I have health records of immunodepressed patients who may have event histories like [high risk demographics] -> [low lymfocyte count] -> [high viral load] -> [clinical events]
From those data I would like to develop models for the risk of [low lymfocyte count], [high viral load] and [clinical events], ultimately leading to decision support for clinicians: how frequent should each patient be monitored for [lymfocyte count] and/or [viral load]? Which patients should receive profylactic treatment due to high risk of [clinical events]?
One problem with this is that the histories of [lymfocyte count] and [viral load] are patchy as patients get those tests done when physicians think they are needed. Those variables play the roles both as predictors and as outcomes in various models I would like to make.
I could create a balanced analysis set where patients with scarce lab tests had test results imputed using MICE, but I am woried that MICE wouldn't be valid as we have informative holes in the lab testing histories. I could also create models for [lab testing propensity], maybe using future [clinical events] among the predictors, but I am not sure what to do with such a propensity model.
Any suggestions for litterature that describe methods I could use?
