@fluidloading@bentatoo31@Denis_Faust@Procto_Log The story of pathological consensus and shortcutting the RCT method of Hill/Fischer with the Petty/Bone science. The lumping paradigm.
The Physician’s War: The Story of the Hidden Battle between Physicians and a Science Based on Pathological Consensus https://t.co/ZgCsoNXsRK
Kent’s framework improves assessment of treatment effects across prognostic risk groups. However, equal predicted risk does not imply equal treatment response. When a risk group contains different target diseases with opposing responses, its average treatment effect remains dependent on the disease mixture. Risk stratification alone therefore does not establish transportability. Now you might say who would mix target diseases? The answer is why not, when it’s all considered “a heterogeneous population under test” subject to HTE?
Daniel you are right and it’s worse than you think. HTE to the statistician includes disease mixture not just covariates. Precision and covariate adjustment cannot repair a structural mismatch. If treatment benefits disease A in the trial mixture but harms B, changing their proportions can reverse the average effect. Transporting a mixture average (E₃) as a disease-specific effect with covariate associated HTE (E₂) risks guideline failure. Learn about this here and forward widely to others bring a focus on the second (devolved) species of the RCT.
https://t.co/Izh5GXeyiu
Quote from my article about AI using EBM protocols in 2019
“For this reason designers of automated systems must recognize that the true complexity of diagnosis and management of adverse conditions, such as sepsis, resides in the portion of diagnosis and care provided by nuanced expert oversight which is difficult to study and reproduce. If the patient is worsening despite adherence to the standard protocol, a physician with naïve trust in RCT may adhere to the guidelines thinking she must stay the course. In the alternative the expert physician knows to detect and track worsening, knows the potential for HTE, and modifies care off protocol if necessary. In addition to responding to worsening, expert clinical nuance is also applied to change diagnosis and/or alter the care in the presence of an unusual presentation, rare cases, and less common mixes of overlapping diseases.
This right here, is why @AndromedaShock is truly pushing the needle forward. The ability to acknowledge the lack of knowledge in someone who is incredibly accomplished, is what allows for steps forward. I often talk about how medicine is hampered by large fragile, egos that, by their very nature, inhibit growth in a science that is so far from being complete and established. We have to be able to challenge existing dogma and not hesitate to acknowledge when we simply don’t know. That is what enables us to find out. Kudos to Glenn.
Super excited about the launch and the rapid growth (almost 500 members in 3 weeks!) of the https://t.co/iVaZzVmphG, but mostly about the building of a community of like-minded, forward looking clinicians of all levels of experience, a way to extend the unbelievable HR experience worldwide and year round. Thanks to the amazing faculty for all your help! As always, we are totally inclusive in terms of fields of expertise and especially growing the concepts of physiology and resuscitation outside of critical care/emergency/anasthesia into the spheres of internal medicine and those who take care of a large number of patients on the wards.
Very well. That’s a good idea Boris. It was fun but I was getting tired of it anyway. However I think it was needed initially. The causal-modeling-free RCT design dogma was “dug in like an Alabama tick”. (Blain)
But you are right I don’t really have only dog in the DT-CI fight. I respect the genius Professor, who doesn’t? I’ll let the math speak for itself going forward.
The issue is reproducibility and safe transportability, not the endless unmeasured-variable debate. Statisticians love safe debate. @f2harrell devoted an entire blog to answering @ildiazm they seem to have forgotten that the true relevance of a debate over chair position is a function of the integrity of the hull.
There is a gaping hole in the hull of RCT design without explicit causal modeling. It has plagued critical care physicians and their patients for decades. Statisticians’ reluctance to debate it leaves the central question… what species of RCT design generates estimands that are less safe to transport?
Exposing the species diminishes the cherished universality of the RCT “gold standard” label.
Learn why and how to detect it.
https://t.co/Izh5GXeyiu
@ildiazm@f2harrell Think you’ll convince the professor? No. Our shared goal should be to convince the audience and @NIH that optimal patient safety requires explicit causal modeling of RCTs. Offering a “safe” debate lets participants appear open to deep scrutiny while avoiding uncomfortable truths.
@Foreman1David@RWJE_BA@f2harrell@yudapearl From my experience, statisticians do not know the difference. Yet their silence is telling. A friend told me he was sent a DM not to engage with me.
(They didn’t know he was my friend.) He indicated that was when he knew they were hiding to protect dogma. Indeed.
@Foreman1David@RWJE_BA@f2harrell@yudapearl “Functional causality” assessment requires “estimand assessment”.
Is the trial generating E₂(i), an average across patient covariates within a target disease, or E₃ , an average across a mixture of diseases?
The guide teaches statiticians & pathophysiologists the difference
It’s nearly impossible to find a statistician who won’t debate criticism of “the RCT”. Statisticians commonly salute the technique in the first paragraph of their writings. This blind faith allowed the an (often pathological) second species of RCT to emerge unnoticed. This is the story of the second species which was formally characterized last year. The math and implications are so shocking and compelling that even the most pugnacious the statisticians have gone silent.
https://t.co/Izh5GXeyiu
Understand David I am not criticizing randomization but simply pointing outlines that it is not safe to use randomization (in the blind) with the entry criteria chosen by one set of experts and the statiticians just saying “that’s the consensus, ok we will plug that in”.
Randomization is not magic and what you randomize matters.
I am bypassing statisticians because they show no present interest failure mode analysis of RCT nor will they debate it.
Their silence is why I produced “the pathologist’s guide to RCT structure” to teach pathophysiologists (acute care physicians) and causal inference leaders.
This guide (written almost at the lay level) can be used to help the CI community understand the evolving present state of DT where new methodologies have become commodities.
David come over to Datamethods and see the patient safety consequences of operating “in the blind” with design theory (DT). In DT the math shapes the output. That’s not the way it should work. Causality as a function of pathophysiology should shape the output.
https://t.co/MwVrdGkp7R
The entire causal inference community should read the linked guide. It’s a look into a hidden ( mathematically embellished) side of RCT and design theory essentially converting sequential mini-trials into smoothed time series despite cause-agnostic gates and shifting disease mixtures over time under treatment and in controls. You can’t make this stuff up. It’s a fully endorsed level of “runaway design” with disregard for causality and pathophysiology that none of you ever imagined existed.
https://t.co/Izh5GXf682
“Causal-phobic”… perfect word. But why? The reason is that there is an entire standard species of RCT which emerged ~35 years ago which incorporates a cause-agnostic gate, which would suddenly come under deep scrutiny if exposed, harming the RCT brand and adding a new required layer of formal causal rigor before RCT approval.
They are uncharacteristically silent on this topic, avoiding debate (and these people love to debate). But here they appear to be exploiting authority bias of the audience, (ie they hope the audience will assume it’s not important or it would be debated by leaders given their known penchant for debate). It’s an old trick to protect dogma. It won’t work.