How to Read a Scientific Study

PICO, Study Design, and the Numbers That Matter

Download PDF

How to Read a Scientific Study

PICO, Study Design, and the Numbers That Matter

Most APPs read the abstract conclusion first and work backward. That is the wrong order. The conclusion is the authors’ opinion. The methods determine whether it is credible. This card gives you a system for reading any study in under five minutes.

Part 1 — The PICO Framework: Read the Abstract in 60 Seconds

Every study asks a question. PICO is the structure of that question. If you cannot find all four elements in the abstract, the study is not well reported.

Letter Element What to Look For
PPopulationWho was enrolled? Age, comorbidities, exclusion criteria. Is this your patient?
IInterventionWhat was done, at what dose, for how long?
CComparisonWhat was the control? Placebo, usual care, active comparator, or nothing?
OOutcomeWhat was measured? Primary endpoint, secondary endpoints. Is this a hard outcome (death, MI) or a surrogate (biomarker, imaging)?
60-second abstract scan: Background (why this study?) → Methods (PICO: who, what, how randomized?) → Results (effect size + CI, not just p-value) → Conclusion (do the authors’ words match the actual numbers?)

Part 2 — Study Design Hierarchy

Study design determines what a paper can and cannot prove. Hierarchy runs from weakest to strongest for establishing causation.

Design Can Prove Cannot Prove
Case report / Case seriesSignal detection, rare adverse eventsCausation, generalizability
Cross-sectionalPrevalence, associations at one point in timeCause and effect (no temporality)
Case-controlAssociation, odds ratios; good for rare diseasesIncidence; subject to recall bias
Cohort (observational)Incidence, risk factors over timeCausation (confounding cannot be fully eliminated)
RCTCausation (randomization controls confounding)Efficacy in populations excluded from trial
Meta-analysis / Systematic reviewPooled effect size, consistency across trialsQuality above the included studies

Practical rule: An observational study cannot establish that a treatment works. It can generate a hypothesis. Only a well-conducted RCT with hard outcomes can change practice.

Part 3 — Numbers That Matter

Metric What It Means Red Flag
Effect sizeMagnitude of the treatment effect (HR, RR, OR, mean difference)Statistically significant but clinically trivial
Confidence intervalRange of plausible true effects at 95% confidenceWide CI = imprecise estimate; crosses null = not significant
p-valueProbability of seeing this result if the null were truep = 0.049 vs. 0.051 is not a meaningful distinction
NNTPatients you must treat to prevent one eventNNT >100 over 5 years rarely justifies risk or cost
Follow-up durationHow long patients were observedShort follow-up misses late harms and late benefits
Lost to follow-upPatients who dropped out or were excluded from analysis>10–20% attrition threatens validity

Internal validity = was the study well conducted? (randomization, blinding, ITT analysis, low dropout)

External validity = does it apply to your patient? (age, comorbidities, excluded populations, real-world dose)

Red flags to look for: Industry funding with no independent statistical analysis. Composite endpoints (death + MI + hospitalization + revascularization) where soft endpoints drive the result. Per-protocol analysis instead of intention-to-treat when dropout was high. Surrogate endpoints (LDL, EF, biomarker) reported as if they equal hard outcomes.
Clinical Rule

Read the Methods before the Conclusion. The conclusion is the authors’ opinion — the methods tell you whether to believe it. An RCT with 30% dropout reported per-protocol is not a clean RCT. A p < 0.05 in a composite endpoint driven by soft outcomes is not practice-changing.

This is one of 13 free reference sheets from the APP Cardiology Academy — no account required.

Browse all resources See the full curriculum