HMS Strategic Planning |
|
Comments and Discussion
January 31, 2008
HMS faculty member (name withheld by request) Evidence-based Clinical Medicine and Research Informs Results Published clinical data continues to challenge students, residents, fellows, and faculty because they don't have the skills and/or the time to identify problems with bias, subject risk level, confounding, and external validity. Moving the students past deferential and categoric thinking to consideration of the distribution of the data, would, in the long run, help them to decide whether the inference drawn was valid. Whether you want to admit it or not, all of our work is based upon a model; being able to interpret a p-value in a context is increasingly important as the sheer volume of good and bad data and models grows. In my three years of running the journal club for the Neonatology Fellows, skill at being able to assess for a fair comparison remains highly variable and all too often devalued. The upshot is that we are still training a generation of physicians who too often look to their colleagues for practice decisions. (Publication bias lives on because we don't police our own research work as well as we should.) Many journals, for example, gloss over p-values when simple summary statistic calculations often uncover trends in p-values that by themselves, or in conjunction with other baseline covariates, can threaten the validity of published RCT's. Third-order selection bias and model selection criteria (efficiency criteria) should be in our vocabulary as much as number needed to treat and the pervasive p<0.05 You can send a comment by clicking here.
|
|