Ethical, physician-aware medical research support
Source-reviewed knowledge center
Each guide shows who prepared it, its current review status, sources, publication date, last-reviewed date, and correction history. Expert review is not claimed until a named reviewer has approved the text.
A primary-source-linked guide to documenting the design and model assumptions behind coefficients, confidence intervals, p-values, and predictions.
Read the complete articleA publication series spanning data definitions, missingness, analysis planning, uncertainty, sample size, outcomes, data quality, and model assumptions.
Coefficients, p-values, and confidence intervals are conditional on design and model assumptions that must be examined, documented, and stress-tested.
Read article →Independent reviewIndependent review can detect systematic errors, undocumented assumptions, and confirmation bias that the person who cleaned the data may no longer see.
Read article →Data cleaningEvery correction, transformation, recode, merge, and exclusion that can affect a result should be traceable to a rule, reason, reviewer, and data version.
Read article →Protocol integrityA three-document crosswalk reveals unexplained changes in outcomes, sample size, eligibility, analyses, and study status before publication.
Read article →Outcome definitionA reproducible endpoint specifies the variable, participant-level metric, method of aggregation, time point, assessor, and decision rules.
Read article →Research feasibilityA scientifically justified target still fails if the institution cannot recruit, retain, measure, and analyze that sample responsibly.
Read article →Sample sizeA sample-size calculation should declare whether the study is designed to test a target difference or estimate a quantity within an acceptable margin.
Read article →Decision communicationDecision-ready communication shows what is known, what remains plausible, which assumptions matter, and what action is justified now.
Read article →UncertaintyAn interval helps readers judge the estimated effect, its precision, and which clinically important possibilities remain compatible with the data.
Read article →Change controlA proportionate amendment pathway distinguishes routine implementation details from changes that alter the scientific question, bias risk, or interpretation.
Read article →Analysis planningA credible report separates tests planned before outcome data were examined from analyses prompted by patterns found later.
Read article →Missing-data governanceInstitutions should define when missingness requires documentation, prevention, sensitivity analysis, escalation, or a limit on the conclusion.
Read article →Missing dataBlank, unknown, not applicable, not collected, and structurally unavailable values represent different events—and collapsing them can undermine analysis.
Read article →Data governanceData definitions need named decision rights, because an unowned variable can change meaning across forms, systems, analysts, and reports.
Read article →Data provenanceA compact, versioned variable specification prevents hidden disagreements about definitions, codes, units, timing, and missingness from reaching the analysis.
Read article →A sample-size formula cannot rescue an ambiguous population, outcome, contrast, time point, or handling of intercurrent events.
Version 1.0 · 24 July 2026 · Initial publication. No corrections recorded.
Statistical significance alone does not establish clinical importance, causality, absence of bias, model validity, or replicability.
Version 1.0 · 24 July 2026 · Initial publication. No corrections recorded.
A defensible project requires the researcher to explain why the design and analysis answer the question—and where they do not.
Version 1.0 · 24 July 2026 · Initial publication. No corrections recorded.
Removing names is necessary but may not be sufficient when dates, rare conditions, geography, free text, or linked variables can identify a person.
Version 1.0 · 24 July 2026 · Initial publication. No corrections recorded.
Naming a tool is not enough. Authors should document what it did, where its output appears, and how humans verified accuracy and originality.
Version 1.0 · 24 July 2026 · Initial publication. No corrections recorded.
Downloadable working tools
Question, design, variables, analysis, ethics, privacy, and ownership.
Download .md ↘DefenseScientific story, methods, interpretation, ethics, and accountability.
Download .md ↘PrivacyAuthority, direct identifiers, indirect risk, and release decision.
Download .md ↘Corrections
