俺也关心这问题哟，翻过一些老外的资料，有的资料基本上都是截尾数据，可老外照做生存分析，难道就没个限制？坛子里的牛人出来说说？

没看到哪本书明确说多少，但截尾数据过多，显然对统计结果是有不利影响的

肯定会有影响的 但没有看到过明确的说法

貌似没有很明确的说法，但是如楼上di所说，删失数据太多肯定不是很好啦。假如是由于失访造成的话，从队列研究的角度去看的话，（记得上课的时候老师们说的是）一般要求失访率不高于10%。如果是由于研究终止造成的话，不太好说，有可能由于观察的时候不够充分，也有可能是某种治疗本身效果比较好。不知我说的对否？

Well, if the truncated data are more than 80%, the whole analysis is meaningless. The main stream way to do the truncated data and censored data are quite stupid, actually. Either they throw away those nondetect data, or use the averange number of the interval instead of the truncated data. Also you could write a macro yourself and run it into SAS. So far, there are a few procedures like LIFEREG, RELIABILITY in SAS are dealing with nondetect data. I think there are a few packages work in R as well too. But because the difficulties of how to treat the trunceted data are so huge, most statisticians in the U.S. would choose the main stream ways which is pretty sad.