“Linked to” versus “causes”: what can a health study actually establish?
Why do observational studies and trials sometimes seem to disagree?
On this page
A study design describes how people, exposures, comparisons and outcomes were selected and measured. An investigator may assign an intervention, observe an existing exposure, or describe a condition. These choices affect the questions the data can answer and the errors that need attention.[1]
What does this actually mean?
The meaning
“Linked to” usually means that two things were associated in the data. “Causes” makes a stronger claim: changing one would change the other under the relevant conditions. Study design helps readers judge that claim. A study can describe a pattern, test a treatment, evaluate a diagnostic test or predict an outcome; those are different jobs.[9]
What observation can miss
In an observational study, researchers do not randomly assign the exposure. People who choose a supplement, diet or treatment may also differ in income, smoking, health or access to care. Such differences can partly explain an association: this is confounding. Adjustment can help, but it cannot guarantee that every important difference was measured correctly.[12]
The direction can run backward
Sometimes illness changes the behavior being studied rather than the behavior causing illness. That is reverse causation. For example, symptoms may lead people to change what they eat before a diagnosis. Looking at when exposure and symptoms occurred helps, but collecting data prospectively is not, by itself, proof that the exposure caused the outcome.[13]
What randomization improves
Random assignment helps make comparison groups similar, including on factors researchers do not know to measure. A well-conducted randomized trial can therefore give stronger evidence about the tested intervention’s effects. Still ask who participated, what the comparison involved and which outcome and duration were studied. The word randomized does not erase poor execution or missing data.[9]
Why results can differ
Studies may test different treatments, populations, doses, outcomes or follow-up periods, as well as having different biases. Disagreement is not always one design “winning.” A published comparison of observational and randomized evidence found no consistent rule that observational treatment effects were always larger. Evaluate the actual question and methods before declaring that a new study overturned an entire field.[11]
What to check
Locate the original study and identify its design, comparison group, outcome and main limitations. A report of a health event after a medicine is useful for surveillance, but sequence alone does not establish causation. FDA explains that adverse-event data need clinical assessment and often other evidence; a report count is not a treatment-risk estimate.[10]
Identify the question and the design before comparing findings
| Family | How it is organized | Main interpretation |
|---|---|---|
| Descriptive report or series | Describes people or events without a comparison group. | Can identify patterns and generate hypotheses; lacks an internal comparative effect estimate.[1] |
| Experimental study | Investigators assign an intervention, randomly or by another rule. | Randomized and nonrandomized assignment need to be distinguished.[1] |
| Cohort | Defines a population and follows exposure and outcomes over time. | Can estimate incidence; exposure groups may differ for other reasons.[2] |
| Case-control | Selects people with an outcome and comparison controls; compares earlier exposure. | Control selection and exposure measurement matter; the sampled case proportion is not population incidence.[2] |
| Cross-sectional | Measures characteristics in a population at a particular period. | Can estimate prevalence; temporal direction may be unclear.[2] |
| Diagnostic accuracy | Compares an index test with a defined reference standard. | Accuracy is distinct from the effects of using a testing pathway.[3] |
| Prediction model | Develops or evaluates predictions from multiple characteristics. | Performance must be evaluated for its intended population and use.[4] |
| Natural experiment and quasi-experimental evaluation | Studies a change outside evaluator assignment using a nonrandomized comparison. | Causal interpretation depends on the assignment process and credible comparison assumptions.[5] |
Study labels do not replace the actual sequence
STROBE recommends describing the important design features early. A cohort can use records that already exist: it still organizes observation from exposure toward subsequent outcomes. Case-control research starts with sampled outcome groups. Calling a paper retrospective does not distinguish these designs. Readers need dates of recruitment, exposure measurement and follow-up, eligibility, missing information and the basis of the comparison. Transparent reporting helps appraisal; STROBE is a reporting guideline, not a certificate that associations are causal.[2]
Qualitative research investigates experiences and context
Qualitative studies use material such as interviews, observations and documents to understand experiences, meanings or social processes. They can investigate why care is difficult to use or how illness affects daily life. The sampling and analysis should fit the question; selected interview accounts do not themselves supply a population prevalence or treatment-effect estimate.[15]
Mixed methods need a connection between the forms of evidence
Mixed-methods research combines qualitative and quantitative work to address a connected question. In an explanatory sequence, measured findings can guide later interviews; in an exploratory sequence, interviews can help develop what a later survey measures. Another design analyzes the two together. The connection can happen through sampling, measurement, analysis or interpretation; merely placing unrelated results side by side leaves the purpose unclear.[16]
Evidence & uncertainty
A prediction model and a causal treatment comparison have different purposes. TRIPOD+AI asks researchers to describe the intended population, outcome, time horizon, data and performance evaluation. A model that predicts an outcome accurately does not by itself show which intervention will change that outcome.[4]
Selected-source overview · Independent review pendingPublished questions
Can a case series estimate treatment benefit against an alternative?
