja Kroke
1Division of Clinical Epidemiology, German Cancer Research Center, Im Neuenheimer Feld 280, D-69120 Heidelberg, Germany.
Annals of Nutrition & Metabolism
|March 27, 2002
Summary
Dietary assessment in multicenter studies faces challenges due to varying measurement errors. Linear regression calibration using a 24-hour recall reference improves comparability of food frequency questionnaire data across centers.
Area of Science:
- Nutritional epidemiology
- Biostatistics
- Public health research
Background:
- Large-scale dietary assessments, often using food frequency questionnaires (FFQs), require population-specific tailoring.
- Multicenter studies face challenges in comparing dietary data due to local variations in assessment tools and measurement errors.
- A common reference instrument is crucial for accurate risk analysis in multicenter dietary studies.
Purpose of the Study:
- To describe the statistical basis of linear regression calibration for multicenter dietary data.
- To present empirical results of applying this calibration to fruit, cereal, and meat consumption data from EPIC Germany.
- To evaluate the impact of calibration on the comparability of dietary intake between different study centers.
Main Methods:
- Utilized a 24-hour dietary recall (EPIC-SOFT) as a standardized reference instrument across all centers.
- Applied linear regression calibration, regressing FFQ data against the 24-hour recall reference.
- Analyzed calibration coefficients (lambda) to assess absolute and proportional scaling bias for specific food groups (fruit, cereals, meat) in two German EPIC centers (Heidelberg and Potsdam).
Main Results:
- Fruit consumption was well-measured by FFQs in both centers (lambda values near 1).
- Cereal and meat consumption showed moderate to poor measurement by FFQs, with significant center-specific biases observed for meat.
- Calibration adjusted FFQ data, altering center rankings and reducing variation for less accurately measured food items.
Conclusions:
- Linear regression calibration is a necessary statistical approach to harmonize dietary data in multicenter studies.
- The study highlights the need for robust calibration methods to address measurement errors and center-specific biases.
- Further development of the statistical framework is required for broader application of dietary data calibration.