RMS Semiparametric Ordinal Longitudinal Model

Dear Professor Harrell, I am a physician using individual participant data from RCTs in metastatic prostate cancer. I plan to analyze two longitudinal ordinal outcomes, skeletal-related events (SRE) and ECOG performance status, with a Bayesian transition model. Both outcomes are recorded at each scheduled visit. SRE is conventionally a composite endpoint, the first occurrence of palliative radiation to bone, pathologic fracture, surgery to bone, or spinal cord compression. My main aim is to estimate the effect of a drug on SRE and ECOG performance status. That drug is given as concomitant therapy and is not randomized. I would value your guidance on how best to define these two endpoints. I want to estimate the drug’s effect on each SRE component, so I redefine and order the components as follows: 0 = alive without SRE 1 = palliative radiation to bone 2 = pathologic fracture 3 = surgery to bone 4 = spinal cord compression 5 = death ECOG performance status, as a reversible scale 0, 1, 2, 3, 4, with death as the absorbing top state. My main concern is the SRE endpoint. Unlike ECOG performance status, which changes gradually so that transitions are mostly between adjacent levels, the four SRE components are discrete clinical events rather than levels of a graduated severity state. If I treat them as transient, a patient who experiences any of levels 1 to 4 returns to 0 afterward, which produces abrupt and often non-adjacent movement on the scale, for example 0 to 4 and back to 0. This does not resemble the smooth trajectory of ECOG. I am therefore unsure whether an ordinal longitudinal state representation suits the SRE endpoint at all, and, if it does, whether the events should be cumulative or transient, and how to define the scale accordingly. I lean toward the transient coding because it can capture recurrent events. I would also welcome your view on whether the SRE endpoint is better served by a multistate model on the event times, while ECOG is kept in the ordinal transition framework. Thank you very much for your time.