←
Stephen Burgess2,833 posts
Stephen Burgess avatar

Stephen Burgess

@stevesphd

Medical statistician, work with genetic data to disentangle causation from correlation. Author of book on Mendelian randomization.

◷ Joined May 2010▧ 155 Photos and videos
Stephen Burgess avatar
Stephen Burgess@stevesphd·26 Nov 2019

Guidelines on performing Mendelian randomization investigations written by an all-star line-up of MR researchers are now available on Wellcome Open Research: wellcomeopenresearch.org/art… - represents a consensus statement after 12+ months of deliberation. Comments welcome!

Stephen Burgess avatar
Stephen Burgess@stevesphd·Nov 4

PS - I've largely stopped posting on Twitter / Bluesky now, and am using LinkedIn instead. I will post important new papers here, but most of my activity will be on LinkedIn.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Nov 4

Many caveats - this was a small sample and not a representative one. Findings surprised us, but equally what patients said and how we reported it will be coloured by our own beliefs and prejudices. But there is an important and consequential message here that is worth exploring.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Nov 4

Conversely, we should not say "X is associated with Y" - patients understood this as a weak causal claim, not a correlative claim. If we mean "individuals with higher levels of X have greater risk of Y", that is what we should say.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Nov 4

Our recommendation from this work is that epidemiologists should avoid trite phrasings in public-facing documents. We should not say "X causes Y" or even "X is a cause of Y". If we mean "increasing your level of X will raise the risk of Y", that is what we should say.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Nov 4

This differs from how we talk about causation in epidemiology, where our primary axis of concern is the degree to which a causal claim is evidenced. Patients did not have a good understanding of correlation - they knew it wasn't causation, but struggled to define what it meant.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Nov 4

They did not interpret the phrasing "X causes Y" to mean "increasing your level of X will raise the risk of Y". They interpreted it as meaning "if X, then Y" - they were happy to accept edge cases where Y does not follow X, but they wanted these to be rare.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Nov 4

Patients interpreted the strength of causal language not in terms of the quality of evidence to separate correlation from causation, but the certainty of the causal link: to what extent will this affect them.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Nov 4

When using phrasings such as "is a cause of", "leads to", "predicts", or "is linked to", the primary axes that patients distinguished these phrasings were: 1) extent to which a mechanistic explanation was known, and 2) degree of certainty that if X happens, that Y will follow.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Nov 4

What do we mean as epidemiologists when we say: "X causes Y"? And what do patients understand by: "turnip consumption is linked with higher risk of bowel cancer" or "swede consumption predicts higher risk of liver disease"? We asked a group of patients and found out.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Nov 4

New paper at AJE Advances: "From Claim to Brain: A Public and Patient Involvement and Engagement Project on how Causal Language is interpreted in Healthcare Studies" led by Emily Bassett with co-authorship from @amymariemason: academic.oup.com/ajeadvances…. Thread follows!

Stephen Burgess avatar
Stephen Burgess@stevesphd·Aug 13

In doing this, we may get a more valid estimate of a less relevant quantity, rather than a less valid estimate of a more relevant quantity. In many cases, the estimand is a good servant, but a poor master.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Aug 13

We encourage a broader framing of target trial emulation: alongside “estimand first” approaches, we should consider “design first” or “assumption first” approaches - where we start by asking "What quantity can we estimate reliably?".

Stephen Burgess avatar
Stephen Burgess@stevesphd·Aug 13

A natural experiment gives an estimate - but the estimate is often a means to an end, not the end in itself. It provides supportive evidence, rather than answering the question of interest. But we would often rather more reliable estimate, even if the estimand is less relevant.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Aug 13

We argue that epidemiological studies can be characterized into early-phase and late-phase in the same way. A natural experiment may give an estimate that is not policy-relevant or translatable to applied practice. But it is still a relevant source of causal information.

Stephen Burgess avatar
Stephen Burgess@stevesphd·Aug 13

In drug development, the goal of a Phase 3 study is to estimate the impact of a well-defined intervention. Whereas the goal of a Phase 2 study is to provide proof-of-concept evidence supporting (or refuting) a given mechanism as a plausible target for intervention.