y**3
发帖数: 267 | 1 Just got a project to build a predictive model of survival data. There are
some cases of multiple events. Can anybody recommend the most used methods
to model repeated events survival data? In addition, this is a rare event
data. Does SAS e miner takes care of repeated events and/or repeated rare
events?
Thanks a lot |
Y******Y
发帖数: 8753 | 2 包子please!
recurrent event model,比较多人用的有几个
1） Anderson-Gill的counting process model,假设recurrent event之间是独立的
2） PWP(Prentice-William-Peterson) model: 这个分两种，一个是gap time model,
一个是total time model
3) WLW (Wei-Lin-Weissfield) model,这个是marginal model
SAS的proc phreg就可以fit这三个model,只要你把data准备成符合sas的counting
process格式。
具体code看这里
http://www.ats.ucla.edu/stat/sas/faq/survival_repeated_events.h |
A*******s
发帖数: 3942 | 3 再加一个shared frailty model
无非就是如下model的censored data版本
fixed effect
stratified model
robust sandwich
random effect
【在 Y******Y 的大作中提到】
: 包子please!
: recurrent event model,比较多人用的有几个
: 1） Anderson-Gill的counting process model,假设recurrent event之间是独立的
: 2） PWP(Prentice-William-Peterson) model: 这个分两种，一个是gap time model,
: 一个是total time model
: 3) WLW (Wei-Lin-Weissfield) model,这个是marginal model
: SAS的proc phreg就可以fit这三个model,只要你把data准备成符合sas的counting
: process格式。
: 具体code看这里
: http://www.ats.ucla.edu/stat/sas/faq/survival_repeated_events.h
|
y**3
发帖数: 267 | 4 Thanks! Will find out if i have Boazi :).
Another specific question:
A patient is on a drug, from date of entry to the date of last fill, there
are time gaps between two refills. that is , the patient is not taking the
drug all the time during the time period. IS this a problem for survial
analysis? IF so, how to deal with it? ( I will medol the Jump of drug levels
) |
Y******Y
发帖数: 8753 | 5 这个drug是作为covariate? 那你把它当作time-dependent covariate来model吧。 最
简单的就是create on/off drug indicator. 反正你的data都要变成counting
process format的。 这个很容易做到。就是在某一段时间这个indicator=1, 下一段时
间这个indicator=0, 在下一段又变成了1， sth like that. |
y**3
发帖数: 267 | 6 thanks again
A chemical level in the drug is the dependent variable. I model the jump
from one level to another leve, which is event. So multiple events. Now
another question, the events types are different as well. For example, the
first jump is from level 1 to 2, another jump is from 2-3 then another jump
is 1-3. Which of the methods for repeated events also takes care of this? |
Y******Y
发帖数: 8753 | 7 check out competing risk model.
another way to do this is Cox-type markov or semi-markov multistate model.
jump
【在 y**3 的大作中提到】
: thanks again
: A chemical level in the drug is the dependent variable. I model the jump
: from one level to another leve, which is event. So multiple events. Now
: another question, the events types are different as well. For example, the
: first jump is from level 1 to 2, another jump is from 2-3 then another jump
: is 1-3. Which of the methods for repeated events also takes care of this?
|
y**3
发帖数: 267 | 8 Many thanks!!!
For competing risk model, I know how to do for single event. But for
multiple events with different types within one record, not sure how? It
seems that sas cant do this. There could be R package?? 楼上大侠能给具体指点
一下吗？
Maybe， to simplify, maybe I just use the first event, then do a competing
risk???
Never knew Cox-type markov or semi-markov multistate model.will check them
out. |
A*****r
发帖数: 795 | 9 Competing Risks/Multistate Models |
y**3
发帖数: 267 | |
