Documentation: Enhancement to ADTTE vignette for complex parameters

[manciniedoardo]

Please select a category the issue is focused on?

Documentation

Let us know where something needs a refresh or put your idea here!

We've had feedback from Roche users that the ADTTE vignette does not cover enough examples for complex censoring rules, in particular that derive_param_tte() is not sufficient for some of their derivations. If working at Roche, see here, where eventually some discussion led to the proposal to use derive_extreme_event() in conjunction with various event() objects.

@bundfussr Could the ADTTE vignette could be enhanced to point users in this direction if they have similar use cases? Maybe make a new "Advanced" section mentioning that this is a possibility, with an example to showcase?

I'm not an expert in the TA so leave to you to figure out the details of what and how to showcase.

[bundfussr]

I think this is not an oncology-specific issue.

In most cases time-to-event analyses are done for negative events like death, progression, AE... The derive_param_tte() function and the vignettes were designed for such cases. However, the team which started the discussion wanted to derive a TTE endpoint for a positive event (time to first response). In this case the censoring rules are slightly different. I would suggest to update the TTE vignette in admiral and enhance derive_param_tte() if necessary.

@manciniedoardo , what do you think?

[manciniedoardo]

@bundfussr thanks for your response. Given your expertise and suggestions, how about the following:

• Update TTE vignette in admiral package as described above (ie. referencing possible uses of derive_extreme_event() for positive events)
• Link to that in TTE vignette in admiralonco (just a small section saying something along the lines of "If you find yourself needing to derive an endpoint for a positive event, eg. time to first response, check out the following example here)
• No updates to derive_param_tte()

[bundfussr]

I agree to update the TTE vignettes. However, I'm not sure if we should use derive_extreme_event() or enhance derive_param_tte(). I think this should depend on what is easier to use and read.

There is also another use case which is not covered by the vignettes. Setting EVNTDESC for censoring is sometimes tricky. For example if you are deriving time to CHG >= 10 and want to distinguish subjects censored because they don't have a baseline value and subject censored because they don't have post-baseline values. I think we should add something for this use case as well.

I would consider both use cases first and try out how they could be solved with existing functions or how they could be solved with enhanced functions. And then we could decide which way we want to go.

@manciniedoardo , does that make sense?