I have two compartmental PK model. Can somebody suggest how to handle missing data? For example missing dose, missing/erroneous concentration, erroneous/missing time, etc.
Thanks for any suggestions.
1 Like
Hello, I’d say this depends on the context and what you’re trying to achieve. Missing observations (samples not collected) are generally not a validity concern in pharmacokinetic modelling, except that they may cause identifiability constraints. The exact time of any observation is ultimately arbitrary with respect to the underlying model.
For missing dose or time-of-dosing, with enough surrounding information in a hierarchical model it may be achievable to estimate these (by defining them as parameters), but it might be difficult to have enough information in practice. For example, in estimating the size of an oral dose, even given the exact time it occurred, you would need an independent clearance estimate for that subject (from an intravenous dose), a strong prior for bioavailability, or similar information. The same principle would apply to errors in predictors.
1 Like