Selected-source overview · Independent review pending. Read the evidence context.
Without an appropriate comparison, improvement could reflect natural history, other care or selection. A series can describe experience and generate a hypothesis, but it cannot supply the same internal comparative estimate as a controlled study.[1]
Evidence behind this answer
- An overview of clinical research: the lay of the land
original methods primer · 2002
Selected original published PDF passages inspected through Northwell academic copy; PubMed indexed record inspected · Taxonomy and comparison/time-direction definitions
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Does retrospective automatically mean case-control?
Selected-source overview · Independent review pending. Read the evidence context.
No. A cohort can examine existing records while organizing exposure before subsequent outcomes. Case-control selection starts with outcome groups. Eligibility, dates, sampling and measurement show the design more reliably than the word retrospective.[2]
Evidence behind this answer
- STROBE: Explanation and Elaboration
original reporting guideline · 2007
Selected original publisher HTML passages inspected · Three observational designs; early design/date/selection reporting; reporting not quality certificate
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Was the case proportion in the 1950 smoking study a population incidence?
Selected-source overview · Independent review pending. Read the evidence context.
No. Cases and hospital comparison patients were deliberately sampled. Their proportions describe the study’s selection, while the exposure comparison investigates association. They do not directly give the population probability of developing lung cancer.[7]
Evidence behind this answer
- Smoking and carcinoma of the lung: preliminary report
original historical case control study · 1950
Selected original published ten-page scan/PDF passages inspected through University of Alabama archive; indexed bibliographic record inspected · Hospital interview sampling, 649 male cases/649 controls, smoking association and alternatives
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Does prospective observation remove all confounding?
Selected-source overview · Independent review pending. Read the evidence context.
No. Measuring smoking before subsequent mortality clarifies sequence, but people still choose exposures and can differ for other reasons. The historical doctors’ cohort also requires care when applying its results to other populations and periods.[8]
Evidence behind this answer
- British Doctors Study
original investigator programme summary · Date not stated
Original investigator programme’s indexed substantive summary and 1954 report bibliographic date inspected; direct programme HTML retrieval failed · Prospective 1951 start, smoking and cause-specific mortality, follow-up programme
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Does good sensitivity prove that screening improves health?
Selected-source overview · Independent review pending. Read the evidence context.
No. Accuracy relative to a reference standard and the benefits or harms of offering a testing pathway are different questions. Participant spectrum, thresholds, verification and the subsequent care pathway all affect interpretation.[3]
Evidence behind this answer
- STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies
original diagnostic accuracy reporting guideline · 2015
Selected original published PDF passages inspected through Queen’s University Belfast repository; original publisher indexed passages and date checked · Accuracy/reference definitions and checklist on recruitment, thresholds, verification and intended use
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Is a useful predictor necessarily a cause to change?
Selected-source overview · Independent review pending. Read the evidence context.
No. Prediction and causal intervention effects have different targets. A model can predict from a marker that is not a manipulable cause. Separate-data performance, calibration and intended clinical use still need evaluation.[4]
Evidence behind this answer
- TRIPOD+AI statement: updated reporting guidance for clinical prediction models
original prediction reporting guideline · 2024
Original publisher indexed substantive passages and official expanded-checklist PDF inspected; publisher full HTML retrieval blocked · Model development/evaluation, intended population/time horizon, discrimination/calibration and expanded checklist
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Is every natural experiment equivalent to a randomized trial?
Selected-source overview · Independent review pending. Read the evidence context.
No. The 2025 framework includes a broader range of changes outside evaluator control. Assignment, comparison trends, concurrent changes and analytic assumptions need examination; the label alone does not eliminate confounding.[5]
Evidence behind this answer
- Using natural experiments to evaluate population health and health system interventions: new framework
original current evaluation framework · 2025
Selected original publisher and PMC indexed substantive passages, plus original NIHR framework chapter indexed explanation inspected; direct full-text retrieval blocked · Broad definition, assignment/context, counterfactual approaches and intended/unintended effects
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Should every question be answered by an RCT?
