James Deardorff of UCSF cautions that the statistical performance of a prognostic model says nothing about what is done with it at the bedside of an elderly patient.
In a commentary published on September 18, James Deardorff, a geriatrician and assistant professor in the Division of Geriatrics at the University of California, San Francisco, invites us to distinguish between two issues that are often confused: the quality of an algorithm and the quality of its use. A model capable of predicting clinical outcomes does not guarantee good results for patients.
His argument is based on the analysis, published in JAMA Network Open, of Epic's end-of-life prediction model. The same tool, he points out, does not carry the same weight at all depending on whether it is used to open a discussion on care objectives or to prioritize transplant candidates. The issue then shifts from statistics to the organization of care.
Finally, he insists on a common blind spot: clinicians must know the performance of an algorithm on each subgroup of patients, and not just its overall accuracy. This is particularly true in geriatrics, where the most vulnerable populations are often underrepresented in training data.
Read the full article on the original source
Go to the source →