[admash]

Inaccuracies and bias in food frequency questionnaire reporting (collectively termed “measurement error”) are well known and can effectively be quantified via validation approaches (see https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5404419/ ) and remediated by well-established statistical techniques such as regression calibration (see https://dietassessmentprimer.cancer.gov/learn/regression.htm... ).

Furthermore, when regression calibration is performed, the magnitude of the calibrated effect is often shown to be greater than the uncalibrated result, so that 1.11 hazard ratio is likely to be larger in reality.

Finally, “small” is clearly relative. I’m certainly quite interested in knowing about things which increase or decrease my risk of having a heart attack or stroke by 10%.

Certainly a blinded randomized controlled feeding trial is preferable, but is not always feasible, particularly when it would entail controlling participants’ diets for long periods of time.