I absolutely adore your intro!!!!!!!!!! My students frequently conflate causation and correlation. One of my replies: Just because you had a stomachache on the night of a full moon does not mean that the full moon caused your stomachache. It is a silly example, but it does seem to help get the point across. As you so eloquently point out, though, literature can blur the line between simple descriptive information and causation.
As a hematologist I applaud this with all hands I can find. I would take it one level further -- in today's electronic world, there is no reason not to provide the detailed information (not the medical record -- just the information on which the "conclusions" are based) on ALL patients -- why restrict it to 100? There would have been no Surgisphere if this had been done, and there is no reason not to do it starting tomorrow. The current NIH push for reproducibility would militate strongly toward doing this -- the re-analyses, new follow-on experiments, etc. would be of deep value to all including the original authors.
There is a well deserved crisis oi publication because of exactly the kinds of jiggery-pokery (intentional or not) much current writing supports. (This oft-called reviewer notes this when contrasting his reviews with those of other reviewers -- the differences are often striking.) Providing all the data all of the time (appropriately pseudonymized) would resolve much of the discredit publications are currently, and often deservedly, taking and would improve the reproducibility (or demonstration of lack thereof) and overall quality dramatically.
There is no shame in starting with observation. But it would be shameful to end there (at least if one were a clinician scientist, who really ought to know better). And we are not in the Middle Ages anymore where one could defensibly confuse correlation with causation.
As for causal inference from observational data, I would fall back onto the Bradford Hill criteria as a starting point.
As the quip goes: some observational studies are correct; we just don’t know which ones.
This is a wonderful article with clear recommendations for the future.
If I could, I would reproduce this piece (with attribution) with “Neonatology” replacing every mention of “Hematology”, and distribute it to our trainees and junior faculty. Ninety percent of neonatology “research” falls into the almanac category.
"The survival benefit persisted on multivariable analysis after exclusion of two inconvenient subjects." Makes me wonder if the the 'inconvenient subjects' may have inconveniently died?
I absolutely adore your intro!!!!!!!!!! My students frequently conflate causation and correlation. One of my replies: Just because you had a stomachache on the night of a full moon does not mean that the full moon caused your stomachache. It is a silly example, but it does seem to help get the point across. As you so eloquently point out, though, literature can blur the line between simple descriptive information and causation.
As a hematologist I applaud this with all hands I can find. I would take it one level further -- in today's electronic world, there is no reason not to provide the detailed information (not the medical record -- just the information on which the "conclusions" are based) on ALL patients -- why restrict it to 100? There would have been no Surgisphere if this had been done, and there is no reason not to do it starting tomorrow. The current NIH push for reproducibility would militate strongly toward doing this -- the re-analyses, new follow-on experiments, etc. would be of deep value to all including the original authors.
There is a well deserved crisis oi publication because of exactly the kinds of jiggery-pokery (intentional or not) much current writing supports. (This oft-called reviewer notes this when contrasting his reviews with those of other reviewers -- the differences are often striking.) Providing all the data all of the time (appropriately pseudonymized) would resolve much of the discredit publications are currently, and often deservedly, taking and would improve the reproducibility (or demonstration of lack thereof) and overall quality dramatically.
Great article. Thanks.
There is no shame in starting with observation. But it would be shameful to end there (at least if one were a clinician scientist, who really ought to know better). And we are not in the Middle Ages anymore where one could defensibly confuse correlation with causation.
As for causal inference from observational data, I would fall back onto the Bradford Hill criteria as a starting point.
As the quip goes: some observational studies are correct; we just don’t know which ones.
This is a wonderful article with clear recommendations for the future.
If I could, I would reproduce this piece (with attribution) with “Neonatology” replacing every mention of “Hematology”, and distribute it to our trainees and junior faculty. Ninety percent of neonatology “research” falls into the almanac category.
Well done.
Thanks for this post - I now better understand almanacs and medical science.
"The survival benefit persisted on multivariable analysis after exclusion of two inconvenient subjects." Makes me wonder if the the 'inconvenient subjects' may have inconveniently died?