Journal Club

Callum Taylor

2024-01-29

Does anything in critical care work?

2008

Multicentre, randomized, controlled trials
evaluating mortality in intensive care:
Doomed to fail?

Ospina-Tascon, Crit Care Med

1241 903 (Not Adult ICU RCT )

338 265 (Not Mortality)

73 4 (Not 50+ Patients)

69 3 (Missed in Initial Search)

72

1987 - 2006

10 / 72 trials

14%

2019

Which Multicentre Randomized Controlled Trials
in Critical Care Have Shown Reduced Mortality?

Santacruz, Crit Care Med

27 / 212 (17 new)

13%

Bayesian Reanalysis (2021)

. Freq +ve Freq -ve
Bayes +ve
Bayes -ve

. Freq +ve Freq -ve
Bayes +ve
Bayes -ve

Journals

2008 - 2018

  1. American Journal of Respiratory and CCM
  2. JAMA
  3. The Lancet
  4. The Lancet Respiratory Medicine
  5. NEJM

Search Strategy

82 Papers

Bayesian Positive Trial?

ARR MCID

Results

. Freq +ve Freq -ve Total
Bayes +ve
2
7		 9
Bayes -ve
2
71		 73
Total 4 78 82

Priors

Bayes +ve / Freq -ve

Power

Bayes +ve / Freq -ve

50% MCID?

Bayes -ve / Freq +ve

ARR < MCID

Conclusion

  1. “Everyone” overestimates trial effect sizes
  2. Most critical care trials don’t improve mortality (5-15%)
  3. Most Bayesian and frequentist trials agree
  4. We can understand why they disagree
  5. We can protect against over enthusiastic Bayes