Correlations between estimated and true dietary intakes: using two instrumental variables

Area of Science:

• Nutritional epidemiology
• Biomarker research
• Statistical modeling

Background:

• Dietary assessment accuracy is crucial for epidemiological studies.
• Traditional methods like food frequency questionnaires (FFQs) have limitations in estimating true dietary intake.
• Instrumental variables offer a potential solution for measurement error correction.

Purpose of the Study:

• To present a novel application of the method of triads for estimating the correlation between a dietary questionnaire (Q) and true dietary intake (T).
• To assess the validity of FFQs compared to repeated 24-hour dietary recalls (R) using biological markers (M, P).

Main Methods:

• Employed the method of triads using three surrogate variables (Q, M, P), where M and P are instrumental (biological) variables.
• Estimating equations were used to compute correlations: Corr(Q,T), Corr(M,T), Corr(P,T), conditional on error correlation assumptions.
• A validation study utilized food frequency data, repeated 24-hour recalls, and biological samples from Seventh-day Adventists.

Main Results:

• Estimated correlation between FFQ and true intake (Corr(Q,T)) ranged from 0.40 to 0.66.
• Estimated correlation between repeated recalls and true intake (Corr(R,T)) was higher, ranging from 0.48 to 0.83.
• Positive correlations were found between errors in FFQ and repeated recall measures.

Conclusions:

• Instrumental variables (M, P) are better suited than reference methods (M, R) for satisfying error correlation assumptions.
• Repeated 24-hour recalls demonstrate greater validity for estimating true dietary intake compared to FFQs.
• Achieving precise estimates of dietary intake correlation (Corr(Q,T)) necessitates large-scale validation studies (2000-3000 subjects).