BiostatisticsTier 1Medical Sciences concept

Bradford Hill criteria for causation

Core concept

A framework of considerations (not a checklist requiring all to be met) for judging whether an observed association is likely causal rather than confounded/coincidental
Originally proposed for smoking and lung cancer -- still the standard framework for causal inference from observational data
No single criterion is necessary or sufficient -- the exam tests recognising which criterion a described scenario illustrates

Criterion	Meaning
Strength	Larger effect size (e.g. OR/RR far from 1) more likely causal
Consistency	Repeated finding across different populations/studies/methods
Specificity	Exposure linked to one specific outcome, not many (weakest/least essential criterion -- many causes have multiple effects)
Temporality	Exposure must precede outcome -- the only criterion considered essential
Biological gradient	Dose-response relationship
Plausibility	Consistent with known biological mechanism
Coherence	Fits with existing disease knowledge, doesn't contradict it
Experiment	Removing exposure reduces incidence (e.g. smoking cessation trials)
Analogy	Similar exposures cause similar effects

Temporality is the only truly essential criterion -- a cross-sectional study cannot satisfy it, which is why cross-sectional designs cannot establish causation regardless of association strength
Trap: assuming "biological plausibility" alone proves causation -- it is supportive, not confirmatory, and plausible mechanisms have been wrong before
When a vignette describes a dose-response relationship (higher exposure, higher/more severe outcome), that is testing recognition of biological gradient, specifically

Directly informs how to interpret observational studies (cohort, case-control) once an association is found -- these criteria are the next analytic step
Confounding, reverse causation and chance are the main alternative explanations that Bradford Hill criteria are used to argue against
Randomised trials sidestep most of these criteria by design (removes confounding, establishes temporality directly) -- one reason RCT evidence outranks observational evidence in the hierarchy

3 more sections, plus exam facts

Premium unlocks every note across every specialty, and the full exam fact library behind it.