Jose de Leon: Training psychiatrists to think like pharmacologists
27. Evidenced-based versus personalized medicine
Donald F. Klein’s response to Jose de Leon’s reply
That Dr. De Leon positively appreciates my work is terrific. The appreciation of informed peers is among our greatest professional rewards. Somewhat shameful, I have to admit that it is obvious that my getting better acquainted with his work is clearly indicated.
In our discussion of personalized and evidence-based medicine, it seemed that De Leon’s major point is that current statistical approaches emphasize main effects and ignore the outliers that he states are central to personalized medicine. De Leon's approach emphasizes pre-test risk stratification by pharmacological understanding. Each outlier is thereby given full attention. However, does this mean that each pre-study risk requires its own independent sample for testing its relationship to some specified outcome?
I need some statistical help, since I made the mistake of definitely claiming ANCOVA had superior power as ANCOVA can apply multiple predictive (equals risk) binary covariates to the same sample. Also, pharmacological pre-stratification using the concepts of hyper- and hypo-metabolism via defined routes has not impressed me by its wide clinical or academic adoption...
On reflection, both approaches suffer from developing an inclusive, correct list of pre-study risks. Unfortunately, there may be many other, perhaps more important risks that we are ignorant about. The use of pharmacological variables for pre-study stratification does not convince me but it is the data that counts, and so far, there has not been enough studies with clinical relevance. A tabulation emphasizing demonstrated human relevance would be helpful.
In principle, ANCOVA does use binary covariates to deal with pre-study stratification. Unfortunately I have no large experience with this approach and have not pursued the extensive, difficult, discussions. My belief that a possible advantage is that ANCOVA may apply many covariates to the same amount of data, is more intuitive than substantive…
Further, are outliers necessarily helpful? I have suggested that many syndromes, especially those that both exacerbate and remit, are due to impairment of a stabilizing, negative feedback, central function by a host of minor saboteurs. Our most useful correlates would be those that cast light on how the stabilizing mechanism interacts with the saboteurs. That is a far cry from some outlier relationship to some aspect of outcome, as GWAS does.
Perhaps more important, neither method gives a narrowly personal insightful description. They may constrict the useful predictive set regarding some aspect of outcome but likely, there may remain considerable multivariate variance within the positively helped outcome set. I think hypotheses, derived from surprising clinical observations are called for. These may yield useful tests that deal with these complexities.
That Dr. De Leon and I agree about the antipathy of funders and publications to the logical necessity of replication is a real pleasure. This emphasis is simply ignored by those in a position to effectively support these efforts.
I did read Dr. De Leon’s editorial and discussions of the recent industrial subversions of the possibilities of both PM and EBM by suppression of negative data.
We need a large informed general discussion of how to deal with this public health menace. We must impact legislation by public outcry. Trial pre-registration attempts to deal with specific systemic flaws but so far has been largely ignored and has not engendered the promised negative fines.
My completely unrealistic Utopian vision is that the entire process of pharmacological discovery and production be taken out of the hands of private industry and given to a new institution within NIH. Unfortunately, this would also amount to totally depriving industry of its profits. They would effectively destroy such a notion. Can anyone figure out an alternative that might actually come to pass?
Donald F. Klein
December 1, 2016