search

scheike / timereg

timereg R package, survival analysis methods
29 stars 5 forks source link

comp.risk issue #6

Open mehlkopf opened 5 years ago

mehlkopf commented 5 years ago

Hello,

at first I want to say thank you for the great tools timereg provides! However, I am having a few problems using comp.risk to analyze competing risk models: I am using "aalen" to run an additive model with an survival object of type Surv(time0, time, failure), and it works fine. However, when I am using the exact same model with comp.risk, I have to switch to Event(time0, time, failure_type) - and this comp.risk model does not converge (neither for "prop" nor for "additive" models). It does, however, converge when using just an Event of type Event(time, failure_type). Is this some type of bug or are there any real issues preventing the competing risk model from converging? Thank you very much!

Best, Andreas

scheike commented 5 years ago

for the competing risks model you need the Event(time,status) call !

You should never use the Event(entry,time,status) call ! Generally, regression models with delayed entry can be handled with weighting as in the Geskus paper.

Any lacking convergence means that the model apparently is hard to identify, and this is typically due to non-parametric components.

From: mehlkopf [notifications@github.com] Sent: Wednesday, February 13, 2019 3:12 PM To: scheike/timereg Cc: Subscribed Subject: [scheike/timereg] comp.risk issue (#6)

Hello,

at first I want to say thank you for the great tools timereg provides! However, I am having a few problems using comp.risk to analyze competing risk models: I am using "aalen" to run an additive model with an survival object of type Surv(time0, time, failure), and it works fine. However, when I am using the exact same model with comp.risk, I have to switch to Event(time0, time, failure_type) - and this comp.risk model does not converge (neither for "prop" nor for "additive" models). It does, however, converge when using just an Event of type Event(time, failure_type). Is this some type of bug or are there any real issues preventing the competing risk model from converging? Thank you very much!

Best, Andreas

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/scheike/timereg/issues/6, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AJvzwAT7vOzWtj5wZVbTtFegtYbE_AbGks5vNB0zgaJpZM4a5fJj.