This sounds like a snarky remark, but it does have a point. Many large-scale nutrition studies rely on surveying thousands of people about their diet and lifestyles.

Unfortunately, this works best for detecting very dramatic health effects, like the link between smoking and lung cancer. For smaller health effects, any meaningful signal in the data can be overwhelmed by other correlations. One famous example is hormone replacement therapy for menopausal women: https://www.med.uottawa.ca/sim/data/EBM_Study_Bias_e.htm