Irrelevant Yet Significant: The DEFENDER study (part 1/2)

How a cognitive bias will soon be presented as Evidence-Based Medicine

I have recently acknowledged there is an ongoing clinical trial of dapagliflozin in patients with organic dysfunction in the ICU. The DEFENDER study rationale and design are here. As we are used to seeing in recent critical care research, the RCT meets the highest standards.

Unfortunately, the highest standards are blind to marginal utility, and the study strays into the same misconceptions I’ve been talking about in this Substack. We can foresee the study will find clinically irrelevant effects. However, there is novelty: the primary endpoint (a compound of death, initiation of hemodialysis, and ICU stay censored at 28 days) will not be assessed by merely counting successes in one group and the other. Instead of reporting the difference in outcome incidence between groups, researchers will determine the winner.

But first, let me walk you through the shortcomings of the study hypothesis. The new P-fishing method is presented in the second part of this essay.

Dapagliflozin is an inhibitor of the sodium-glucose cotransporter-2 (SGLT-2), a transmembrane protein in the proximal renal tubules that reabsorbs glucose and sodium. It helps diabetic patients by promoting glucosuria and helps those suffering from cardiac failure because it is a diuretic.

By the early 2000s tight glycemic control was a main topic in critical care. The trend was mainly driven by one study¹, however, as a display of compliance and deep commitment to evidence-based critical care, we would try to apply the tight glycemic control protocol. Regrettably, the protocol required testing capillary glucose by the hour in every ICU patient. All of them, all night long, every night, giving insane high insulin doses to reach normal glycemia. I regret I was so young and susceptible! Later it was found useless and dangerous, and more liberal approaches to glycemic control prevailed. Someone will argue that this time it is different because, you know, maybe the problem was insulin, etc. I don't buy it. No one in the ICU is dying of moderate hyperglycemia, and the unavoidable consequence is that no one will survive because you induced glucosuria with dapagliflozin.

The other effect of dapagliflozin is inducing diuresis. Promoting diuresis is my preferred intervention. I joke I allow residents to give 40 to 400 mg IV furosemide a day without even telling me. There is evidence that a restricted approach to fluids in early shock and ARDS, and active de-resuscitation in late phases of shock are helpful. However, it is difficult to make the case for mortality reduction. Anyway, I may consider dapagliflozin to induce diuresis and ease glycemic control with a single drug.

Above is a rationale of potential benefits considering the drug mechanism of action and some priors related to these mechanisms. Now, we enter the realm of cognitive bias. The DEFENDER study rationale claims several other mechanisms: improved metabolic efficiency and endothelial function, inhibition of pro-inflammatory pathways and sympathetic activity, decreasing production of reactive oxygen species, increased erythropoietin production, reduced biomarkers of inflammation, and reduced pathological findings of lung injury.

Wait a minute! They say we may improve "metabolic efficiency" with dapagliflozin! How can I know that my patient needs more metabolic efficiency? oh, everyone needs it, right? God… It’s so close to Integrative Medicine that I urge to knell in contrition before my Cecil's.

The authors state 10 different mechanisms. In a review article, you will find more, surpassing 20. These include some of the darlings of critical care physicians: glycemic control, enhanced diuresis, mitochondrial dysfunction, sympathetic tonus, inflammatory markers, etc. The effect appears so complete that apparently, all patients might benefit. Accordingly, the DEFENDER study tested the drug in unselected patients with worsening organic dysfunction (renal, cardiovascular, or respiratory) of any cause. It is nothing but surreal.

Maybe every acute renal, cardiovascular, or respiratory failure has the same mechanism and dapagliflozin will fix it by blocking glucose and sodium tubular reabsorption. That would be the most spectacular discovery in decades.

The DEFENDER study epitomizes the cognitive bias I termed Blindness to Marginal Utility. It scatters multiple plausible interventions around multiple plausible and possibly concurring causes of organic dysfunction but doesn't even try to identify a dominant cause of death, worsening renal failure, etc to be treated. In other words, it offers tens of marginal treatments. Due to the exponential distribution of the relative size of therapies' marginal efficacy, we know that adding 10 marginal interventions doesn't get the same clinical effect as adding one dominant intervention. You don't need 10 mechanisms of action for antibiotics or thrombolytics. One is enough if you target the dominant cause.

I won't get into a more trivial bias commentary, like pointing to the fact it is an open study, making it prone to performance bias in a setting where researchers are highly motivated to publish the benefits of the drug. The study may even turn out to be a Black Swam, a strike of pure luck, serendipity, or the Providence. In that case, if DEFENDER shows an absolute mortality benefit (1) not linked to enhanced diuresis and (2) large enough to be spotted with basic inferential statistics, I swear I will spend the rest of my days prescribing dapagliflozin for everyone who crosses the ICU door.

The DEFENDER study is a special one. Not because the hypothesis it brings is so helplessly weak. Bad hypotheses are everywhere in Critical Care literature. DEFENDER is special because it will bring an innovative method to show you a beautiful P-value.

In the next post, you'll discover why DEFENDER is the first Irrelevant Yet Significant study.

1. I was in the last year of my residency when NEJM published this study in 2001. It is incredible how the mainstream intensivists swallowed an NNT of 3 for hospital mortality. The hypothesis is so weak, the study is so biased, and the results so surreal that, although I was unable to formulate clearly what I felt was wrong, this study created in my young self a deep disdain for those in the academy.