Table 3 Approaches of Competing Risks Flexible Parametric Models
Model | Measures of associations | What is the model useful for? | How to model? | Advantages | Disadvantages | Which Stata commands? |
|---|---|---|---|---|---|---|
Cause-specific hazard flexible parametric model | hazard | Etiological questions: which covariates have a causal effect on the occurrence of the event | CSHFPM | Easy to perform (on the original data) and interpret, using the standard FPM | Fitting separate models for each event | stpm2 |
| Â | Â | Â | Unified CSHFPM | Fitting one model instead of separate models, using the standard FPM, Ability to handle shared covariate effects | Considering the same knot positions for all events, complex implementation (on the stacked data), Potential convergence problems | Stratified stpm2 |
Cause-specific subdistribution hazard flexible parametric model | Subdistribution hazard and cumulative incidence function (risk) | Prognosis questions: What fraction of patients are at risk to experience the event at a particular time | SDHFPM1 | Fitting a separate model for the event of interest, using the standard FPM | Intensive computation (not ideal for large data sets), no constraint on the sum of CIFs | stcrprep and stpm2 |
| Â | Â | Â | SDHFPM2 | Fitting a unified model for all events (when the focus is on all events), Easy to perform (on the original data) and interpret, Less computation (ideal for large data sets) | convergence problems for small sample sizes, no constraint on the sum of CIFs | stpm2cr |