Selected-source overview · Independent review pending. Read the evidence context.
No. The research question may concern experience, prevalence, mechanisms, implementation or exposures that cannot ethically be assigned. The complex-intervention framework supports choosing methods around the important uncertainty and context rather than a universal design preference.[6]
Evidence behind this answer
- A new framework for developing and evaluating complex interventions: update of MRC guidance
original complex intervention framework · 2021
Selected original PubMed and PMC indexed substantive methods passages inspected; direct publisher retrieval blocked · Context, programme theory, stakeholder input, uncertainty, refinement, economics and method choice
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Why do observational studies and trials sometimes seem to disagree?
Selected-source overview · Independent review pending. Read the evidence context.
They may involve different populations, interventions, comparators, outcomes or follow-up. Observation is also vulnerable to confounding and reverse causation, while trials have their own design and conduct limitations. Compare the actual questions and methods; observational studies do not invariably overestimate effects, and randomized does not mean flawless.[11][12][13]
Evidence behind this answer
- A comparison of observational studies and randomized, controlled trials
comparative methodological study · 2000
Indexed abstract; direct retrieval attempted · Abstract: comparison and conclusions - Bias and causal associations in observational research
methodological review · 2002
Indexed original abstract; direct PubMed retrieval returned incomplete body · Confounding; selection and information bias; temporal ordering - Clustered environments and randomized genes: a fundamental distinction between conventional and genetic epidemiology
methodological paper · 2007
Indexed original abstract; direct PubMed retrieval returned no body · Background: lifestyle confounding and reverse causation
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Does an adverse-event report prove a medicine caused the event?
Selected-source overview · Independent review pending. Read the evidence context.
No. It records an event after exposure or a suspected association. FDA distinguishes an adverse event from an adverse drug reaction caused by treatment. Reports can help identify signals, but causality assessment requires more context and may draw on several data sources.[10]
Evidence behind this answer
- Understanding CDER’s Postmarket Safety Surveillance Programs and Public Data
agency explanation · Date not stated
Selected official page passages · Adverse event versus reaction; report limitations; multiple evidence streams
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Does research saying linked to prove that one thing causes the other?
Selected-source overview · Independent review pending. Read the evidence context.
No. “Linked to” describes an association, not proof of cause. Another factor may influence both things (confounding), or illness may change the exposure being studied (reverse causation). Causal interpretation requires examining timing, study design and alternative explanations.[13][14]
Evidence behind this answer
- Clustered environments and randomized genes: a fundamental distinction between conventional and genetic epidemiology
methodological paper · 2007
Indexed original abstract; direct PubMed retrieval returned no body · Background: lifestyle confounding and reverse causation - Finding and Using Health Statistics — Causation
agency methods explanation · Date not stated
Selected official tutorial passages · Causation; Guidelines for Assessing Causation, lines 71–90
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Can interviews explain why a treatment is difficult to use?
Selected-source overview · Independent review pending. Read the evidence context.
They can investigate experiences and context that a symptom average misses. Readers still need to know whose accounts were included and how the analysis produced its interpretations.[15]
Evidence behind this answer
- Standards for Reporting Qualitative Research: A Synthesis of Recommendations
Original standards-development methods study · 2014
Original seven-page article publicly hosted by Stanford retrieved; complete abstract and selected introduction, methods/results discussion and Table1 passages read. · Original PDF introduction; research design/methods; Table 1 sampling, reflexivity, analysis and transferability
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
Does adding a few interviews automatically make a study a useful mixed-methods study?
Selected-source overview · Independent review pending. Read the evidence context.
The methods need a clear connection to the research question and to each other. Explain how one informs sampling, measurement, analysis or interpretation of the other.[16]
Evidence behind this answer
- Achieving Integration in Mixed Methods Designs—Principles and Practices
Original methodological framework with illustrative examples · 2013
Original full text retrieved after native challenge; complete abstract and selected integration/design/fit passages read. · Abstract; research questions; sequential/convergent designs; fit of integration; disclosures
This answer summarizes selected sources. An independent appraisal of the complete evidence has not been completed.
