{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreif3nuj234vxjsu4f3g2qivydbfxwxmtzljrwqlthgrpeby5hv4qgq",
    "uri": "at://did:plc:wwyqal4cnqhuwyacdj7rqq3n/app.bsky.feed.post/3mmp5f27l6ny2"
  },
  "path": "/t/rms-semiparametric-ordinal-longitudinal-model/4819?page=6#post_120",
  "publishedAt": "2026-05-25T12:31:17.000Z",
  "site": "https://discourse.datamethods.org",
  "textContent": "What a wonderful setup and question. A lag-1 Markov process would model the previous SRE state as categorical using 4 indicator variables, so it allows for large jumps. But the different types of endpoints comprised in the ordinal scale may possibly have different risk factors or different weights for risk factors, making it harder to model as proportional odds. Partial proportional odds may be needed for certain covariates.\n\nBut you’ve raised a bigger problem which I’ve wondered about for heart attacks in a cardiovascular study. If a patient has a heart attack one day are they logically coded as back to normal the next day? How long does a heart attack last? I’ve only been able to figure this out in the context where the consequences of the heart attack are included in the outcome scale. For example, hospitalization might be considered as slightly worse than a heart attack not requiring hospitalization (which is rare) and as long as the patient stays in the hospital, the hospital status will override heart attack and give a reasonable amount of “badness” in the score. If the patient has to go into intensive care, then even worse. See if this gives you a direction to try.",
  "title": "RMS Semiparametric Ordinal Longitudinal Model"
}