Method

How I read a health paper in 15 minutes.

2026-04-12 · 5 min read · Data & Metrics

Most health papers do not need an hour. The ones that do, you will know inside the first five minutes anyway. What follows is the workflow I have drifted into after a few hundred papers. Not because it is clever. Because it survives contact with the typical paper.

The order is the trick

Read in this order. Not the order the paper is printed in. This one.

  1. Title. What is the claim?
  2. Funding and author affiliations. Thirty seconds. A supplement-industry-funded trial of that same supplement goes on a much shorter leash unless the design is unusually tight.
  3. Methods, specifically the design. RCT, observational, or mechanistic. Crossover or parallel. Blinded. Pre-registered. This alone tells you what the paper is allowed to conclude, before you have read a single result.
  4. Sample size and the primary outcome. Is it powered? Is the primary outcome the one in the title, or is the title quietly promoting a secondary?
  5. Effect sizes in the results table. Effect sizes. Not p-values. We will come back to why.
  6. Limitations. Authors are usually honest here, and the limitations they own up to are usually the ceiling on how seriously you should take the whole thing.
  7. The abstract, last. Now you can see how it has been framed, and the gap between "what they showed" and "how the abstract sells it" tells you a lot.
I do not read the abstract first. It is the most heavily edited part of the paper and the part most optimised to walk you to a conclusion before you have checked any of the working.

The five questions I am really answering

  1. Who was studied, and is it me or a population I care about?
  2. What was the dose, and is it realistic outside a lab?
  3. What was the comparator, and is it the right one?
  4. How big is the effect, in units I can picture?
  5. Has anyone replicated it?

Any one of those answered badly and the paper is interesting but I am not changing behaviour over it. All five answered well, and it is still one paper. I will wait for two more before I act.

What to ignore

Press releases. Twitter threads with bold conclusions. "Studies show…" articles that never link the study. Citation count. High citation count tracks "provocative" far more closely than it tracks "true." Anything built on the phrase "could potentially."

What to read slowly

Meta-analyses with a funnel plot. Pre-registered RCTs. Anything where the authors put their analysis code online for you to break. Replications, especially replications by people not friendly with the original authors. Null results, which are unfashionable, hard to publish, and for exactly those reasons unusually informative when they do show up.

The shortcut. Read the methods first. If the methods are bad, you can put the paper down. Fifty minutes saved. That single habit, methods-before-everything, is the highest-leverage thing I have picked up in five years of reading this stuff.

Try the paper sanity check on something you read recently. If the verdict surprises you, sit with it for a minute.