Does anyone know of a paper about using EHR events as proxy state for a multi-state recurrent event survival modelling approach?
Thank you
you mean just how to do recurrent events analysis on these data? this might be useful: https://pubmed.ncbi.nlm.nih.gov/24453096/ “We describe methods of analysing repeat hospitalizations, and illustrate their value in one major trial.”
1 Like
That is an excellent paper on recurrent events.
A general approach is longitudinal ordinal outcome modeling. This can, for example, consider death as more serious than hospitalization, allows for death as a terminating event, and counts multiple hospitalizations. It provides a simple estimand: as a function of time what is the probability of hospitalization or death and what is the probability of death. See here for more information. This approach can also make use of partial information. For example if in a given week you did not know whether or not the patient was hospitalized but you knew she was alive, and you coded home/hospital/death as Y=0,1,2, Y would be left censored at Y=1, i.e., we just know that Y <= 1 for that week.
When a patient is hospitalized multiple times, this elevates the overall estimated probability of being in the hospital at a given time. We model an overall time effect using e.g. a restricted cubic spline function.
1 Like
thanks for that. It also reminded me of a SiM paper i think you mentioned once: Analysis of failure time data with ordinal categories of response It’s quite relevant for an analysis im looking at right now
1 Like
Yes that is a cool paper when there is a single outcome and the ordinal scale measures the severity of that one outcome.