I have an observational study where I want to compare 2 groups (A versus B) on whether they tend to get diagnosed with lower or higher stage cancers. For each patient, they have one outcome: the time until they are diagnosed with a stage T1, T2, T3 or T4 cancer, or they are right censored at last available followup time. So basically I have 4 mutually exclusive time-to-event outcomes that follow a natural order, where higher stages are more severe cancer.
My understanding is that the ordinal transition model is a perfect approach for this type of data because it can estimate a single treatment effect (proportional odds ratio) to answer my question of whether group A tends to have higher/lower cancer stages compared to group B.
Questions:
- Importantly, I want to use Age as the underlying time scale, so “time 0” would be the time a patient is born. However, I need to properly account for “age at study entry” using this approach. If fitting a Cox model, the way to do this is using left-truncation. Is it possible to do something similar with an ordinal transition model?
- Does anyone have R package recommendations or example code for implementing an ordinal state transition model with multiple time-to-event outcomes similar to my setting? I think Dr. Harrell has recommend using VGAM (frequentist) or rmsb (bayesian), but I’m having a hard time finding good example code to get started with these approaches for a context similar to mine.