Get PulmCCM’s Weekly Email Update
Stay up-to-date in pulmonary and critical care. No spam.
“What time is this, to trade the handshake for the fist?”
As remarked by the authors of the recent PLUS trial, a systematic review and meta-analysis of randomized controlled trials comparing balanced resuscitation fluid to 0.9% saline was reported in the New England Journal of Medicine Evidence. This methodical tour de force was crucial given the outwardly incongruous conclusions of the SMART, SPLIT, BaSICS and PLUS trials. Might the small mortality benefits observed in some of these large randomized trials become significant with greater statistical power? Might a Bayesian paradigm give us a slightly different perspective when thinking about the lumped results? Given the hundreds of millions of litres of IV fluid administered every year, even very small effect sizes could have implications at the population level.
Thus, the authors of the systematic review and meta-analysis approached their synthesis with the following question: in critically-ill adults, does the use of balanced crystalloid solutions compared with 0.9% saline reduce mortality and/or the occurrence of acute kidney injury?
What they did
The authors systematically searched Medline, EMBASE, and five clinical trials registries to identify RCTs, crossover trials, and cluster RCTs. Studies were eligible for inclusion if they: recruited critically-ill adult patients [from the ED, OR, ICU]; compared balanced crystalloid solutions [e.g., Plasma-Lyte, Hartmann’s, LR] with 0.9% saline for resuscitation and/or maintenance therapy. Nonrandomized trials, trials of patients who were not critically-ill, and trials in which there was concern about scientific misconduct were excluded.
The primary outcome was 90-day, all-cause mortality in trials deemed to be ‘low-risk-of-bias.’ If 90-day mortality was not available, reported mortality at the point nearest to 90 days was used.
There were multiple secondary outcomes, including: the proportion of patients with AKI as defined in the original trial, mortality at the longest interval, the proportion of participants requiring new renal replacement therapy, ventilator- and vasopressor-free days to day 28, quality of life, and functional outcomes.
What they found
The authors retrieved 1779 records plus three unpublished studies. Ultimately, the authors included 11 studies and two conference abstracts, for a total of 35,884 participants. Six studies with 34,450 participants were deemed to have low risk of bias.
The pooled estimated risk ratio [RR] for 90-day mortality [or closest time point] for balanced crystalloid solutions compared with 0.9% saline in the six studies with low risk of bias was 0.96 [95% CI, 0.91 to 1.01].
When using a Bayesian paradigm with vague priors for the primary outcome, the authors calculated an 89.5% probability that balanced crystalloid solutions were associated with lower mortality compared with saline.
With regards to acute kidney injury, and considering only the trials adjudicated as low risk of bias, treatment with balanced solutions compared with saline demonstrated a 0.96 RR for developing AKI [95% CI, 0.89 to 1.02] and a RR of being treated with renal replacement therapy of 0.95 [95% CI, 0.81 to 1.11]. No other secondary outcomes or sub-group analyses revealed clinically or statistically-significant differences between the groups.
Does this data change practice? Will those who feel, deep in their bones, that there is truly no difference between bags of saline and lactated ringers now reach for the latter when resuscitating a critically-ill patient?
After reading these different trials, I was immediately reminded of the provocative commentary written by Jean-Louis Vincent over a decade ago entitled: We should abandon randomized controlled trials in the intensive care unit. That piece made an impression on me at the time and is a commentary that everyone, I think, should read. While I am not proclaiming that the randomized controlled trial is a failed paradigm, we should recognize the limitations itemized by Professor Vincent. In doing so, we grasp that even with meta-analyses, comparing patients and interventions and outcomes between trials is a defiant endeavour. This makes clinical translation difficult, opposing facile interpretation for any particular patient, leaving the clinician with the same old preconceptions.
Additionally, the results of both BaSICS and PLUS play into what Professor Vincent called the ‘pendulum effect’ where an RCT in the ICU is favourable for a particular intervention only to be followed by multiple ‘negative’ studies. This then leads to therapeutic nihilism amongst intensivists, where trials first-reporting beneficial results are treated skeptically and implementation is resisted until replication which will, in anticipation, show no benefit. This enforces a positive feedback cycle where initial resistance is later rewarded by the dopamine surge associated with ‘I told you so.’ No doubt, it is safe to wager against any intervention in the critically-ill. In a brief review of major RCTs in emergency and critical care medicine, comprising 100,000 patients, mortality was 40% in the lumped intervention arms and 39% in the lumped control arms. With this sentiment, why do anything at all?
Another criticism of randomized trials worth emphasizing is the statistical ontology upon which these studies typically found themselves. Traditional, frequentist approaches are often held sacrosanct. Readers and researchers genuflect before the p-value, thresholding the world into accepting or rejecting the null hypothesis; is this a false dichotomy? In parallel, with the rise of adaptive platform trials during the COVID-19 pandemic, a separate statistical paradigm is emerging in the critical care universe. The difference between frequentist and Bayesian perspectives was superficially considered in the context of the COVID STEROID 2 trial. As a loose summary of Bendtsen, perhaps the best distinction between the two statistical genres is that the frequentist position authorizes the statistics to infer for us; for the Bayesian, clinical inference materializes after the stats.
And so, with these two mathematical attitudes the reader imposes her or her own predeterminations to the results of the meta-analysis. There is no statistically-significant difference in 90-day mortality between those randomized to 0.9% saline and balanced crystalloids; also, there is a 90% probability that the average treatment effect of using balanced crystalloids is to reduce mortality.
Alas, it looks as if the practice of critical care medicine will be ‘choose-your-own-adventure,’ ad infinitum.