How Preprints Are Screened Before Posting
Preprint screening checks scope, ethics signals, privacy, duplication, and possible harm, but it does not certify the research through peer review.
Health & Evidence Library
Tools for reading studies, interpreting tests and effect estimates, recognizing uncertainty, and asking whether evidence applies.
207 guides · Page 1 of 4
Preprint screening checks scope, ethics signals, privacy, duplication, and possible harm, but it does not certify the research through peer review.
Journal embargoes coordinate reporting around publication; prior-publication rules decide whether earlier sharing affects manuscript eligibility.
A screening test's value depends on population risk, test thresholds, treatment benefit, follow-up capacity, harms, and the full care pathway.
A pooled estimate can mislead when studies answer different questions, use incompatible outcomes, contain major bias, or vary in effect direction.
The STAR*D debate shows how outcome definitions, analysis populations, missing data, and protocol fidelity can materially change a trial's headline.
A practical guide to claims, electronic records, registries, devices, and linked data, including what each source records and systematically misses.
How procalcitonin and C-reactive protein can support antibiotic stewardship without replacing diagnosis, severity assessment, cultures, or reassessment.
Trustworthy science communication makes evidence traceable, calibrates certainty, quantifies effects, discloses interests, and corrects errors visibly.
A useful biomarker must be measurable, valid for a defined purpose, and capable of improving a decision. Association alone is not enough.
Major replication projects found smaller effects, mixed repeatability, and practical barriers. Their results require more nuance than a single failure rate.
Good clinical evidence fits a precise question and withstands scrutiny for bias, precision, relevance, transparency, replication, and applicability.
External validity asks whether a study's results apply to a defined target beyond the study, across people, settings, treatments, outcomes, and time.
Learn how ctDNA molecular residual disease testing estimates recurrence risk, why a negative result cannot prove cure, and when treatment utility needs trials.
Verification bias occurs when disease confirmation depends on the test result or patient features. Learn how it distorts sensitivity and specificity.
Learn how the USPSTF combines age, risk factors, and 10-year cardiovascular risk when recommending statins to prevent a first heart attack or stroke.
How standardized mean differences combine outcomes measured on different scales, and why scale direction, variability, bias, and context matter.
Selection bias explained through recruitment, follow-up, referral, missing data, conditioning, and the checks that make study results more credible.
Competing events change the meaning of a survival curve. Cumulative incidence, cause-specific hazards, and Fine-Gray models answer different questions.
Umbrella reviews synthesize reviews, but overlap, variable methods, stale searches, and inherited bias can overstate certainty.
Trial registration records the planned question, while structured results reporting reveals completed studies and outcome changes.
A multivariable regression table may display many coefficients, but each causal question can require a different adjustment set and interpretation.
Placebo responses mix context, natural change, and measurement. Learn what placebo effects can change, what they cannot prove, and why controls matter.
Nocebo effects can amplify symptoms through expectation and learning. Good communication reduces avoidable harm without hiding genuine treatment risks.
After approval, real-world data can detect rare harms and study use in broader populations, but design and data quality determine credibility.
Learn how noninferiority margins, assay sensitivity, confidence intervals, missing data, and analysis populations determine whether a trial is credible.
Disease, product, quality, and trial registries serve different purposes. Their credibility depends on coverage, definitions, and follow-up.
Earlier diagnosis automatically lengthens measured survival from diagnosis. Learn how mortality trials separate true screening benefit from lead-time bias.
Instrumental-variable studies can address unmeasured confounding only under demanding assumptions. Learn relevance, independence, exclusion, and LATE.
Unexpected findings on scans, tests, and sequencing can trigger useful care or harmful cascades. Learn how context and follow-up guidance shape decisions.
Immortal time bias can make treatment look protective when pre-treatment survival is credited to treatment. Learn how to detect and prevent it.
A hazard ratio compares instantaneous event rates, not simple probabilities. Learn proportional-hazards assumptions, censoring, and better companion measures.
Learn how patient selection, reference standards, thresholds, verification, and prevalence determine whether diagnostic accuracy results will travel.
A practical guide to iodinated, gadolinium-based, ultrasound, and gastrointestinal contrast, including kidney and reaction risk.
Confounding by indication can make helpful treatments look harmful, or ineffective treatments look useful, in observational medical research.
Length-time bias makes screen-detected cancers look less lethal because screening preferentially finds tumors with longer detectable phases.
Why raw retraction counts are increasing, how paper mills and better detection contribute, and which metrics reveal whether correction works.
Why retraction notices fail to reach every PDF, reference list, review, guideline, and database, and how to stop unreliable evidence spreading.
What Research Use Only means for laboratory products, why it cannot support patient claims, and which evidence a clinical test needs.
How pulse oximeters estimate oxygen, why darker pigmentation can bias readings upward, and how safer interpretation reduces missed hypoxemia.
How awareness of treatment assignment affects psychotherapy trials and which design choices can preserve credible comparisons.
Why overall survival is uniquely persuasive in oncology, where it can mislead, and how to read earlier endpoints alongside it.
How indication, population, endpoints, duration, background care, safety questions, and estimands differ between obesity and diabetes programs.
How low pretest probability, false positives, incidental findings, overdiagnosis, and follow-up harms can make an untargeted test panel unsafe.
Why a global confirmatory program may need more participants than the primary efficacy calculation, from safety and regions to subgroups and missing data.
What FDA's 2024 complete response said about efficacy, safety reporting, functional unblinding, durability, and the path for new evidence.
How a 2008 FDA safety policy created the modern diabetes CVOT, what those trials test, what they discovered, and how U.S. policy later changed.
Bias labels describe recurring reasoning patterns, but diagnostic error also reflects knowledge, data, workload, teamwork, and system design.
Current medical-publishing rules keep authorship human, require accountable verification, and call for transparent disclosure of AI assistance.
What the USPSTF found insufficient about universal adult suicide-risk screening, and why that does not limit assessment when concern exists.
Data monitoring committees review interim benefit, harm, and futility using prespecified rules while protecting trial integrity.
ICMJE defines who qualifies for authorship; CRediT records contributor roles. Learn how accountability, AI use, groups, and acknowledgments fit.
Real-world evidence can describe care, compare outcomes, and detect harms, but confounding, selection, missing data, and timing can mislead.
How duplicate publication, fragmented reporting, and legitimate secondary publication affect readers and evidence synthesis.
Proportional hazards is an assumption about a stable rate ratio over time. Curves, residuals, and time-based alternatives reveal when it fails.
Automated consistency checks can flag impossible means and mismatched p values, but every flag needs context and human review.
PROBE trials keep treatment open but blind endpoint assessment. This limits some judgment bias without repairing every consequence of open treatment.
A subgroup analysis tests whether effects differ across groups. Credibility depends on interaction tests, planning, precision, and replication.
A p value measures how unusual a result is under a stated statistical model. It is not the probability that a hypothesis is true.
A lab reference interval describes results in a defined reference population. It is not a universal border between health and disease.
A practical guide to confidence intervals, power, equivalence tests, and the difference between no evidence of effect and evidence of no effect.