What a Negative Control Tells You
A negative control asks a related question whose causal answer should be null, helping reveal confounding, selection, measurement, or analytic bias.
Health & Evidence Library
Tools for reading studies, interpreting tests and effect estimates, recognizing uncertainty, and asking whether evidence applies.
207 guides · Page 2 of 4
A negative control asks a related question whose causal answer should be null, helping reveal confounding, selection, measurement, or analytic bias.
Learn how a funnel plot displays small-study effects, why asymmetry has several causes, and when visual or statistical tests are informative.
A control group estimates what would happen without the tested intervention and defines the comparison a study can legitimately claim.
A confidence interval describes estimate precision under a model. It is not the probability that a claim is true or that one result will repeat.
How laboratory work, clinical trials, regulatory review, recommendations, safety surveillance, and causal assessment fit together.
How win-ratio analyses rank outcomes, compare patient pairs, handle ties and censoring, and sometimes hide which component drives a result.
Sensitivity and specificity can change across severity, alternatives, settings, and pathways. Learn how spectrum effects alter diagnostic accuracy.
Why one study rarely settles a scientific question, how chance and bias shape results, and how replication and synthesis strengthen conclusions.
Diagnostic accuracy is measured against a reference standard. When that standard makes errors, sensitivity and specificity can mislead.
Evidence hierarchies rank designs by the biases they usually control, but the right design depends on the question and quality still matters.
The fragility index counts outcome changes needed to cross a significance threshold. It is intuitive, narrow, and easy to overinterpret.
When studies stay unpublished because of their results, the visible literature can exaggerate benefits and hide uncertainty. Here is how to detect it.
An estimand states the exact treatment effect a trial seeks, including population, treatment, endpoint, later events, and summary measure.
Why associations among countries, counties, or hospitals cannot automatically describe the people within them, and when group data remain useful.
Causal questions compare what happened with what would have happened under another action. Learn potential outcomes, assumptions, and target trials.
Target trial emulation starts with the randomized trial researchers wish they could run, then maps that protocol transparently to observational data.
A biomarker can shorten a trial without guaranteeing patient benefit. Learn validation, trial-level prediction, failure modes, and confirmatory evidence.
A p-value addresses compatibility with a statistical model. Clinical importance depends on effect size, uncertainty, baseline risk, harms, burden, and values.
SPRINT lowered cardiovascular events and mortality in selected high-risk adults. Learn its targets, measurement protocol, harms, and limits.
Funnel-plot asymmetry can have many causes. Learn what Egger's test and trim-and-fill estimate, their assumptions, and safer interpretation.
Randomized journal trials tested masked authors, signed reviews, and public reports. Their narrow results do not settle whether peer review works overall.
Learn how sensitivity, specificity, predictive values, prevalence, thresholds, and likelihood ratios change the meaning of a diagnostic test.
The ten-events-per-variable rule is not enough. Learn how shrinkage, outcome frequency, predictor parameters, and precision guide model sample size.
How the 2026 dyslipidemia guideline combines PREVENT risk, risk enhancers, Lp(a), apoB, and coronary calcium in primary prevention.
Restricted mean survival time reports average event-free time through a chosen horizon and remains interpretable when hazards are not proportional.
How research ethics turns contextual vulnerability into added safeguards while avoiding unjustified exclusion from evidence and benefits.
How current US policy distinguishes fabrication, falsification, and plagiarism from detrimental practices, disagreement, and honest error.
A practical guide to computational reproducibility, new-data replication, terminology differences, and what each can and cannot establish.
A diabetes-specific guide to reproducible analysis, replication, assay harmonization, CGM definitions, transparent protocols, and data sharing.
How CONSORT 2025 and the EQUATOR Network make health research checkable, and why complete reporting does not guarantee a sound study.
Relative risk compares probabilities; an odds ratio compares odds. They are similar for rare outcomes and can diverge sharply for common outcomes.
Extreme measurements often move toward average on repeat testing. Learn why this happens, how it mimics treatment benefit, and how studies control it.
What esketamine trials show about rapid symptom change, monotherapy, relapse prevention, masking, selected populations, safety, and regulatory limits.
A practical guide to target trials, confounding, immortal time, result-level judgments, and the draft six-domain ROBINS-I V2.
How objective response rate is defined, measured, confirmed, and interpreted alongside duration, independent review, missing scans, safety, and survival.
What ANDROMEDA-SHOCK tested, why its primary mortality result was negative, what the confidence interval permits, and how later evidence should be separated.
How network meta-analysis combines direct and indirect comparisons, why transitivity is essential, and why treatment rankings can overstate certainty.
Interpret pragmatic and explanatory trials by matching eligibility, setting, delivery, follow-up, outcomes, and analysis to the study purpose.
How I-squared, tau-squared, Cochran's Q, and prediction intervals describe different parts of between-study variation in a meta-analysis.
How QUADAS-2's four domains identify bias in diagnostic studies, and what changed when QUADAS-3 became the recommended version in 2026.
How Turner compared FDA records with journal articles, quantified selective publication, and showed why trial registries matter.
How propensity scores support observational comparisons through matching or weighting, and how to audit balance, overlap, and residual bias.
How PREVENT differs from the Pooled Cohort Equations in population, outcomes, inputs, race, kidney measures, time horizon, calibration, and guideline use.
How protocols, registries, and statistical analysis plans constrain flexibility, when deviations are legitimate, and why post hoc findings can still be useful.
Why good predictive performance does not prove clinical benefit, and how impact studies test workflow, decisions, outcomes, harms, cost, and equity.
How prediction intervals describe between-study heterogeneity, why they differ from confidence intervals, and when their apparent precision can mislead.
A practical guide to accuracy, trueness, precision, repeatability, reproducibility, bias, uncertainty, calibration, and traceability in clinical measurement.
How comments, corrections, expressions of concern, and retractions work after publication, plus a practical method for weighing public criticism.
Peer review adds expert scrutiny but cannot certify truth, integrity, or replication. Learn how to weigh preprints, revisions, and corrections.
How human, animal, plant, and environmental systems interact, how climate shapes health pathways, and how to interpret risk without overclaiming causation.
NRI and IDI can describe changes between prediction models, but calibration, thresholds, optimism, and clinical utility determine whether changes matter.
Missing trial outcomes can change an estimate when absence relates to health or treatment. Imputation and sensitivity analyses make assumptions visible.
Meta-research studies research methods, reporting, verification, evaluation, and incentives so scientific reform can be tested with evidence.
Meta-regression relates study features to effect estimates, but ecological bias and sparse study counts limit individual-level conclusions.
How medication lists, reconciliation, interaction checks, barrier-aware adherence support, and clear follow-up reduce preventable harm.
Mediation analysis separates direct and indirect effects, but causal interpretation depends on timing, measurement, and confounding assumptions.
Judge systematic review currency from its last search, new evidence, current question, methods, certainty, and update process.
How IPD meta-analysis checks and harmonizes participant records, preserves trial structure, studies effect modifiers, and handles missing data sets.
How to appraise event definitions, collection, denominators, recurrence, time at risk, withdrawals, coding, missing data, and rare harms in trials.
How duplicated, spliced, or selectively adjusted figures are detected, investigated, and interpreted without assuming intent.