1/7 #PREPAREII showed that a 500-ml fluid bolus does not prevent cardiovascular collapse during tracheal intubation in the ICU. I love efficient study designs - this trial had two features that in my opinion are worth replicating in future #criticalcare #RCTs (a thread 🧵):
2/7 Sample size: the authors calculated that 750 patients would be needed to detect a MCID with an 80% power. After a blinded interim analysis it became clear that the event rate is lower than expected. Smart study design feature no.1️⃣: adaptive sample size re-estimation.
3/7 The goal of sample size re-estimation is to prevent the dreaded "underpowered study" - a situation where the trial can no longer rule out that a clinically meaningful effect exists due to the signal/noise ratio in the data being to low. You can do it using both frequentist &
4/7 Bayesian statistics. In PREPARE-II, the authors looked at the overall event rate without knowing how many cardiovascular collapses occurred in the intervention/control group. ⬆️ sample size is not statistical trickery - it saves 💰 otherwise wasted on an uninformative trial.
5/7 Smart study design feature no.2️⃣: heterogeneity of treatment effect (HTE) analysis showing how often we are mistakenly focused on “subgroup effects”. A picture tells a thousand words: can you point to any SBP & baseline risk “subgroups” on these graphs? 
6/7 No - and you don’t need to do it! Both blood pressure measurements & risk estimates are continuous variables that should not be arbitrarily dichotomised (e.g. SBP <100 vs >100 mm Hg). #Dichotomania = throwing information away & ⬇️ power fharrell.com/post/errmed/
7/7 Congratulations to the @PCCRG for finishing another amazing trial in this area. Our Visual Abstract of the #PREPAREII trial is available on Instagram instagram.com/p/Cg1b4YdNn5o/… #CCR22 @pericrit @JonathanCaseyMD @brian_driver @toddrice_ICU @DerekRussellMD
@PCCRG @pericrit @JonathanCaseyMD @brian_driver @toddrice_ICU @DerekRussellMD *Sample size re-estimation: “(…) DSMB will evaluate the rate of the primary outcome in the no fluid bolus group. (…) Based on this information, the DSMB recommended increasing the total sample size from 750 to 1065 patients.” bmjopen.bmj.com/content/10/9/e… jamanetwork.com/journals/jama/…
• • •
Missing some Tweet in this thread? You can try to
force a refresh
