Signal vs. Noise: How to Read a Nutrition Study
Most nutrition headlines are reverse-engineered from press releases. A reporter gets a summary from a university PR department, a PR department summarises a paper, a paper summarises a trial, and a trial measured something only vaguely related to what the headline claims. By the time it reaches your feed, four layers of interpretation have transformed “modest correlation in a specific sub-group” into “scientists saying X causes Y.”
Here is the framework I use to cut through it.
Step 1 — Who paid for it?
Follow the money first, not the conclusions. Conflict-of-interest disclosures are buried in supplementary sections for a reason.
Myth
Industry funding doesn't bias nutrition research — journals have ethics rules.
A 2019 systematic review in PLOS Medicine found that industry-funded studies were five times more likely to reach conclusions favourable to the sponsor than studies with no financial conflict. Journals require disclosure; they do not adjudicate the analysis.
Step 2 — Randomised controlled trial vs. epidemiology
These are not equivalent evidence, even when both appear in peer-reviewed journals.
Myth
Epidemiological studies prove that eating red meat causes cancer.
Large cohort studies (like the ones behind IARC’s processed-meat classification) show association, not causation. They cannot control for the full dietary pattern — people who eat more red meat in Western samples also eat more refined carbohydrates, smoke more, and exercise less. Confounding is almost never fully adjusted away.
Signal
Effect sizes in nutrition epidemiology are almost always smaller than what makes headlines.
A relative risk of 1.15 means a 15% increase in relative risk — which sounds alarming until you learn the absolute baseline risk is 0.4%. The absolute risk increase is 0.06 percentage points. This arithmetic is almost never in the headline.
Step 3 — What was actually measured?
Surrogate endpoints are not the same as clinical outcomes.
Myth
This study proves saturated fat raises heart disease risk because it raised LDL.
LDL-C is a surrogate marker — a proxy — not a direct measure of cardiovascular events. Studies that use LDL as the outcome measure are measuring a proxy. Randomised controlled trials that reduced LDL via diet alone have not reliably reduced mortality. The LDL → heart disease arrow is more complicated than the model suggests.
Signal
Fasting insulin and HOMA-IR are stronger early-warning metabolic markers than total cholesterol.
These markers reflect insulin sensitivity directly, and insulin resistance precedes most of the metabolic conditions (T2D, NAFLD, PCOS, hypertension) that kill people at scale. Most standard lipid panels do not include them. Ask your doctor for them by name.
Step 4 — Duration and dose
A six-week intervention tells you about six-week adaptations.
Signal
Short-term dietary interventions often show adaptation effects, not steady-state effects.
Switching to a low-carbohydrate diet for six weeks will cause measurable changes in LDL particle size, triglycerides, and HDL — but some of these continue to change over months as metabolic adaptation occurs. Studies that end at six weeks are capturing a snapshot of transition, not a final state.
The short checklist
When you see a nutrition headline, ask:
- Who funded it? — look for the COI disclosure
- What type of study? — RCT > prospective cohort > cross-sectional > ecological
- What was measured? — surrogate markers vs. clinical endpoints
- What was the effect size? — absolute risk, not just relative risk
- How long did it run? — weeks vs. years matter
Most stories collapse somewhere in that list. The ones that survive it are worth reading closely.