How to Read a Health Study Without Getting Fooled

Step 1: Find the actual question

Translate the paper into a sentence: In which population, does which exposure or intervention, compared with what, affect which outcome, over what time? This PICO-style frame exposes missing pieces quickly.

A headline may say a food ‘improves health’ while the study measured one laboratory marker for four weeks. That outcome may be interesting, but it is not the same claim.

Step 2: Match the design to the claim

• Randomized controlled trial: can support causal inference when allocation, adherence, follow-up and measurement are sound.
• Cohort study: follows exposures and outcomes over time; useful for association and risk patterns, still vulnerable to confounding.
• Case-control study: starts with an outcome and looks backward; efficient for rare outcomes, sensitive to selection and recall bias.
• Cross-sectional study: one-time snapshot; cannot reliably establish which came first.
• Systematic review or meta-analysis: structured synthesis whose strength depends on search, selection, bias assessment and study comparability.

Step 3: Inspect who, what and how

Read the methods before trusting the conclusion. Check inclusion criteria, sample size, randomization, blinding, comparison group, duration, dropouts, adherence and whether the outcome measure was valid.

• Does the sample resemble the people in the headline?
• Was the comparison fair and planned in advance?
• Did groups differ at baseline?
• Were many outcomes tested but only a few highlighted?
• Were missing data or dropouts handled transparently?
• Was the study long enough to observe a meaningful outcome?

Step 4: Read magnitude and uncertainty

A p-value addresses compatibility with a statistical model; it does not tell you whether an effect is large, important, unbiased or likely to apply to you. Look for the absolute difference, relative difference, confidence interval and the number of events.

A large relative change can describe a tiny absolute change when the starting risk is low. A wide confidence interval may include both trivial and meaningful effects. If only a surrogate marker changed, ask whether changing that marker has been shown to improve outcomes people care about.

Step 5: Put one result into the evidence map

• Read the stated limitations—and look for important ones the authors did not emphasize.
• Check trial registration or a published protocol when available.
• Look for conflicts of interest without assuming funding alone invalidates data.
• Compare with systematic reviews, guidelines or independent replications.
• Notice whether the conclusion language matches the design: associated with, improved, reduced, or caused.