A historical case-control study compared smoking histories
Doll and Hill’s 1950 investigation interviewed hospital patients with lung carcinoma and comparison patients. Its main male comparison included 649 patients in each group. Smoking, particularly heavier smoking, was much more frequent among cases. This design investigated earlier exposure among selected outcome groups; it did not randomly assign smoking or directly measure a population incidence rate. The authors examined possible biases and competing explanations, including hospital selection and how smoking histories were obtained. The finding was consequential, but its interpretation required more than the number of cases.[7]
Prospective follow-up asked a related question differently
The Oxford investigators describe the British Doctors Study as a prospective programme beginning in 1951. Smoking information was obtained before subsequent cause-specific mortality follow-up. Early reports demonstrated the prospective relationship with lung-cancer mortality; later follow-up addressed other causes and long-term outcomes. Observing exposure before death reduces one temporal ambiguity found in a snapshot. It does not turn self-selected smoking into random assignment. Doctors in a historical British cohort also do not represent every population, period or product. The two smoking investigations supplied different evidence rather than two names for the same design.[8]
Test accuracy requires a target condition and a reference
STARD 2015 asks for the intended test use, participant selection, index and reference methods, thresholds, blinding and participant flow. Sensitivity and specificity summarize classification relative to the reference standard in the studied setting. An imperfect reference, different disease severity, selective recruitment or missing verification can affect estimates. A highly accurate test can still lead to unnecessary procedures or fail to improve outcomes if the subsequent care pathway is ineffective. Thus a diagnostic accuracy study answers a different question from a trial comparing the benefits and harms of offering that test.[3]
Validation must concern data beyond model fitting
TRIPOD+AI covers regression and machine-learning prediction models. Development, tuning, internal evaluation and evaluation in a separate dataset should be described distinctly. Discrimination concerns ranking people with different outcomes; calibration concerns agreement between predicted probabilities and observed outcomes. Performance in development data may be optimistic, and performance can change across settings or groups. A predictor need not be a manipulable cause. These reporting requirements help readers assess the model’s intended role without assuming that a good prediction guarantees benefit from acting on it.[4]
A policy change can create an evaluation opportunity
A natural experiment concerns the change being studied; a quasi-experimental design describes how an evaluator constructs a nonrandomized comparison. The 2025 MRC/NIHR framework uses a broad definition rather than requiring every assignment to be effectively random. Evaluators should investigate why the change occurred, who was exposed, credible comparison groups, timing, context and possible concurrent changes. Interrupted time series, difference-in-differences and other approaches require their own assumptions. Before-and-after improvement alone may reflect an existing trend. The new framework also considers implementation and unintended effects, so a policy’s average outcome is not the entire evaluation.[5]
Match methods to the uncertainty that matters
The 2021 MRC framework treats development, feasibility, evaluation and implementation as connected activities. Programme theory explains how an intervention is expected to work and under which circumstances. Stakeholder perspectives, context, refinement and economic consequences shape the research question. Randomization can strengthen a comparative effectiveness question, while qualitative or process work can explain delivery, experience and mechanisms. Some harmful exposures cannot ethically be assigned. A study should therefore be judged by whether its methods address the actual question, rather than by one universal ranking of every design.[6]
Ask how the interpretation was developed
The Standards for Reporting Qualitative Research ask authors to explain selection, setting, data collection, analysis and supporting material. Reflexivity means examining how researchers' backgrounds, assumptions and relationships may influence the work. An informative report makes those influences visible and links interpretations to evidence. A reporting checklist does not by itself prove the interpretation is sound or transferable to every setting.[15]
Different findings can deepen the question
Fetters and colleagues describe findings that confirm each other, expand understanding of different aspects, or conflict. Disagreement should prompt examination of methods, assumptions and context. Interviews about treatment burden and an average symptom score can answer different questions; neither should silently replace the other.[16]
References
- An overview of clinical research: the lay of the land
Original methods primer
Grimes and Schulz; The Lancet · Content/publication date: 2002
Accessed: 2026-10-14 UTC · Selected original published PDF passages inspected through Northwell academic copy; PubMed indexed record inspected
Cited sections: Taxonomy and comparison/time-direction definitionsHistorical stable design definitions; old evidence hierarchy not adopted as current universal guidance. Academic PDF is the same paper, not a separate allowance.
