How Do We Know if Something Actually Works? A Guide to the Bradford Hill Criteria

By Dr. Steven Long, DO, MHA, CPT
Beyond Health | Precision Medicine for High-Performance Living

Modern medicine is flooded with headlines: “Red wine lowers heart disease risk.” “Artificial sweeteners cause cancer.” “Intermittent fasting reverses aging.”

But how do we know what’s causal and what’s correlated?

In 1965, British epidemiologist Sir Austin Bradford Hill proposed a framework to help scientists and readers distinguish between association and causation — a critical skill for interpreting both large population studies and smaller mechanistic trials.

Nearly 60 years later, the Bradford Hill criteria remain one of the most powerful tools for assessing whether an observed relationship is truly causal — not just statistical noise, bias, or coincidence.

At Beyond Health, we use this same framework when evaluating research — from cardiovascular medicine to nutrition, hormone therapy, and exercise physiology — to decide which findings are strong enough to change how we practice.

1. A Brief History: The Origins of Bradford Hill’s Framework

Sir Austin Bradford Hill was a physician and statistician best known for his work linking cigarette smoking to lung cancer in the mid-20th century.
At that time, many scientists doubted that smoking caused cancer — because experimental proof (like randomized controlled trials) was impossible.

Hill proposed a set of nine criteria that could strengthen or weaken the argument for causation based on available evidence.
This framework became foundational in epidemiology, public health, and clinical research — helping transform observational data into actionable medicine.

Today, these principles help guide interpretation of everything from hormone replacement therapy to dietary interventions and longevity studies — especially when randomized trials are limited or impractical.

2. The Nine Bradford Hill Criteria — Explained with Examples

     1. Strength of Association

The stronger the association between exposure and outcome, the more likely it is causal.

• Example: Smokers have a >20-fold higher risk of lung cancer compared to non-smokers — a powerful association unlikely due to confounding.
• By contrast, a study showing a 10% difference in risk for a dietary factor (e.g., coffee consumption) may warrant skepticism.

Clinical application:
 A strong odds ratio or hazard ratio supports causation but doesn’t guarantee it.
 Small associations should be interpreted cautiously unless multiple other criteria are also met.

     2. Consistency

If multiple studies — in different populations, settings, and methodologies — find similar results, confidence increases.

• Example: The link between LDL cholesterol and atherosclerosis has been confirmed through epidemiologic data, genetic studies (Mendelian randomization), and clinical trials.
• Conversely, if results fluctuate widely across studies, the relationship is less convincing.

Application to smaller studies:
 Even small, well-controlled trials gain strength if their results align with larger observational data.

     3. Specificity

If a specific exposure leads to a specific outcome, it strengthens causal inference.

• Example: Asbestos exposure specifically causes mesothelioma, a rare cancer.
• However, modern science recognizes that many conditions (like heart disease) have multifactorial causes — so lack of perfect specificity doesn’t disprove causation.

Clinical takeaway:
 Specificity helps when present but is no longer considered essential; biology is rarely that simple.

     4. Temporality

Cause must precede effect.

• Example: Elevated fasting insulin must exist before the development of insulin resistance and metabolic syndrome — not the other way around.
• In contrast, if both variables are measured at the same time, causation can’t be inferred.

Practical note:
This is the only absolute criterion — time must flow in the right direction.

     5. Biological Gradient (Dose-Response Relationship)

Increasing exposure should increase (or decrease) the effect.

• Example: The more cigarettes smoked per day, the greater the lung cancer risk.
• Similarly, increasing LDL particle number (ApoB) leads to a proportionate increase in cardiovascular events (Ference et al., Eur Heart J, 2017).

For smaller studies:
 Finding a dose-response strengthens credibility even if the sample size is modest.

     6. Plausibility

The observed effect should make biological sense based on current understanding.

• Example: Estrogen therapy may protect arterial health when started near menopause because estrogen receptors remain active and endothelial repair is responsive — supporting the “timing hypothesis.”
• Conversely, implausible claims (e.g., a vitamin curing multiple unrelated diseases) deserve skepticism.

Clinician’s note:
 Biologic plausibility evolves — a finding once dismissed may gain support as mechanisms become clearer (as with gut microbiome research).

     7. Coherence

The association should not conflict with existing scientific knowledge about disease mechanisms.

