The dropout problem
In 2013, Stanford published a meta-analysis of 135 dietary trials. The headline: no difference between low-fat and low-carb at twelve months. The buried finding: the average dropout rate across all 135 trials was 40 percent. The analyses were conducted on the survivors.
We have read every major nutritional RCT since 1980. The sugar industry ones that blamed fat. The dairy industry ones that found calcium prevents obesity. The PREDIMED trial — a landmark Spanish study meant to prove the Mediterranean diet prevents heart attacks — was retracted and republished after randomization failures. Our problem is not data. Our problem is that nutritional data is almost comically difficult to generate. You cannot blind someone to what they are eating. Metabolic ward studies involve eight to twenty people. So we rely on food frequency questionnaires where participants underreport calories by 30 to 40 percent. We are building a science on self-reported data from people who cannot remember lunch.
The ancestral camp concludes formal research is useless. We reject the inference. Ancestral populations did not live long enough to develop the chronic diseases we study — survivorship bias, the most basic error in epidemiology. The intuitive camp makes a claim we take seriously: diet-related stress is harmful. But stop thinking about food
is not a prescription for a diabetic whose body sends corrupted hunger signals. The industrial critics are right about the subsidies. Knowing Coca-Cola funded biased research does not tell you what to eat.
Where we concede ground: Ancel Keys — the mid-century physiologist whose research shaped US dietary fat guidelines for decades — cherry-picked seven countries from twenty-two in his Seven Countries Study. We knew. The guidelines were issued anyway.
What would change our mind: A preregistered, industry-free trial showing an ancestral diet outperforming current guidelines on cardiometabolic biomarkers.
Read the full synthesis: Why can’t anyone agree on a healthy diet?