- STROBE: Explanation and Elaboration
Original reporting guideline
Vandenbroucke and colleagues; PLOS Medicine · Content/publication date: 2007
Accessed: 2026-10-14 UTC · Selected original publisher HTML passages inspected
Cited sections: Three observational designs; early design/date/selection reporting; reporting not quality certificateExplanation/elaboration and shorter statement are an integrated reporting family; journal/PMC mirrors do not add allowances. No independent epidemiologic appraisal or formal grade claimed.
- STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies
Original diagnostic accuracy reporting guideline
Bossuyt and colleagues; The BMJ · Content/publication date: 2015
Accessed: 2026-10-14 UTC · Selected original published PDF passages inspected through Queen’s University Belfast repository; original publisher indexed passages and date checked
Cited sections: Accuracy/reference definitions and checklist on recruitment, thresholds, verification and intended useCurrent core STARD 2015 listing checked at EQUATOR; specialized extensions do not replace every core use. Repository PDF and multijournal mirrors remain this same statement.
- TRIPOD+AI statement: updated reporting guidance for clinical prediction models
Original prediction reporting guideline
Collins and colleagues; The BMJ · Content/publication date: 2024
Accessed: 2026-10-14 UTC · Original publisher indexed substantive passages and official expanded-checklist PDF inspected; publisher full HTML retrieval blocked
Cited sections: Model development/evaluation, intended population/time horizon, discrimination/calibration and expanded checklistOfficial supplement belongs to this statement and shares its allowance. BMJ q902 representation and PMC mirrors not treated as independent publications. Prediction accuracy is not a causal treatment claim.
- Using natural experiments to evaluate population health and health system interventions: new framework
Original current evaluation framework
Craig and colleagues; MRC/NIHR; The BMJ · Content/publication date: 2025
Accessed: 2026-10-14 UTC · Selected original publisher and PMC indexed substantive passages, plus original NIHR framework chapter indexed explanation inspected; direct full-text retrieval blocked
Cited sections: Broad definition, assignment/context, counterfactual approaches and intended/unintended effectsNew 2025 framework replaces use of 2012 guidance as current recommendation. BMJ report and NIHR detailed framework are one guidance family for conservative attribution, not extra variants.
- A new framework for developing and evaluating complex interventions: update of MRC guidance
Original complex intervention framework
Skivington and colleagues; MRC/NIHR; The BMJ · Content/publication date: 2021
Accessed: 2026-10-14 UTC · Selected original PubMed and PMC indexed substantive methods passages inspected; direct publisher retrieval blocked
Cited sections: Context, programme theory, stakeholder input, uncertainty, refinement, economics and method choice2021 framework is a genuinely distinct publication from 2025 natural-experiment framework. Later reprint not counted as new evidence or new allowance.
- Smoking and carcinoma of the lung: preliminary report
Original historical case control study
Doll and Hill; British Medical Journal · Content/publication date: 1950
Accessed: 2026-10-14 UTC · Selected original published ten-page scan/PDF passages inspected through University of Alabama archive; indexed bibliographic record inspected
Cited sections: Hospital interview sampling, 649 male cases/649 controls, smoking association and alternativesOriginal DOI 10.1136/bmj.2.4682.739; PMID 14772469; PMC 2038856. Reprints and archive copies remain one report. No population incidence or experimental effect calculated from case sampling.
- British Doctors Study
Original investigator programme summary
Oxford Clinical Trial Service Unit and Epidemiological Studies Unit · Content/publication date: not stated
Accessed: 2026-10-14 UTC · Original investigator programme’s indexed substantive summary and 1954 report bibliographic date inspected; direct programme HTML retrieval failed
Cited sections: Prospective 1951 start, smoking and cause-specific mortality, follow-up programmeSource page publication/update date unverified. Historical 1954 report date is not assigned as this programme webpage’s publication. Original 1954 paper and later follow-ups not presented as separately appraised studies or extra allowances.
- Understanding Clinical Studies
Agency explanation
NIH · Content/publication date: not stated
Accessed: 2026-10-15 UTC · Selected official page passages
Cited sections: Randomized and observational designs; strengths and weaknesses - Understanding CDER’s Postmarket Safety Surveillance Programs and Public Data
Agency explanation
FDA · Content/publication date: not stated
Accessed: 2026-10-15 UTC · Selected official page passages
Cited sections: Adverse event versus reaction; report limitations; multiple evidence streams - A comparison of observational studies and randomized, controlled trials
Comparative methodological study
Benson and Hartz · Content/publication date: 2000
Accessed: 2026-10-15 UTC · Indexed abstract; direct retrieval attempted
Cited sections: Abstract: comparison and conclusionsHistorical selected comparison; not a universal guarantee of design agreement.
