Has the STAMINA study just pronounced ARDS’ death?
The Brazilian study raises interesting questions about the future of the synthetic syndrome.
THE ZEITGEIST… one day you look at the people around you and realize something has changed. At fifty years old, I started to get used to this feeling. I can't engage in conversation with people the age of my kids (19 and 20 years old) without feeling I am the strange one. Young condescendence increasingly softens my days as a professor and as a physician.
One of these days, I was lecturing the students about the importance of giving early antibiotics if the patient meets sepsis criteria. A student, incited to delineate her strategy, asked me if she could give early antibiotics to every infected patient instead of using triage tools like qSOFA. She argued it would be easier to implement and probably more efficient. Well, of course, yes, why not?
Without knowing or caring, she buried down 30 years of sepsis definitions and exposed their utter inconsistency and practical irrelevance. The rest of the class agreed matter-of-factly. It promptly became a non-issue for them. For me, however, it struck a darker note. Now, it seems all I have studied about sepsis, my Master’s thesis and articles, uncountable rounds and classes and meetings sessions, everything became part of the stories this aging physician one day will present before the polite reverence of the younger doctors.
Tempus Fugit!
The Times They Are a-Changin'
The successive ARDS definitions are mistakes that allured generations of intensivists but failed to generate real advancements in patient care. It is increasingly embarrassing to realize how mindlessly we accepted the oxymoronic “heterogeneous syndrome” thing and how we have merrily prompted ourselves to apply one-size-fits-all generic strategies to heterogeneous patients, clinical trial after clinical trial, patient after patient.
All we learned in 60 years of ARDS was to avoid ZEEP and not to harm patients with excessive pressure or volume. No treatment emerged. Moreover, some 30 years ago the VILI zeitgeist seized us. Since then, we have mostly discussed ways to minimize lung injury while buying some days for the patient to recover. In 2024, we are witnesses to the exhaustion of the protective ventilation approach. There is nothing to add. The STAMINA study, however, attempts to add nuance to protective ventilation.
The questions I pose here are: Is STAMINA an ARDS study? If so, is it reproducible?
Dr. Lawrence Lynn raises the reproducibility problem in ARDS clinical trials precisely because of heterogeneity (see figure below). He states ARDS is not a disease but a collection of diseases with similar presentations. These diseases fulfill a set of arbitrary threshold definitions like a pre-specified PaO2/FiO2 ratio.
However, it makes no sense to say that a patient with a PaO2/FiO2 ratio slightly above the arbitrary threshold is free of the disease and a patient slightly below is not. We should not define a disease by the severity of its presentation while disregarding any biological mechanism.
ARDS is a Synthetic Syndrome, a set of arbitrary thresholds mimicking a disease definition. Synthetic Syndromes are not discovered in a breakthrough that brings up a new and promising disease model. Instead of being discovered, Synthetic Syndromes are arbitrarily defined and redefined by a consensus of experts.
The syndrome is a teleology, that is, its definition stems not from the nature of the disease, but from the need to enroll patients in clinical trials.
We are pretty confident that all clinical trials enrolled patients with hypoxic respiratory failure, bilateral chest infiltrates, etc. However, we don’t know how the disease developed. We ignore the underlying causes of such clinical presentation.
Moreover, we ignore the diseases set in each studied population, rendering RCTs irreproducible. That’s the apical error made by Petty sixty years ago. He lumped several different diseases under the umbrella term ARDS, for the convenience of finding patients for clinical studies. In hindsight, today it is obvious that this strategy would fail.
Obvious? Not for everyone. Sixty years ago, randomized clinical trials were transforming the art of medicine into a science. It was not long since the RCT debuted in a landmark tuberculosis study. Of course, the RCT method was to be applied to a single disease. However, here is Petty's mistake. He presented his lumping ARDS definition as a "disease equivalent" and statisticians fooled their selves into believing it was enough to proceed.
I think this historical mistake has not yet been fully acknowledged by data scientists involved in ARDS research.
You get the full story and much more in the book:
STAMINA is the first post-ARDS study
Now, back to the STAMINA study. The trial is actively recruiting and results are expected soon.
Population: Patients with ARDS due to pneumonia (includes Covid)
Intervention: A ventilatory strategy based on a driving pressure goal
Control: ARDSNet Low PEEP table
Outcome: Ventilation-free days at the 28th day
STAMINA patients are dying from infection and ventilatory failure. Hence, the dominant interventions must be antibiotics (as necessary) and mechanical ventilatory support. Due to the exponential distribution of marginal treatment effects, we should expect to see decreasing yields for each additional therapy. Of course, these patients will receive the best ICU care available in the world. Each little intervention they receive, including the study intervention, is subjected to marginal utility as expected in The Additive Paradigm.
STAMINA compares two different strategies for delivering mechanical ventilation. These diverse approaches should result in different airway pressures and volumes to set a meaningful difference between groups. Hence, STAMINA tests the marginal clinical effects of pressure and volume units. As we know, they will not allow for extremes in volumes and pressures. I expect small between-group differences in pressures and volumes, held within a reasonable and safe range. Small and safe differences predictably produce a marginal effect, marginal effects are irrelevant, etc.
One eyebrow-raising feature of the STAMINA study is the small sample size1. Marginal effects are usually clinically irrelevant and require large sample sizes, translating to large amounts of time and money. However, depending on how data is treated the irrelevant effects may become statistically significant and shine in a beautiful and publishable paper. There's a trend of “augmented analysis” or another name like this, normalizing procedures people used to call P-fishing in the good 'n' old times.
Beware of fancy inferential analysis. They are a disguise for weak hypotheses
My frequent reader is probably used to the marginal utility causality analysis and aware of my despise for fancy "creative” inference. I won't go further. I brought up the STAMINA trial to discuss ARDS definitions and RCT reproducibility.
The decision to enroll only patients with pneumonia greatly reduced heterogeneity and from a practical standpoint makes STAMINA a pneumonia study.
STAMINA studies the marginal effect of two ventilatory strategies in patients with pneumonia receiving mechanical ventilation. We don't need ARDS to describe the study! It is a pneumonia study, testing an intervention retrospectively extrapolated from ARDS studies.
Recalling our reproducibility discussion, it is easy to see that STAMINA's case mix is so diverse from previous ARDS studies that it shouldn't be regarded as one. It avoids the “heterogenous syndrome” mistake and is not reproducible inside the ARDS paradigm.
STAMINA is the first post-ARDS study. It inaugurates a new era, moving away from a failed disease definition.
Let’s accelerate. It’s time to abandon ARDS definitions.
The times they are a-changin'.
More about the book
More about marginal utility
Clinically irrelevant yet statistically significant
The study protocol projects a 3-day difference in the primary outcome, ventilation-free days, from 4.7 to 7.8. This is an obvious case of Delta Inflation. It's time for innovative inferentials!
Moreover, the protocol specified mortality as a secondary outcome. It plans to detect a 5% decrease in mortality (60 to 55%) with 90% power. My old Stata estimated the sample size at 4,100 patients, using old-fashioned sample size estimates.
I have long argued that much of the research we do is a lot like ARDS research in that either the definition of the thing studied or the intervention is, at the end of the day, poor.
We can get into situations were we study poorly understood or more likely a final common pathway for several dieases. We fall in to the trap of "oh, if we only have 25,000 more people we could have shown a statistically significant difference"
Back to the past. In 1984 we were titrating peep to best compliance and FiO2 no greater than 60%.