Monday, November 18, 2013

Dead in the Water: Colloids versus Crystalloids for Fluid Resuscitation in the ICU

It is a valid question:  at what point has a concept been tested ad infinitum such that further testing is not worthwhile?  There are at least three reasons why additional study of a concept may not be justified:

  1. Because the prior probability of success is so low (based on extant trials) that a subsequent trial is unlikely to influence the posterior probability that any success represents the truth.  (This is a Bayesian or meta-analytic worldview.)
  2. Because the low probability of success does not justify the expense of additional trials
  3. Because the low probability of success violates bioethical precepts mandating that trials must have added value for patients and society
And so we have, in the November 6th edition of JAMA, the CRISTAL trial of colloids versus crystalloids for resuscitation in the ICU.  As is customary, I will leave it to interested readers to peruse the manuscript for details.  My task here is to provide some background and nuance.

 This is the fifth large and noteworthy RCT to have investigated this concept:
  1. The SAFE trial (NEJM, 2004) randomized 6997 patients to albumin versus saline and found that relative risk (RR) of 28-day mortality was 0.99 with P=0.87.
  2. The VISEP trial (NEJM, 2008) randomized 600 patients to 10% pentastarch (an HES solution) versus modified Ringer's lactate solution and found the RR of 28-day mortality was 1.10 with P=0.48.  There was a trend (P=0.09) towards higher 90-day mortality in the pentastarch group.
  3. The 6S trial (NEJM, 2012) randomized 804 patients to tetraspan (an HES solution) versus Ringer's acetate and found a RR of 28-day mortality of 1.17 in the tetraspan group (P=0.03)
  4. The CHEST trial (NEJM, 2012) randomized 7000 patients to HES versus saline and the RR for 28-day mortality was 1.06 with P=0.26.
  5. Finally, the CRISTAL trial randomized 2857 patients to colloids versus crystalloids and the RR for 28-day mortality was 0.96 with P=0.26.
That's 18,258 patients randomized and the trials show either no effect with a RR point estimate almost dead on 1.0, or they show a trend towards harm in mortality and other secondary outcomes (with the exception of the CRISTAL trial which had some positive signals in secondary outcomes).  To even hope of budging the point estimate of a meta-analytic result, a future trial is going to have to have tens of thousands of patients and/or a marked divergence from the results summarized above.

But why, other than natural statistical variation, might there have been a difference in trends in secondary outcomes between CRISTAL and prior trials?  The most obvious and likely possibility is that CRISTAL was the only of the five trials that was not blinded. There are at least three reasons blinding is essential to avoid bias in trials:
  1. To prevent ascertainment bias in outcomes.  If an outcomes assessor has knowledge of the treatment assignment, and that outcome is soft or mutable, s/he may have a conscious or unconscious tendency to assess outcomes in accord with his predilections.  The CRISTAL trial had blinded assessors.
  2. To avoid the possibility that other processes of care are influenced by providers asymmetrically on the basis of knowledge of treatment assignment.  For example, clinicians in the study may give more pressors to patients assigned to either HES or saline.  It is for this reason that Alan Morris, a famed ARDSnet researcher at IMC in Salt Lake City is always talking about "adequately explicit protocols" for care for studies in the ICU that cannot be blinded.  It is only if you know that care is uniform in both groups that you can have confidence that any differences observed between randomized groups is due to the randomization variable rather than other processes of care.  Differences in process of care is a good explanation for the difference in results between CRISTAL and the other trials.
  3. To avoid the possibility that a differential placebo effect occurs if patients know their treatment assignment and have predilections about one of the interventions.  This is unlikely in the case of CRISTAL.
My best guess is that lack of blinding led to differences in processes of care between the randomization groups in CRISTAL and this explains the differences in secondary outcomes.

When I see that there was a significant difference in 90-day mortality  in the CRISTAL trial favoring colloids (one of at least NINE secondary outcomes), I am unmoved.  Rather than call it an exploratory finding (with the suggestion that maybe we should explore it further with additional trials), I'm going to call it a necropsy finding - something you find when analyzing a concept that is dead in the water.

I commend the investigators of all the trials mentioned above for giving me overwhelming evidence to guide my choice of resuscitation fluids to the safest and cheapest.  But please - you've outdone yourselves - more trials will not shed more light on this topic.  The answer is clear.  Crystal clear.


  1. In yesterday's JAMA the letters to the editor regarding this article can be found. Here's the first one:

  2. great blog!!! I agree with the above