This talk covers several ways to make clinical trials more efficient and to reduce the chance of ending with an equivocal result. Some of the approaches covered are Bayesian sequential designs allowing for study extension if results are promising, not being tied by type I assertion probabilities/α spending, using high-information longitudinal ordinal outcomes, and covariate adjustment.

Comments from the chat:

Babak Oskooei: Type I error rate is not a perfect measure - every statistician is aware of that. But it is a performance measure that has been traditionally used. It helps in ensuring that the finding are replicable/reproducible etc.
David Colquhoun: what was the link to simulations -bit.ly/?
Svetlana Cherlin: https://hbiostat.org/r/hmisc/gbayesse...
Marianna Nodale: Regarding the longitudinal odds model, how do you assess whether your model is misspecified? It is often the case that odds ratios at each level of the outcome variable is different, possibly with a consistent direction at all levels (worse odds towards one end of the scale, ie death) or possibly wavering either side of zero. Rare to have similar odds across the scale.
Saul Richmond: Are you comfortable using Natural History as an alternative to placebo controlled trials in ultra-rare diseases in which recruitment is challenging? Any thoughts?
Francois Maignen: Are the major current limitations of clinical studies the inherent responsibility of regulatory agencies (eg FDA, EMAat are the reasons underlying the overreliance of regulatory agencies on sample sizes, alpha errors, etc.?
Alain Amstutz: What do you think about covariate adjusted average marginal estimates - I guess you were talking only about covariate adjusted conditional estimates?
Phoebe Kitscha:found this very interesting thank you :) wondering how some of this could work for research funders, we as a funder of trials are seeing more applications with bayesian designs but committee views are quite mixed on them, and we havent got a handle yet on how we deal with it when sample sizes need to increase (in terms of increase in funding required) but this has given me lots of things to think about
Eamonn: Hello Frank, great talk. What advice would you give to changing the status quo? For example, change from baseline is still a very common analysis and I believe the FDA do not seem to be criticial of the approach? Secondly, are there examples of (Bayesian) trials in which analyses have been performed as data is flowing in? Thank you
David Colquhoun: Apologies for my incoherence. The distinction I was trying to make is explained here https://royalsocietypublishing.org/do...