Self-reported exposure data in epidemiologic studies: measurement errors and missing values
Year: 2008
Type: dissertation
Abstract: In observational epidemiology many studies rely on self-reported exposure information, which is prone to measurement errors and missing values. This may subsequently lead to bias in relative risk estimates of disease or other outcome measures. Therefore, it is important to assess data quality. The subject of this thesis in epidemiology was measurement error and missing data analysis with emphasis on self-reported data on diet and melanoma risk factors. The data were collected by postal health- and lifestyle questionnaires used in the Norwegian Women and Cancer Study (NOWAC), an ongoing population-based cohort study. The aspects of measurement error studied were reproducibility and recall bias, based on two separate data collections: a test-retest study (3 month time interval) in a sample of 1,995 women, and a nested case-control study of 208 cases of malignant melanoma and 2,080 matched controls. Reproducibility was studied for food frequency questionnaire (FFQ) data, and melanoma risk factors. The effect of different imputation methods for missing values in FFQ data was also investigated. A common procedure for microarray data, k nearest neighbors imputation (KNN), was adapted and applied to FFQ data. Recall bias was studied for melanoma risk factors. The observed level of reproducibility for the FFQ data was within the reported range for similar questionnaires, but may attenuate disease risk estimates, as illustrated in an example using regression calibration. Comparing different imputation methods, the largest differences were seen for KNN compared to the null value (no consumption), and for food items with a high proportion of missing (>30%). The overall reproducibility of melanoma risk factors was found to be acceptable. Indications of recall bias in the reporting of melanoma risk factors were found, but the results were not consistent for the different exposures.
Author: Christine L Parr
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W576068589