We're looking for some fantastic critical care clinicians (of any discipline) to join our Trustee board. Please share with your relevant networks. @ICS_updates @FICMNews @UKCCRGroup @BACCNUK.

RT @UKRoxTrial: The UK-ROX trial has officially hit its recruitment target of 16,500 patients after 3 incredible years. 🌟. A huge THANK YOU….

Join millions who have switched to Grok.

RT @Septictrial: Today is #WorldSepsisDay! Our CI @agordonICU sheds light on how the study aims to find more effective treatment for sepsis….

RT @LaneCarrandi: Are you involved in the delivery of health services or health services research? I'd be grateful if you took 10-15 min to….

RT @callum_kaye: Some great responses to our survey with @hsru_aberdeen. Still time to share your views here: @UKC….

Although maybe out of scope, the brief section in the @covidinquiryuk Module 1 report on research omits that a big success of UK (and global) research in the pandemic — @remap_cap — was funded as pandemic preparedness research as part of the €32m EU @PREPARE_EUROPE project.

RT @ICNARC: Know any former ICU patients or family members who would like to be involved in improving ICU at a national level? Please put t….

I’ve actually oversimplified here- we need 57 levels to replicate the win ratio hierarchy: death=57, LOS+28 for survivors that received KRT, LOS for survivors that did not receive KRT.

It gives you an odds ratio as an effect estimate, which is generally better understood. And, even when the proportional odds assumption is violated, it gives you results that are consistent with the non-parametric approach. 7/7.

But you could also use a GLM approach, such as a proportional odds model - this would have been my primary analysis of choice for DEFENDER. This allows you to include covariates, such as the stratification variables from the randomisation, which is always a good idea. 6/n.

First, you could use a non-parametric approach: Wilcoxon rank sum with the probabilistic index (or concordance) as the effect estimate. Inference from this would be almost identical to the win ratio - the only real difference is how ties are handled. 5/n.

With these, you can very easily construct a single hierarchical ordinal outcome… assign deaths 30, survivors who receive KRT 29 and everyone else their ICU LOS 1-28. This massively opens up the options for analysis. 4/n.

So why wouldn’t I have used it in the #DEFENDERtrial and what would I have done instead? DEFENDER had a hierarchy of two binary outcomes (death and KRT) and one ordinal (ICU LOS to 28 days). 3/n.

The win ratio is useful for analysing a hierarchy of multiple long-term time-to-event outcomes, e.g. death, MI or hospitalisation for heart failure, which avoids the typical problem of time to first event that treats all these as equally important. 2/n.

I feel the need to wade in and dampen some of the enthusiasm for the win ratio at #CCR24. But first, in the interest of balance, when *do* I think it is a good approach… 1/n.

A bit of chat at #CCR24 yesterday about RAR - and I expect we may get some more this morning. I’d really recommend reading this balanced review by @SofiaSVillar1 and colleagues at @MRC_BSU

RT @HDR_UK: Check out the fantastic outputs of 19 funded orgs who were supported to implement the #UKHealthDataResearchAlliance Transparenc….

I’m delighted to be attending the @HDR_UK Transparency Showcase tomorrow to present our work to improve #TransparentDataAccess on the new @ICNARC website: #UKHealthDataResearchAlliance.

RT @CritCareReviews: Delighted to announce that Prof Kathy Rowan @KathyRowan101 will deliver the 2024 John Hinds Lecture at CCR24. "ICNARC….

RT @ATSBlueEditor: Temporal Trends in Mortality of Critically Ill Patients with Sepsis in the United Kingdom, 1988–2019. 🔗 .