Case-Cohort Design - longitudinal ordinal outcome with interval-censoring

A Case-Cohort design was used to study several biomarkers and an uncommon outcome. The biomarkers were measured for all cases, as well as a random sample of the cohort, referred to as the subcohort (sampling without replacement, and no stratification). The outcome is actually ordinal (grade 0, 1, 2, 3, 4), but case status was defined as grade >= 2 for purposes of the case-cohort design/sampling. Baseline participant demographics and clinical variables were measured for all participants in the cohort.

In addition to relative risk, I would like to estimate the predicted probability of the outcome within T years conditional on biomarker values. We originally intended to use a time-to-event analysis. Can I use the Markov longitudinal ordinal models that you have presented? I’m not sure given that a) case status and therefore sampling was based on a dichotomized version of the ordinal outcome (grade >=2), and b) not sure how to incorporate the inverse of the sampling fraction as a weight to account for the Case-Cohort design. Otherwise I was hoping to use Weibull due to interval censoring in outcome assessment, but also for that model I don’t see how to incorporate the inverse of the sampling fraction as a weight.
Suggestions would be greatly appreciated!
Thank you.

A lot of work has been done for case-cohort designs with binary time to event outcomes. I haven’t seen work done with more general outcomes but there may be a paper that addresses this for an ordinal outcome measured at one time.

1 Like

I have seen R packages for case-cohort designs with binary time to event outcomes, interval-censoring, and predicted cumulative incidence, and even two out of the three. But I haven’t found one that can do all three.

Full wish list includes:

  • Case-cohort design
  • Interval censored outcomes
  • Predicted cumulative incidence conditional on covariate combinations
  • Penalized maximum likelihood estimation or other approach to adjustment for overfitting
  • Evaluation of time-dependent discrimination and calibration

If anyone is aware of such a package please let me know. Even the first 3 would be great.

1 Like