• Example: The causal role of hypertension in stroke aligns with physiology, pathology, and experimental models.
• In contrast, a claim that contradicts well-established physics or biochemistry (e.g., “detox patches”) fails coherence.

Interpretation:
 A coherent theory fits both population data and known physiology.

     8. Experiment

Causation is more likely if removing or modifying the exposure changes the outcome.

• Example: Statins lower LDL cholesterol and reduce cardiovascular events — a clear experimental confirmation.
• Similarly, smoking cessation reduces lung cancer risk over time, supporting the causal model.

Application in small studies:
 Even short-term interventions showing measurable improvement (e.g., lower hs-CRP after exercise training) strengthen causal inference.

     9. Analogy

If a similar exposure causes a similar effect, causation becomes more plausible.

• Example: Multiple viral infections (e.g., HPV, hepatitis B, H. pylori) are known carcinogens; discovering another infection linked to cancer (e.g., EBV and lymphoma) is therefore credible.
• Likewise, new drugs in the same class as proven agents can be expected to have related mechanisms and effects.

Practical takeaway:
 Analogy helps bridge gaps in evidence — useful when direct data are limited.

3. Applying Bradford Hill to Modern Research

The Bradford Hill framework remains essential for interpreting both large-scale epidemiologic studies and smaller mechanistic or clinical trials.

For large observational studies:

• The criteria help distinguish true risk factors from spurious correlations (e.g., “red meat causes cancer” vs. confounded lifestyle factors).
• They guide which associations deserve experimental follow-up.

For small clinical or mechanistic trials:

• Even limited data gain weight if they demonstrate plausibility, temporality, and dose-response.
• Small studies showing consistent mechanisms (e.g., reduced inflammation, improved endothelial function) can meaningfully inform clinical hypotheses.

In practice:
 When evaluating any claim, ask:

1. Does the relationship make biologic sense?
2. Is it reproducible and dose-dependent?
3. Does changing the exposure change the outcome?

If the answer to several is “yes,” the evidence is likely causal, even without massive sample sizes.

4. Why This Matters for Clinicians and Patients

Modern medicine demands that we separate signal from noise.
 The Bradford Hill criteria help readers:

• Evaluate the quality of evidence, not just its size.
• Recognize where causality is likely, not just where correlation exists.
• Translate data into rational, safe, evidence-based care.

For patients, understanding this framework builds trust — it clarifies why some findings change practice while others fade with time.

At Beyond Health, we use these principles to assess every intervention — whether it’s hormone therapy, nutrition, or exercise physiology.
 Our standard is not “does it sound good,” but “does it meet Bradford Hill’s test for causation?”

5. Beyond Health’s Perspective

Precision medicine requires critical appraisal as much as cutting-edge technology.
 The Bradford Hill framework keeps us grounded — helping separate meaningful data from hype and ensuring our patients receive care that is both innovative and evidence-driven.

By applying these principles, readers can better integrate new science responsibly:

• Support therapies with demonstrated causality.
• Question those driven by correlation alone.
• Combine clinical experience with mechanistic logic to guide real-world outcomes.

At Beyond Health, this framework underlies every recommendation we make — from the supplements we prescribe to the biomarkers we monitor.

It’s how we ensure that modern longevity medicine stays rooted in scientific integrity.

Conclusion

Correlation is easy to find.
 Causation is hard to prove.
 The Bradford Hill criteria remain medicine’s compass for navigating this complexity — ensuring that we act on evidence, not assumption.

Whether evaluating a landmark cardiovascular trial or a new metabolic supplement, these principles remind us to ask the right questions — not just whether something works, but why it works, and whether the evidence truly supports that conclusion.

At Beyond Health, we believe that understanding causality isn’t just academic — it’s foundational to safe, effective, and ethical precision medicine.

References

1. Hill AB. The Environment and Disease: Association or Causation? Proc R Soc Med. 1965;58(5):295–300.
2. Ference BA, et al. Low-Density Lipoproteins Cause Atherosclerotic Cardiovascular Disease: 1. Evidence from Genetic, Epidemiologic, and Clinical Studies. Eur Heart J. 2017;38(32):2459–2472.
3. Vandenbroucke JP, et al. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): Explanation and Elaboration. PLoS Med. 2007;4(10):e297.
4. Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. 4th ed. Philadelphia: Lippincott Williams & Wilkins; 2021.
5. Hernán MA, Robins JM. Causal Inference: What If. Boca Raton: Chapman & Hall/CRC; 2020.