Pooling dietary data using questionnaires with open-ended and predefined responses: implications for comparing mean...

Area of Science:

• Nutritional Epidemiology
• Genetics
• Biostatistics

Background:

• Diet-gene analyses necessitate large sample sizes, often achieved by pooling data from consortia.
• Food frequency questionnaires (FFQs) are common dietary assessment tools, utilizing predefined and open-ended responses.
• Pooling FFQs with mixed response types requires categorizing open-ended responses into predefined categories.

Purpose of the Study:

• To evaluate the impact of categorizing open-ended food frequency questionnaire responses into noncontiguous predefined categories.
• To assess the effect on estimates of mean difference in intake and odds ratios in diet-gene association studies.
• To provide recommendations for data pooling strategies when FFQ response types differ.

Main Methods:

• Simulated dietary intake data from 1,664 controls in a lung cancer case-control study.
• Modeling of open-ended responses falling into gaps between noncontiguous predefined response categories.
• Comparison of bias and inflation of false positives for mean intake differences and odds ratios.

Main Results:

• Significant inflation of false positives was observed when estimating mean differences in dietary intake.
• Bias in estimating odds ratios was found to be acceptably small, even with mixed response types.
• The study highlights the differential impact of data pooling on various statistical measures.

Conclusions:

• When pooling FFQ data with mixed response types, focus on odds ratio estimation to minimize bias.
• Inferences regarding mean intake differences may be unreliable due to inflated false positives.
• Careful consideration of data harmonization is crucial for valid diet-gene association studies.