Rounds & Square Pegs

Read in this order, not top to bottom

Start with the sample size and the setting. Who was studied, where, and how many. This tells you what the findings can and cannot be generalised to before you invest time in the rest.

Next, find the outcome measure. What, specifically, did they measure, and how. If you cannot state the primary outcome in one sentence after reading this section, the paper has not told you yet, or you need to reread it once.

The table that matters most

Find the table comparing baseline characteristics between groups, usually Table 1. If the groups look meaningfully different before the intervention even starts, treat every later comparison with more scepticism. This single table does more to tell you whether a study is trustworthy than most of the discussion section.

Worked example: a fake discharge pathway study

Imagine a paper comparing two discharge pathways at a 420-bed urban hospital:

Pathway A (standard): usual discharge planning, social work referral if flagged.
Pathway B (enhanced): pharmacist med reconciliation at bedside, structured follow-up call at 72 hours.

Primary outcome: 30-day readmission rate. Secondary: length of stay, patient-reported preparedness (a 5-point scale).

Sample: 312 patients per arm, enrolled over 14 months. Inclusion: adults on general medicine, expected discharge home. Exclusion: hospice, transfer to another acute facility, left against medical advice.

Here is what Table 1 might look like (invented numbers):

Characteristic	Pathway A (n=312)	Pathway B (n=312)
Age, mean (SD)	68.4 (12.1)	66.9 (11.8)
Female, n (%)	171 (54.8)	158 (50.6)
Charlson index, mean (SD)	3.2 (1.8)	2.6 (1.6)
Index admission LOS, mean days (SD)	5.1 (2.4)	4.8 (2.2)
Prior admission in 12 mo, n (%)	89 (28.5)	72 (23.1)
Lives alone, n (%)	94 (30.1)	88 (28.2)

What to notice before you read the results:

Groups are similar on demographics and living situation. Good sign.
Pathway B has a lower mean Charlson index (2.6 vs 3.2) and fewer prior admissions (23.1% vs 28.5%). Not enormous, but the enhanced pathway may have enrolled slightly healthier patients, or randomisation failed in ways the authors did not emphasise.
If the paper then claims a 40% relative reduction in readmissions, you ask: could sicker patients have landed disproportionately in Pathway A? Table 1 gives you permission to ask that out loud.

You do not need to run statistics. You need to notice whether the groups were comparable on the things that predict your outcome before anyone intervened.

What to skip on a first pass

Skip the full statistical model specification unless you specifically need to evaluate the analysis. Skip most of the limitations paragraph until you have decided the study matters to you; limitations sections are frequently boilerplate.

Stop reading: a decision tree

Use this when you have ten papers and an hour, or when a colleague forwards something with "thoughts?"

Start: read abstract sample + setting + primary outcome
  |
  +-- Sample clearly wrong for your context? (paediatric study, you run geriatrics)
  |     --> STOP. File under "not applicable."
  |
  +-- Primary outcome not something you care about?
  |     --> STOP unless the method itself is what you came for.
  |
  +-- Find Table 1 (or equivalent baseline table)
  |     |
  |     +-- Groups wildly different on key predictors of outcome?
  |     |     --> READ RESULTS WITH SKEPTICISM. Maybe stop if you have better options.
  |     |
  |     +-- Groups look comparable?
  |           --> Continue to results. Check effect size, not just p-value.
  |
  +-- Results: is the effect clinically meaningful, not just statistically significant?
        |
        +-- No? --> STOP. Do not cite this in a meeting.
        |
        +-- Yes? --> Read discussion for confounders the authors admit. Decide if you believe it.

The tree is blunt on purpose. You are not trying to become a methodologist. You are trying to decide, quickly and defensibly, whether a given study should change what you believe or do.

What to do with all this