Association or Causation


后来想搜一下有没有比较通用的逻辑模型时,发现了 Sir Austin Bradford Hill的演讲 The Environment and Disease: Association or Causation? 其中提到了在流行病学调查中,如何证实观察到的相关现象存在因果关系,即 nine “viewpoints” 或者叫做 Bradford Hill criteria:

  1. Strength of Association: A strong association between a factor and a disease is more likely to be causal. For example, the strong relationship between smoking and lung cancer supports a causal link.

  2. Consistency: An association is more likely to be causal if it is consistently observed in different studies, populations, and circumstances. Replication of findings across various settings adds credibility to the causal hypothesis.

  3. Specificity: Causality is suggested if a specific exposure is associated with a specific disease. However, this criterion is less applicable in complex situations where a single cause can lead to multiple effects.

  4. Temporality: For an association to be causal, the cause must precede the effect. This is a critical criterion as it establishes the chronological sequence of events.

  5. Biological Gradient (Dose-Response Relationship): A causal relationship is supported if an increasing amount of exposure leads to an increasing risk of the disease. For example, higher levels of radiation exposure leading to higher rates of cancer.

  6. Plausibility: The association should be biologically plausible based on existing knowledge. This means that there should be a logical and biological mechanism that explains how the exposure could lead to the disease.

  7. Coherence: The association should be coherent with existing knowledge and theory. It should not seriously conflict with the known history and biology of the disease.

  8. Experiment: Causality is more likely if an experiment (such as a randomized controlled trial) can demonstrate that changing the exposure leads to a change in the outcome.

  9. Analogy: Sometimes, if a similar exposure-disease relationship has been established for another situation, it can lend support to a causal relationship in the current situation.

Researchers use these criteria as a guideline rather than a checklist. Not all criteria need to be met to establish causality, and some are more critical than others. Temporality is often considered essential, while specificity may be less relevant in multifactorial diseases. These criteria help to systematically evaluate the evidence and strengthen the argument for or against a causal relationship in epidemiologic research.

因为在逻辑上具有某些共通性,所以在基础实验中研究特定条件下,分子A是不是通过某个通路对事件B进行了调控,可以使用其中的一些原理,如Strength of Association, Consistency, Specificity, Temporality, Biological Gradient。需要用到的试验手段有比如:

  1. Perturbation experiments:
    • Inhibit or activate key components of the proposed pathway.
    • Observe the impact on the phenomenon.
    • Use techniques like genetic knockout, RNA interference, or pharmacological inhibitors.
  2. Gain-of-function experiments:
    • Overexpress or artificially activate pathway components.
    • Examine whether this enhances the phenomenon.
  3. Loss-of-function experiments:
    • Inhibit or remove pathway components.
    • Observe if the phenomenon is diminished or abolished.
  4. Rescue experiments:
    • Restore the function of a disrupted pathway component.
    • Determine if the phenomenon is rescued.


转载请注明来源:Association or Causation