George Davey Smith

Does the (live) shingles vaccine protect against dementia? Fergus Hamilton and colleagues provide important evidence in a new preprint using hospital data from the whole of England in a regression discontinuity design from which we *don't* show evidence of a meaningful protective effect /cont

15/n a study reflect a choice. In summary, we don't see an effect on hospital coded dementia. we don't think this is because of our data, analytic method, or approach. This does not rule out a small effect (but the effect size would be smaller than that reported as point estimates previously) /cont

Does the (live) shingles vaccine protect against dementia? Fergus Hamilton and colleagues provide important evidence in a new preprint using hospital data from the whole of England in a regression discontinuity design from which we *don't* show evidence of a meaningful protective effect /cont

13/n We don't have an easy explanation for these differences. We discuss in the paper a lot more, but one explanation is - that RD designs - like all observational designs - have more analytic choices than RCTs, which means results are often more variable than an ITT RCT /cont

Does the (live) shingles vaccine protect against dementia? Fergus Hamilton and colleagues provide important evidence in a new preprint using hospital data from the whole of England in a regression discontinuity design from which we *don't* show evidence of a meaningful protective effect /cont

11/n Our result is not entirely consistent with the prior literature. In Wales, Eyting reported a 1.3% absolute risk reduction (95% CI 0.2-2.7), so although our confidence intervals overlap, our point estimate of 0.06 sits outside their CI (as does theirs to ours). However, given /cont

Does the (live) shingles vaccine protect against dementia? Fergus Hamilton and colleagues provide important evidence in a new preprint using hospital data from the whole of England in a regression discontinuity design from which we *don't* show evidence of a meaningful protective effect /cont

9/n We then tested whether their result would stand if in Wales they only had hospital data, if they had the same study entry as us, etc: essentially we tested many of the biases of HES. Essentially, these decisions have an effect on absolute effects but NOT relative effects /cont

Does the (live) shingles vaccine protect against dementia? Fergus Hamilton and colleagues provide important evidence in a new preprint using hospital data from the whole of England in a regression discontinuity design from which we *don't* show evidence of a meaningful protective effect /cont

7/n To test this, we limited our analyses to those who were plausibly vaccinated (e.g. only those 79+), and started follow up at 1 year post Sep 2013 (this would give ~6 months for the vaccine to work). This analysis was null /cont

Does the (live) shingles vaccine protect against dementia? Fergus Hamilton and colleagues provide important evidence in a new preprint using hospital data from the whole of England in a regression discontinuity design from which we *don't* show evidence of a meaningful protective effect /cont

5/n Importantly, we replicated this in an *independent* extract from the same dataset, analysed by a different team at Imperial College London, with a different data granularity (week vs month) and with a different inclusion lookback period. This is reassuring! /cont

Does the (live) shingles vaccine protect against dementia? Fergus Hamilton and colleagues provide important evidence in a new preprint using hospital data from the whole of England in a regression discontinuity design from which we *don't* show evidence of a meaningful protective effect /cont

3/n First, we show a robust effect on shingles diagnoses using this date of birth cut off - this accumulates over and is about the same relative effect as reported in the trials. Importantly, this is *hospital* coded shingles - which is ~10% of all shingles, so the outcome is rare. / cont

Does the (live) shingles vaccine protect against dementia? Fergus Hamilton and colleagues provide important evidence in a new preprint using hospital data from the whole of England in a regression discontinuity design from which we *don't* show evidence of a meaningful protective effect /cont

With England vs Argentina in the world cup don’t get too excited if it gets to a penalty shoot out .. here an analysis of the 1998 shoot out vs Argentina using regression discontinuity shows increase in heart attacks pubmed.ncbi.nlm.nih.gov/12493655/

Yet another spurious non-linear Mendelian randomization paper has been retracted by a journal academic.oup.com/jpubhealth/a... Kudos to the editors of the Journal of Public Health for following through and retracting spurious publications. What is shocking is that many other journals /cont

A comprehensive new tool for exploring how genetic variation influences outcomes, of relevance to everything driven by phenotypes, which encapsulates many domains, from how drugs work to how natural selection works ... / cont