Making inferences about covariate effects for drug labeling is not a trivial challenge. We’ll explore the problem, review methods, and present some possible solutions at ACoP8 (Monday Oct. 16, 2017, 10am - Session 1a: Quantification of covariate effects for labeling). Whether or not you’re able to join us in Ft. Lauderdale, we invite you to join in the discussion here before or after the conference session.
Thanks for your nice presentation at ACOP8.
Indeed we still have a long way to optimize and communicate our covariate models.
One of my attempts is located here: https://pharmacometrics.shinyapps.io/interactiveforestplot/
code is on github: https://github.com/smouksassi/interactiveforestplot.
One of our recurrent problems is that we talk about between subject variability in terms of precentage what does it mean to have a BSV of 30 % ? as a pharmacist I prefer to think about how much percent of my patients will fall say between 0.8-1.2 or between 0.5-2.0. Or alternatively and instead of reporting a single value like 30 % we can report where 50 % of patient lies with BSV=30 % this will be : 0.82-1.2 and then where 90 % of my patients will be: 0.61-1.7.
This helps me understand whether my patients risk to be outside of therapeutic limits because of random unexplained variability. It will be good to keep this discussion live and see how we can have better labels that really help the clinicans in making better decisions.