Last week, there was a very interesting, and very wet, conference on Causation and Complexity in Sydney. This was the 10th Munich-Sydney-Tilburg Conference in the Philosophy of Science (MuST10). Some of the talks related to EBM+ and RWT. Before I introduce these talks, let’s review the difference between EBM+ and RWT:
- EBM+ holds that, because evidence of mechanisms is used – often implicitly – to support or undermine interventions in medicine and public health, this evidence should be made explicit so that it can be exposed to scrutiny and criticism. Currently, evidence-based medicine (EBM) focuses on evidence that arises from clinical studies. The EBM+ programme seeks to extend the range of evidence under scrutiny to include evidence of mechanisms originating from sources other than clinical studies, such as evidence arising from in vitro experiments or from biomedical imaging. This goal should be uncontroversial – it is a natural extension of the EBM strategy of making evidence and its evaluation explicit.
- RWT (the Russo-Williamson thesis) is more controversial. This says that, in order to establish a causal claim in the health sciences, one needs to establish both that the putative cause and effect are appropriately correlated and that they are appropriately connected by some mechanism or structure of mechanisms. The motivation here is that some correlations are due to bias, confounding or other non-causal considerations; in a genuine causal relationship there is a mechanistic connection in addition to a correlation. RWT has been the object of some criticism in the literature, not least because it conflicts with EBM, which tends to view evidence of correlation produced by clinical studies as strictly superior to evidence of mechanisms produced by other means. This draft paper explains the conflict and offers a defence of RWT.
In sum, EBM+ and RWT are distinct theses. While RWT makes EBM+ particularly pressing, one can endorse EBM+ even if one doesn’t accept RWT. EBM+ is consistent with contemporary EBM, but RWT is not.
Back to MUST10. On the first day of the conference, there was a session on causation and evidence in medicine, with four talks.
First, I gave a talk on Models in systems medicine. Systems medicine applies data-intensive functional genomics techniques – e.g., transcriptomics, metabolomics and proteomics – to medicine. Systems medicine appeals to many different kinds of model, and the relationships between these models can be rather confusing. I argued that one can structure the development of models in systems medicine in line with EBM+ and RWT, by using mechanistic models to inform the development of causal models. This talk was based on a paper forthcoming in Disputatio.
Next up, Sydney Katherine Green (Antwerp) gave a nice talk on The Impossibility of Causal Claims in Psychiatry. She argued that if RWT is true, then no causal claims in psychiatry have been established. Some members of the audience seems to be broadly skeptical about psychiatry and willing to accept the conclusion. However, it might be possible to resist the conclusion. Green required that evidence of mechanisms be independent of evidence of correlation. Perhaps this is too strong a requirement: it seems in principle possible that the same item of evidence could provide evidence both of correlation and mechanism. Arguably, RWT makes no such requirement, in which case some causal claims in psychiatry may be deemed to be established after all.
Roland Poellinger (Munich) gave a talk on Similarity and Analogy in Pharmacology. He presented a formal Bayesian framework for taking various kinds of evidence into account when evaluating a causal claim in pharmacology, and looked in detail at some kinds of analogical evidence.
Barbara Osimani (Munich) spoke about Reliability, robustness and the reproducibility problem. The talk concerned the problem of replicability and showed how the formal Bayesian framework mentioned above can take various sorts of reliability of evidence into account.
This formal framework seems very much in line with EBM+, because it includes evidence of mechanisms as one of the kinds of evidence to be taken into account. On the other hand, this formal framework does not implement RWT, because it admits the possibility that causality can be established by establishing only correlation, or by establishing only mechanism.
Incidentally, the Sydney Opera House is a good example of the use of concrete reinforced with steel to build structures that could not be built with either concrete or steel on their own. Analogously, EBM+ views evidence of correlation and evidence of mechanisms as mutually reinforcing. See my piece on Penguins and Causality. (The engineers who worked on the Sydney Opera House also worked on the penguin pool at London Zoo.)