SAS 生存分析截尾数据的处理

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

各位大神请教一个问题，这是我们的project里的生存分析的问题。
在t0 到t30间以0 和1 记录事件是否发生。 o3为事件第一次发生的时间点。 比如o3=15，代表t15的时候第一次发生该事件。 那么如果一个事件从0到30都没发生， 应该如何处理？
这应该属于type one right censoring，但是难在所有数据我只知道是大于30的缺失，而我查到的所有资料都是右截尾数据的值是不同的，比如264+， 220+。我觉得这种情况应该还比较普遍啊，比如实验在t=30的时候结束，有的数据就是没发生event。。。
如果我要用censor，那么就必须给o3赋值，这个值我不知道，我只知道是30+
如果不赋值，那么就自动处理为missing，censor就不能用了。我原来有100个数据，missing 后只剩下77个。missing处理的话，代表这部分信息的缺失吗？ 用PHREG和LIFEREG有处理的办法吗
本来我给所有missing赋值为31， 做出来的残差图有明显的趋势，对于所有O3=31的点拟合有偏，所以是应该直接drop这些点吗？？

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：截尾数据 生存分析 missing lifereg Project 记录 如何 资料

No worry about that.
It is very typical for survival data, such as fixed observation window or end of study, etc.
if no event at time t30, just set it as censored. Remember that S(infinite time) = 0. that is, all should experience event, sooner or later. Censored = (event time > observed time). Missing = unknown status. But we do know that the events do not occur to them.
Jingju

jingju11 发表于 2014-5-5 13:20
No worry about that.
It is very typical for survival data, such as fixed observation window or end  ...

Thank you for your reply, but if I set them as censored, I have to set a value for these points. They can not be".", that will be ignored. But if I choose a value, such as 31 or other value larger than 30, it will influence the output and residual plot. there is an obvious trend in the residual plot, formed by all points valued 31, is that Ok?? It looks so strange so I really got confused. I never find a data like this before.

I may not fully understand your question. But say,

You had observed subjects periodically for maximum 31 times, and noted as t0, t1, ..., t30.
for subject 1, at t15 you observed an event on him/her. so, you have a record like

subject=1, time=15, event=1

for subject=2, you followed him to the end of the study (at t30) and no event observed. you have a record like:

1. subject=2, time=30, event=0

复制代码

for basic survival analysis, the data like above is enough, that is, 3 variables subject, time and status(event).
for example, in KM method in sas you wrote it as

1. proc lifetest...; time time*event(0); ...; run;

复制代码

see, that is, event =0,  censored. Remember that, in survial we only consider two statuses of censored or not censored(event).
Here variable 'subject' may not be required. But when you had  multiple records from one subject,you may think that, between subjects, observations are independent; but not within subjects. for example,  time-dependent covariates, or multiple events, etc. Those need special stastics treatment.
JingJu

请问楼主最后是怎么解决这个问题的？

求问怎么解决