jama-study-design

Study Design & Internal Validity (jama-study-design)

When to trigger

• Choosing or defending the design: RCT vs cohort vs case-control vs diagnostic
• A reviewer will ask whether the analysis is intention-to-treat or per-protocol
• Bias, confounding, or missing data threaten the primary inference
• The design and the reporting checklist need to be aligned before writing

Design-specific safeguards JAMA reviewers expect

Randomized clinical trials

• A single, pre-specified primary outcome; secondary outcomes clearly labeled
• Adequate randomization (sequence generation) and allocation concealment
• Blinding of participants, clinicians, outcome assessors where feasible — state what was blinded
• Intention-to-treat as the primary analysis; per-protocol only as secondary/sensitivity
• A priori sample-size / power calculation tied to the primary outcome
• Pre-defined stopping rules and handling of interim analyses

Cohort / case-control (observational)

• Explicit confounding control: prespecified covariates, adjustment strategy, DAG reasoning
• Clear definitions of exposure, outcome, and follow-up windows; avoid immortal-time bias
• Selection-bias and information-bias appraisal; how participants entered the sample
• Association, not causation — design and language must respect this

Diagnostic-accuracy studies

• Pre-specified reference standard, applied to all participants, blinded to index test
• Consecutive or random enrollment; report spectrum of disease
• Pre-defined thresholds; report sensitivity/specificity/predictive values with CIs

Systematic reviews / meta-analyses

• Pre-registered protocol (e.g., PROSPERO), pre-specified eligibility and outcomes
• Comprehensive, reproducible search; duplicate screening and extraction
• Risk-of-bias assessment; pre-planned heterogeneity and sensitivity analyses

Decision table

Question	Design / action
Does an intervention cause an outcome?	RCT; if infeasible, strong quasi-design
What is the prognosis / risk of an exposure?	Prospective cohort with confounder plan
How accurate is a test?	Diagnostic-accuracy study (vs reference)
What does the totality of evidence show?	Systematic review ± meta-analysis
Rare outcome, exposure already occurred	Case-control (watch selection bias)

Checklist

• Primary outcome is single, pre-specified, patient-relevant
• For RCTs: randomization, allocation concealment, blinding stated; ITT is primary
• Sample-size / power calculation present and tied to the primary outcome
• Confounding/bias strategy explicit and pre-specified (observational)
• Reference standard and blinding defined (diagnostic)
• Protocol pre-registered (RCT and systematic review)
• Missing-data handling pre-specified (not improvised post hoc)
• Design choice matches the EQUATOR checklist you will use

Anti-patterns

• Promoting a secondary or post hoc outcome to "primary" after seeing results
• Per-protocol analysis presented as primary for an RCT
• "Adjusted for everything" with no pre-specified covariate rationale
• Causal claims from an observational design
• Reference standard chosen or applied after knowing the index-test result
• No power calculation, then attributing a null result to "trends"

Worked example: locking the design (illustrative)

Vignette (illustrative): a multicenter randomized clinical trial, N = 3,400 adults with acute kidney injury across 18 sites, early vs standard renal-replacement timing; pre-specified primary outcome 28-day all-cause mortality, 28.5% vs 31.2%, absolute risk difference -2.7 percentage points (95% CI, -6.1 to 0.7). For a Journal of the American Medical Association Original Investigation, the design safeguards a JAMA reviewer checks: a single pre-specified primary outcome, documented allocation concealment, intention-to-treat as primary (per-protocol only as sensitivity), and an a priori power calculation tied to mortality. The 95% CI crossing zero means the trial does not establish benefit — the Conclusions must say "no significant difference," not "a trend toward benefit."

Reviewer pushback and the JAMA fix

• "Is this intention-to-treat or per-protocol?" Fix: make ITT the primary analysis and relegate per-protocol to clearly labeled sensitivity.
• "Primary outcome looks chosen after seeing the data." Fix: cite the registry/protocol; demote any post hoc outcome to exploratory.
• "Causal language from an observational design." Fix: switch to associational verbs and frame confounding as a residual threat.

Calibration anchors (hedge where uncertain): a single pre-specified primary outcome, allocation concealment, ITT-as-primary, and an a priori power calculation are durable JAMA expectations; the matching EQUATOR checklist (CONSORT/STROBE/STARD/PRISMA) follows from the design — confirm specifics against current author guidelines.

Operating pass for JAMA

Run this as a concrete capability pass. First lock the clinical question, patient population, estimand or endpoint, safety/ethics issue, and reporting checklist; then test whether the manuscript addresses clinical reviewers who ask whether the evidence changes patient care, policy, or medical decision-making while satisfying reporting standards.

• Primary move: Return a claim-evidence-risk ledger; every recommendation must point to a manuscript location or missing artifact.
• Decision ledger: return claim / evidence / blocker / next edit rows so the next pass can patch the manuscript directly.
• Sibling comparison: compare against NEJM for field-changing clinical medicine, Lancet for global-health breadth, specialty journals for narrower clinical domains; if the neighboring outlet has the stronger audience claim, recommend re-routing before polishing.
• Verification floor: before submission-ready advice, re-open resources/official-source-map.md for volatile rules and name the one unresolved fact that could change the recommendation.

Output format

【Design】RCT / cohort / case-control / diagnostic / systematic review
【Primary outcome】... (pre-specified: yes/no)
【Key validity safeguards in place】...
【Validity gaps to fix】...
【Causal vs associational claim】...
【Matching EQUATOR checklist】CONSORT / STROBE / STARD / PRISMA
【Next skill】jama-reporting-standards