Hello, I’m a health economist by training, but some of my PhD was a bit more biostats-heavy (prediction modelling, mostly). I was hoping to get some advice on potentially improving upon some of the methods used in health economic modelling, typically either Markov models or discrete-event microsimulations, in cases where you have relatively rich line-by-line patient data. Apologies in advance if this isn’t very well articulated - I’m still feeling my way around the problem.

As a toy example, let’s say you want to model the progression of some cohort through X cancer diagnosis in Y health system, and produce generalised estimates of their costs and survival outcomes. The goal of the analysis would be to produce estimates that can be easily extrapolated to the broader population, of whom you assume that your data constitutes a random sample. The data might be formatted such that each line constitutes an interaction of the patient with the health system, with an associated timestamp, cost, and state (e.g. Stage I, Stage II, remission, death, and so on).

Most methods I have seen in this area tend to use Monte Carlo methods to sample from distributions for model input parameters, namely costs and state transition probabilities, but I thought perhaps this is where a Markov Longitudinal Ordinal Model (as per https://hbiostat.org/proj/covid19/ordmarkov.html) could be a bit more helpful. Monte Carlo takes random draws from input parameters, which seems like it might needlessly inflate the variance of your final estimates. A way to share information across state transitions based on individual covariates seems like a significant improvement to existing methods.

I guess my question is, how might you envisage this approach working? Is it a misuse of the longitudinal ordinal model? Are there any resources you might suggest to get me started? Thanks in advance.