When Hell Freezes Over: Trials of Temperature Manipulation in Critical Illness

The bed is on fire

Two articles published online ahead of print in the NEJM last week deal with actual and attempted temperature manipulation to improve outcomes in critically ill patients.

The Eurotherm3235 trial was stopped early because of concerns of harm or futility.  This trial enrolled patients with traumatic brain injury (TBI) and elevated intracranial pressure (ICP) and randomized them to induced hypothermia (which reduces ICP) versus standard care.  There was a suggestion of worse outcomes in the hypothermia group.  I know that the idea that we can help the brain with the simple maneuver of lowering body temperature has great appeal and what some would call "biological plausibility" a term that I henceforth forsake and strike from my vocabulary.  You can rationalize the effect of an intervention any way you want using theoretical biological reasoning.  So from now on I'm not going to speak of biological plausibility, I will call it biological rationalizing.  A more robust principle, as I have claimed before, is biological precedent - that is, this or that pathway has been successfully manipulated in a similar way in the past.  It is reasonable to believe that interfering with LDL metabolism will improve cardiovascular outcomes because of decades of trials of statins (though agents used to manipulate this pathway are not all created equal).  It is reasonable to believe that intervening with platelet aggregation will improve outcomes from cardiovascular disease because of decades of trials of aspirin and plavix and others.  It is reasonable to doubt that manipulation of body temperature will improve any outcome because there is no unequivocal precedent for this, save for warming people with hypothermia from exposure - which basically amounts to treating the known cause of their ailment.  This is one causal pathway that we understand beyond a reasonable doubt.  If you get exposure, you freeze to death.  If we find you still alive and warm you, you may well survive.

But the situation with this trial and its underlying hypothesis (which is never actually stated or supported in the manuscript - omissions that are all too common) is even more curious.  What causal pathway were we trying to interrupt?  Do we know any causal pathways in traumatic brain injury?  Is the hypothesis that lowering ICP with hypothermia improves outcomes?  If so, is it simply through reduction of ICP?  Do we know that reducing ICP improves outcomes by any means?  Or was the hypothesis that lowering body temperature has other salutary effects on neural injury?  It certainly sounds as though we're shooting in the dark (with perhaps some collateral damage in this case) and we're going after elevated ICP not because we know anything about its relationship to causal pathways, but rather because we can measure it and manipulate it with the implicit assumption of causality.  The problem is that even if ICP is causal to outcomes of interest, there are several reasons why manipulating it may fail to achieve the desired result, chief among them time dependency of the causal pathway.  It could well be that by the time ICP becomes elevated, the horse is already out of the barn.

Go ahead and fix the door - this horse is already out of the barn.

I an not being a nihilist here.  I ride motorcycles and bicycles (with a helmet) but I imagine I'm still at risk of TBI, so I welcome a therapy for it.  But a reasonable person accepts the data and considers that, after all, they do not support the hypothesis, and attempts to think of all the reasons the hypothesis was wrong, rather than rationalize why it still might be right, under slightly different trial conditions, with different temperatures, etc.  A MAFSR (Meta-Analytic Futility Stopping Rule) may be in order here.  And perhaps it's time to return to the basics and see if mannitol and hypertonic saline and hyperventilation do any good - perhaps they're harmful too, and if so it opens the possibility that elevated ICP is protective in TBI, as heretical as that may sound.

Speaking of protective alterations in physiological numbers, wrought by evolution over the eons:  the HEAT trial randomized 700 ICU patients with fever and suspected infection to receive intravenous acetaminophen (APAP) versus placebo and found no difference in the primary outcome of ICU-free days.  This result does not surprise me because I think it is abundantly clear that fever is an adaptive response to infection, and any contrary hypothesis constitutes a normalization fallacy unless supporting evidence can be provided.  But, I fear that this trial failed to shed any light whatsoever on the hypothesis even if it was doomed from the outset.  The reduction in temperature achieved with IV APAP was highly statistically significant, but clinically minuscule - a mere difference of 0.2 degrees Celsius between the groups.  It is certainly optimistic to think that a reduction in temperature of this nominal magnitude could possibly influence clinical outcomes.  (Indeed if this trial had been positive, I would have said that APAP was beneficial not through temperature lowering but through some other pleiotropic effect.)  What this trial does prove is a welcome result, one which has a very high prior probability of being true:  that Tylenol is worthless in reducing fever in the ICU, as years of clinical experience reveals.  So now I can, with even greater confidence, tell the nurses who call me about fever in a patient already treated for known infection that no, I'm not giving Tylenol.  Call me back when the bed catches fire and I'll reconsider.

For further reading, see:  Fever, external cooling, biological precedent, and the epistemology of medical evidence