The Bradford Hill Causality Screen
Nine criteria that separate a real signal from epidemiologic nonsense.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 7
- Confidence
- 85%
Attia leans on the criteria developed by the mid-twentieth-century scientist Austin Bradford Hill — nine factors used to scrutinize observational data and adjust confidence that a relationship is causal rather than merely correlated. Among them he names the strength or magnitude of the observation, its reproducibility, its biological plausibility, whether an analogy exists, and whether there is a dose effect. He uses the screen in both directions. A study claiming twelve hazelnuts a day cuts death risk by 10 percent fails it and gets called epidemiologic nonsense. The finding that very strong people outlive very weak people passes it — strong association, consistent across studies, dose-responsive, biologically plausible — which is why he treats strength as likely causal rather than a mere marker of health.
Origin
The nine criteria are owed to Austin Bradford Hill, a scientist of the middle part of the twentieth century. Attia says he walks through them a couple of times in Outlive because so much nutrition and exercise data is unavoidably epidemiological.
Core principles
- 01Correlation in observational data is the starting point, not the conclusion.
- 02Most nutrition and exercise data will never come from randomized trials.
- 03Confidence in causality is a dial, not a switch.
- 04Magnitude, consistency, plausibility, analogy and dose response all move the dial.
- 05A confounder like fair skin behind red hair explains most spurious associations.
How to run it
- 1
Classify the study design
Establish whether there was an intervention. An observational study, retrospective or prospective, follows groups without intervening, so it cannot on its own establish cause.
- 2
Measure the magnitude
Ask how large the actual observed effect is. A small effect in observational data is the easiest thing in the world to manufacture from confounding.
- 3
Check reproducibility and consistency
Does the same association appear repeatedly across studies and populations? Attia notes grip strength comes up over and over again as a longevity proxy, which is part of why he trusts it.
- 4
Demand biological plausibility and analogy
Is there a mechanism that would explain the relationship, and is there an analogous accepted relationship elsewhere in biology?
Watch out Plausibility alone is weak; it is one dial among nine, not the verdict.
- 5
Look for dose response
If more exposure produces more effect in a graded way, causality becomes far more credible. Attia cites a monotonic decline in dementia onset and dementia death as grip strength increases.
- 6
Name the confounder
Before accepting a causal story, state the most likely alternative explanation. Attia's teaching case is red hair and melanoma, where fair skin is the actual driver.
Pro tip If you cannot name a plausible confounder, look harder before concluding there isn't one.
- 7
Convert relative risk to absolute risk
A headline that a food doubles your cancer risk may mean one in five billion becoming two in five billion. Ferriss's point is that behaviourally that difference does not matter at all.
In the wild
Attia's invented illustration: observe a group of redheads and a group of brunettes for ten years and find the redheads get far more melanoma. The tempting conclusion is that red hair causes skin cancer and brown hair reflects the sun. The real answer is that red hair is associated with fairer skin, and skin is what drives the risk. To actually test causality you would have to randomize people and dye their hair. The example makes the confounder visible in a case where nobody is emotionally invested in the answer.
→ A clean mental template for spotting the confounder behind an association.
Attia contrasts two epidemiologic findings. A study claiming twelve hazelnuts a day reduces death risk by 10 percent he dismisses immediately as epidemiologic nonsense — implausible mechanism, implausible magnitude for the exposure. The finding that very strong people live longer than very weak people passes the same screen: large association, repeatedly consistent, dose-responsive, mechanistically plausible. Same study design, opposite verdicts, because the criteria and not the design decide.
→ The screen lets one method of evidence produce both a rejection and a confident acceptance.
Common mistakes
Scientism: only RCTs count
Ferriss describes well-educated non-scientists who insist anything outside a randomized controlled trial is unsubstantiated nonsense. Many of the relevant trials will never be run.
Treating a marker as a lever
Attia is careful to distinguish strength being a marker of health from training strength improving outcomes. The criteria are what let him argue for the latter.
Reacting to relative risk headlines
A 100 percent increase on a negligible base rate is still negligible, but the headline reads identically to a meaningful one.
Is it for you?
Best for
Anyone making health or lifestyle decisions from published research and media coverage of it.
Not ideal for
Settings where a well-powered randomized trial already exists and directly answers the question.
From the transcript
“what Bradford Hills criteria allow us to do is look at nine factors such as the strength of the observation so what's the actual magnitude…”
“when I look at epidemiologic data that say people who are really really strong live longer than people who are really really weak by going…”
From the episode
#661: Dr. Peter Attia — The Science and Art of Longevity, Optimizing Protein, Alcohol Rules, Lessons from Glucose Monitoring with CGMs, Boosting Your VO2 Max, Preventing Alzheimer's Disease, Early Cancer Detection, How to Use DEXA Scans, Nature’s Longevity Drug, and More