Fluids in Sepsis. Dogma in doubt?

 

How is the study quality?

Well it is a great study. It set out to prove something that the authors thought was true (fluid boluses in septic African kids will save lives) so that they would have the ammunition they need to get this fairly basic and presumably life-saving intervention happening in hospitals that lack any coherent emergency medicine or triage or critical care system. Then, low and behold, it proved the opposite. Fluid boluses were actually killing these patients.

Now you have to love a paper that proves the opposite of what it set out to prove. Why? Because the fact that it hasbeen published means that it was a powerful enough result to overcome the prejudices and assumptions of the investigators and because it is a result that brings truly new knowledge.

As far as internal validity goes this paper was pretty good. The numbers were huge (how long would it take do you think to find that many truly septic children in a first world setting?), the randomisation was robust, the groups well matched, the intervention focussed on a single thing, protocol breaches minimal, analysis on intention to treat and differences between the two groups both statistically convincing and clinically meaningful. Inclusion and exclusion criteria were defined in the body of the article. There was only one primary end-point. No huge claims were made on the basis of secondary end-points or post-hoc sub-group analysis.

When I read an article like this there are a few things that jump out at me as potential weak spots (but which were not in this case):

  • Three groups (albumin, saline and no bolus)- multiple groups should always make you wary. There are two potential errors that crop up in RCTs using multiple groups. First you can declare that there is no difference because you ended up splitting up your population into too small groups (underpowered study). That was not a problem in this case: the numbers were huge and power was adequate and  they did find a difference anyway. Secondly, multiple comparisons are a great way to find a random difference (compare new drug, new drug on alternate days, new drug with salt tablets, new drug on Wednesdays and placebo- WOW! new drug on Wednesdays is better than placebo! Get the press release out.) That was also not a problem in this case as both bolus groups were essentially the same compared to placebo and whether looked at separately or together the result was consistent.
  • 2 strata (severely hypotensive kids who all got a bolus and the rest who got albumin, saline or placebo). This means you are essentially running two experiments. You have to be careful to ensure that the results from one are not muddled up with the question being asked by the other. (e.g. you give a drug or placebo to two groups of people with, oh, let's say stroke and in one group you are looking at 24 hour mortality and in the other you are looking at 4 month function and mortality and you find no difference in 24 hour mortality in the first group but when you have a post-hoc look at 24 hour mortality in the second group you find a difference and report that).  In this case the two strata issue was not much of a one as the severly hypotensive group was trivially small compared to the huge size of the main group.
  • Early cessation. Always beware early cessation. Was it really becasue of safety? Was it because the pendulum happended to have oscillated over to the result we were looking for? Not a problem here I think.

 

So why should it not change our managment?

Well the big question you must ask yourself is how similar is my population of sick kids to this one?

Remember that these are sub-saharan African kids. And they were really sick. They defined "prostration" as inability to sit up (or breast feed in pre-sitting kids). They defined severe malnutrition as kwashiorkor or visible wasting. The mean baseline Hb was  71 with 75% of kids having a Hb <10 and 30% an Hb <5. Half of the kids had malaria and 4% had HIV.

What will a 20-40mL/kg fluid bolus do to a baseline Hb of 4g/dL? There was a transfusion protocol for these kids but they got their transfusion relatively slowly (the protocol was 10mL/kg over 4 hours and looking a the fluid volumes delivered I am not certain how many of these kids would have actually got that blood in that time).  

And what disease did these kids have? Half had malaria. The other causes are not reported and were probably hard to define given the resource limitations of the setting but one would suspect that meningitis and pneumonia featured highly. Both of these are associated with SIADH meaning that a fluid bolus might have significant negative implications.

Remember also that these kids had no access to ICU. Most pneumonias do not die of septic shock but of respiratory failure and ventilatory support was not available to these kids.

So why should we not write it off completely?

Interestingly, the kids with HIV fared no differently than those without it. APO and raised ICP were not found to be more common in the kids who died (although how they were detecting raised ICP in this setting is not clear). In fact it is not clear what killed the kids in the bolus group. Until more research is done we are going to have to keep this study in mind when taking care of even fat healthy first world kids.

What this study really brings home for me is that in really sick kids we need to think past the simple APLS scenarios where a fluid bolus and a dose of ceftriaxone fixes everything and reach for vasopresor and intrope support early. Treating an adult with temperature of 40, HR 130 and BP 60/- we would not expect to give 20-40mL/kg of fluid and a shot of antibiotics and save the day. We would be reaching for the norad before the first litre was though. The fact that most of us rarely see a very sick child makes us hesitant managing them in the same way. This study reminds me that we need to overcome that.