[Jones 2000](https://pubmed.ncbi.nlm.nih.gov/11143786/) traces an interesting history of cardiology as it hosted one of the decisive contests between 'evidence based' and 'clinical' medicine during the adoption of CABG surgery. EBM proponents argued for the primacy of _evidence from RCTs_ of 3-year survival, while clinical proponents got behind _visual evidence of mechanistic effects_ from angiography with the logic:
blood flow = health
no blood flow = disease
restored blood flow = cure.
The RCT is interpretable only as evidence about average outcomes (would the study subjects have benefited on average if everyone were treated), while the mechanism is evidence of success in individual cases (was *this* patient's blood-flow restored by the procedure). In the care of a specific individual, it's not clear that survival should be the primary target, since quality-of-life could matter more for some patients depending on their preferences, which (unfortunately for rationalists) may be impossible to quantify/elicit. If restoring flow benefits both survival _and_ quality-of-life, and is accessible on a per-patient level, then the clinical heuristic is arguably more important than the results of the RCT. On the other hand, if restoring flow benefits neither, why does CABG work at all; it seems like any efficacy has to pass through this mechanism.
RCTs work formally, but it seems like they are frequently insufficient at producing knowledge about actions in reality. Understanding these shortcomings will help us to understand in which settings decision making can (should) be automated effectively / ethically...
Having the results of the randomised trials to hand, we then apply the average result to the individual patient in front of us. But couldn't that average outcome be comprised of patients who were harmed as well as helped by the treatment? Why should the average outcome - which is presumably experienced by few individual trial participants - apply to my patient? (Notwithstanding that my patient's characteristics differ in known & unknown ways from the trial participants, but I see that as a different issue).
While we're at it, what about hearing that your coronary stenting procedure has a 1:1,000 chance of a serious complication - how does the certainty-craving human brain use that information - that you're unlikely to have a complication but if you do it'll be life-changing - to inform the dichotomous decision as to whether to proceed with treatment?
The broader conundrum is, how can the human brain best use statistics and probabilities to inform individual decisions?
[Jones 2000](https://pubmed.ncbi.nlm.nih.gov/11143786/) traces an interesting history of cardiology as it hosted one of the decisive contests between 'evidence based' and 'clinical' medicine during the adoption of CABG surgery. EBM proponents argued for the primacy of _evidence from RCTs_ of 3-year survival, while clinical proponents got behind _visual evidence of mechanistic effects_ from angiography with the logic:
blood flow = health
no blood flow = disease
restored blood flow = cure.
The RCT is interpretable only as evidence about average outcomes (would the study subjects have benefited on average if everyone were treated), while the mechanism is evidence of success in individual cases (was *this* patient's blood-flow restored by the procedure). In the care of a specific individual, it's not clear that survival should be the primary target, since quality-of-life could matter more for some patients depending on their preferences, which (unfortunately for rationalists) may be impossible to quantify/elicit. If restoring flow benefits both survival _and_ quality-of-life, and is accessible on a per-patient level, then the clinical heuristic is arguably more important than the results of the RCT. On the other hand, if restoring flow benefits neither, why does CABG work at all; it seems like any efficacy has to pass through this mechanism.
RCTs work formally, but it seems like they are frequently insufficient at producing knowledge about actions in reality. Understanding these shortcomings will help us to understand in which settings decision making can (should) be automated effectively / ethically...
All to say: we should fund the humanities.
I like how you stuck the landing!
Having the results of the randomised trials to hand, we then apply the average result to the individual patient in front of us. But couldn't that average outcome be comprised of patients who were harmed as well as helped by the treatment? Why should the average outcome - which is presumably experienced by few individual trial participants - apply to my patient? (Notwithstanding that my patient's characteristics differ in known & unknown ways from the trial participants, but I see that as a different issue).
While we're at it, what about hearing that your coronary stenting procedure has a 1:1,000 chance of a serious complication - how does the certainty-craving human brain use that information - that you're unlikely to have a complication but if you do it'll be life-changing - to inform the dichotomous decision as to whether to proceed with treatment?
The broader conundrum is, how can the human brain best use statistics and probabilities to inform individual decisions?
I love that last question and will do my best to at least give partial answers over the coming weeks.