Written by a large bunch of great people, edited by Anna Heath, @NataliaKunst and me, and comes with an R package to use the methods at https://t.co/WxHIUVUY8C
...Nearly forgot to add, I'm doing a talk about the survextrap package next week at RSS Northern Ireland https://t.co/AWBV21PDr1 . Teams link https://t.co/fh3BAc3oPb
The package should also be useful as an easy way to fit a flexible Bayesian survival model in general, even if you don't need extrapolation. It is inspired a lot by rstanarm , and tweaks the model in https://t.co/OyaF2Orlfi in various ways to facilitate extrapolation.
Hoping people will try it and use it. Please feed back, in particular if you would like to use it, but something about it looks too hard! The idea is to make principled methods easily accessible.
You can (a) build in multiple data sources (e.g. trial, registry, population, elicited...) in a general format (b) fit models with a single command, (c) output results with a single command, in a friendly tidy format.
Announcing a new paper "survextrap: a package for flexible and transparent survival extrapolation" https://t.co/gpQCutCHJ2 . Proud of this, as I've been interested in the problem for many years, but only recently had time to get stuck into a solution. Tweet summary below.
The paper introduces an R package https://t.co/joidaieqL1 that does all this. It uses a flexible Bayesian model, fitted with https://t.co/Hi6venW62F. The "hazard" of death is allowed to vary over time using a spline, and can vary smoothly outside the data.
Ideal tool would be able to (1) combine several data sources (2) fit it all well (3) acknowledge uncertainty where data are weak. And not forgetting: (4) be easy to use!
Health policy decisions often involve "extrapolating" short term data from trials. To do this properly, we have to build in long term information. There's been lots of methods papers on the topic, but no ideal tool.
3yr postdoc job opportunity @MRC_BSU@Cambridge_Uni in Bayesian methodology
Potential topics/directions: computational methods for data integration/Markov melding; methodology for prior specification; methods for multi-stream EHR data
Closes 6 Nov
https://t.co/vBl96y0REP
If you like multistate / survival modelling you may be interested in this position https://t.co/QLZSd5xECX collaborating with some good people in a neighbouring department.
Coming to the end of an enjoyable three days of #RSS2023Conf . Too much good science and people to pick highlights, just generally feeling that stats and statisticians are alright!