Hello,
I've recently stumbled upon this:
https://github.com/mattrosen/rfSLAM
Article Clinical risk prediction with random forests for survival, l...
It's a rather novel approach to random forests in survival analysis, not relying on the PH like some others do.
My concern lies within the usage of their so called CPIU's. In the limited documentation available it states:
"Each individual can have many CPIUs during the period of follow-up. In the example sudden cardiac arrest (SCA) risk prediction problem, we consider follow-up time of 8 years and specify the time intervals to be 6 months long so that each individual has a CPIU representing each half-year of observation."
My dataset, alas, does not have a fixed follow-up time. The follow-up time varies from several hours to multiple weeks. With the focus lying around 3 days. Would this still be a reliable model for such a dataset?
Also in https://github.com/mattrosen/rfSLAM/blob/master/utilities/make_into_cpiu.R
Which describes how to turn your dataset into cpiu, they keep referring to death. I do, however, not have such an event. Furthermore, one patient may have multiple events.
Would it be interesting to use RF-SLAM, or should I use
https://cran.r-project.org/web/packages/LTRCforests/LTRCforests.pdf
Kind regards,
Matt
- Badges
- Science topic