- Bias and causal associations in observational research
Methodological review
Grimes and Schulz · Content/publication date: 2002
Accessed: 2026-10-15 UTC · Indexed original abstract; direct PubMed retrieval returned incomplete body
Cited sections: Confounding; selection and information bias; temporal orderingOriginal methods primer; not a clinical effect estimate.
- Clustered environments and randomized genes: a fundamental distinction between conventional and genetic epidemiology
Methodological paper
Davey Smith and colleagues · Content/publication date: 2007
Accessed: 2026-10-15 UTC · Indexed original abstract; direct PubMed retrieval returned no body
Cited sections: Background: lifestyle confounding and reverse causationOnly the methodological distinction is used; no genetic causal claim transferred.
- Finding and Using Health Statistics — Causation
Agency methods explanation
National Library of Medicine · Content/publication date: not stated
Accessed: 2026-10-15 UTC · Selected official tutorial passages
Cited sections: Causation; Guidelines for Assessing Causation, lines 71–90Study-design and temporal-order explanation; no clinical effect estimate.
- Standards for Reporting Qualitative Research: A Synthesis of Recommendations
Original standards-development methods study
O'Brien, Harris, Beckman, Reed and Cook · Content/publication date: 2014
Accessed: 2026-10-20 UTC · Original seven-page article publicly hosted by Stanford retrieved; complete abstract and selected introduction, methods/results discussion and Table1 passages read.
Cited sections: Original PDF introduction; research design/methods; Table 1 sampling, reflexivity, analysis and transferabilityTransparent reporting supports appraisal; it is not an automatic quality score or one mandatory qualitative method.
- Achieving Integration in Mixed Methods Designs—Principles and Practices
Original methodological framework with illustrative examples
Fetters, Curry and Creswell · Content/publication date: 2013
Accessed: 2026-10-20 UTC · Original full text retrieved after native challenge; complete abstract and selected integration/design/fit passages read.
Cited sections: Abstract; research questions; sequential/convergent designs; fit of integration; disclosuresMethodological framework with examples, not a trial proving mixed methods always improve decisions. No disclosures reported.
Scope & review
- Population
- Readers interpreting health-study claims rather than choosing personal treatment
- Exposure or intervention
- Descriptive/analytic observational, experimental, natural-experiment, diagnostic and prediction study purposes
- Comparison
- Internal control groups, reference standards or validation data depending on the question
- Outcomes
- Prevalence/incidence, exposure associations, accuracy, prediction and intervention outcomes distinguished
- Timeframe
- Historical smoking examples and source-specific current reporting/evaluation frameworks; no universal follow-up period
- Jurisdiction
- International methodological education; UK MRC/NIHR frameworks not global law; no personal test or treatment schedule
Selected-source educational explanation. Official guidance, original study findings and editorial interpretation are distinguished. No exhaustive search, independent systematic synthesis or qualified medical review has been completed.
Sources checked on their individual access dates. Selected sources were checked; ongoing surveillance and qualified review have separate statuses on the coverage page.
Revision history
- 2026-10-14 UTC: Added a bounded study and causal methods foundation guide; qualified clinical and research-methods review remains pending.
- 2026-10-15 UTC: Added the plain-English series explanation with definitions, distinctions, practical questions and selected primary sources; qualified review remains pending.
- 2026-10-20 UTC: Added supported foundational, practical and evidence explanations after reading full target bodies and relevant current/pending adjacent coverage. Original sections, questions and sources preserved as exact prefixes; qualified review pending.
What this article still leaves open
- Which design and comparison actually produced this claim?
- Does the inference concern prediction, description or changing an outcome?
- What alternatives and setting differences remain despite a transparent report?
Suggest a correction
Submit a content correction to the editorial queue, or download a local note. Leave out personal medical details. Submitted text is stored privately with a browser receipt session and is visible to this browser and the editor. Clearing cookies removes access to your submission history. This is not an account or a medical service.
Your submitted corrections for this article
More in the plain-English series
Keep making sense of it.
Explore Prevention & care explanations See every explanationExplore the topic map
Browse the health domains connected to this article. The map also identifies topics whose articles remain unwritten.