IV fluid choice part 2: The SMART trial

Yesterday, I covered the SPLIT trial, comparing saline with Plasma-Lyte 148. Today we are going to look at SMART, the trial that provoked the numerous cringe-worthy popular headlines…

The paper

Semler MW, Self WH, Wanderer JP, et al. Balanced Crystalloids versus Saline in Critically Ill Adults. The New England journal of medicine. 2018; 378(9):829-839. PMID: 29485925 [free full text]

The Methods

This is a pragmatic, unblinded, cluster randomized, multiple-crossover trial done in 5 ICUs at a single university hospital.

Patients: All adult patients admitted to the ICU.

Intervention: Balanced crystalloid (Ringer’s or Plama-Lyte-A) for any crystalloid infusion.

Comparison: 0.9% saline for any crystalloid infusion.

• Each ICU used one fluid during the even months and one fluid during the odd months. The only part that was randomized was which fluid you started with. Patients admitted from one month to the next had the type of fluid switched on the first day of the month, and so were exposed to both.
• The volume and rate of both fluids was decided by the treating physician.
• Clinicians were prompted to consider relative contraindications to balanced solutions: hypokalemia and traumatic brain injury. Clinicians could decide to use the alternate fluid if they wanted.

Outcome: The primary outcome was a composite of death, new renal-replacement therapy, and persistent renal dysfunction (a creatinine >200% of baseline).

The Results

15,802 patients were enrolled in the trial.

Only a very small amount of fluid was used: about 1 litre in total during the entire ICU stay in both groups. About 5% of patients received both fluids because they were in the ICU during 2 different months.

The primary composite outcome occurred in 14.3% of the balanced crystalloid group and 15.4% of the saline group (OR 0.91, 95% CI 0.84-.099, p=0.04). When I put these numbers into a fragility index calculator, the fragility index is 0, meaning the calculator thinks these results were not statistically significant. I get a p value of 0.06.

Mortality was not statistically different: 10.3% in the balanced group vs 11.1% with saline (p=0.06).

Need for dialysis was not statistically different: 2.5% balanced vs 2.9% saline (p=0.08).

Persistent renal dysfunction was not different (6.4% vs 6.6%, p=0.60).

None of the secondary outcomes (ICU free days, ventilator free days, vasopressor-free days) were statistically different.

In the subgroups, benefit was seen in both sepsis patients and patients with a history of renal replacement therapy, although no mention was made of adjustment for multiple comparisons. (They do tell us that with 14 subgroups, the chance of a p value less than 0.05 by chance alone is 51.2%).

My thoughts

There are many physiologic reasons that normal saline isn’t all that “normal”, and many intensivists are convinced of its evils. That being said, to date there actually hasn’t been any good evidence of clinical harms from normal saline. The SPLIT trial was the one big previous RCT, comparing Plasma-Lyte 148 to saline in 2278 ICU patients, and there was no statistical difference between the two groups. (Young 2015)

This is a very large study looking at an important question, but it has some significant flaws. The trial was not blinded. Furthermore, patients were not truly randomized, but rather assigned a fluid based on the month in which they were admitted, and clinicians could decide to use the alternate fluid if they saw fit. These facts mean that differences in patient selection or treatment could easily have been introduced if the treating physicians had pre-existing beliefs about fluid management.

When I started reading the methods, I was very concerned that they were using a composite outcome. Often, we see less clinically important outcomes, like lab values, drive these results, while the truly patient oriented outcomes remain unchanged. Composite outcomes can increase the power of a study, by increasing the number of events seen, but if you need a composite outcome in a trial of 15,000 patients the difference between the groups is likely to be trivially small. However, in this trial, the difference reported seems to be driven primarily by the important outcomes of mortality and renal replacement therapy, although neither of those outcomes were individually statistically significant.

Only half of these patients were admitted from the emergency department, so extrapolation to ED practice isn’t easy. I will also note that this seems like a very healthy group of ICU patients, with only ⅓ using mechanical ventilation, and only ¼ receiving a vasopressor. That  doesn’t sounds like any ICU I have worked in.

The amount of fluid used was tiny, and not in keeping with most ICU practice I have seen. I would not have expected to see a difference in outcomes from only a single litre of fluid. Would we have seen bigger differences if larger volumes of fluid were used? Or does the tiny amount of fluid used decrease the biologic plausibility of this finding? (For reference, in SPLIT the median fluid volume was 2 L per patient, and no difference was seen). (Young 2015) Either way, we should be wary of trying to extrapolate these results to patients requiring large volume resuscitation.

We always want to see large trials, but one problem with large trials is the possibility of seeing statistically significant results with very small numbers. The difference here is small, if it is real. My biggest concern is that when I run the numbers myself, the result is not statistically significant, with a p value of 0.06 and a fragility index of 0. I am sure that a statistician could clarify why I am getting different results (I think this is because this is actually an observational study, and the authors used logistic regression in an attempt to match the two groups), but if you have to split statistical hairs to make a trial statistically significant, I am very skeptical of the results.

Bottom line

For me, this is not a practice changing study. It is the first study of its kind to show clinically important differences when choosing balanced fluids over normal saline, but the various methodologic flaws mean that these results can’t overrule the previous SPLIT study. If anything, this shows that saline is mostly safe (at least in small volumes), and will probably continue to be my fluid of choice until we see more data. (I understand another large RCT – PLUS – is in the works). I do not know how to extrapolate these results to large volume crystalloid resuscitations, where the physiologic problems with saline might be more relevant.

Other FOAMed

The Bottom Line

St Emlyn’s

REBEL EM: Is the Great Debate Between Balanced vs Unbalanced Crystalloids Finally Over?

PulmCrit- Get SMART: Nine reasons to quit using normal saline for resuscitation

EMNerd: The Case of the Unbalanced Solution

EM Journal Club

The Resus Room: The Crystalloid Debate

References

Young P, Bailey M, Beasley R, et al. Effect of a Buffered Crystalloid Solution vs Saline on Acute Kidney Injury Among Patients in the Intensive Care Unit: The SPLIT Randomized Clinical Trial. JAMA. 2015; 314(16):1701-10. [pubmed]