Hi Carlo,
Thank you for your reply. Sorry, I didn't describe the problem clearly.
I understand the estimation can be done, but my worry is that without any treatment, the estimated coefficient would be biased.
X is an indicator variable and it changes over time.
Let's suppose X=1 can prolong people's life (think of a drug).
And different patients are treated with X=1 at different time (they will only be treated drug when they are found to be at risk).
My dataset is right truncated at 2008.
I can observe everyone who have failed in the past when there was no drug.
I'll observe all patients who are treated (X=1) but failed by 2008.
However, I don't observe any patients who are at risk with X=1 and live beyond 2008.
In this case, estimating the impact of X, would most likely be estimated downward, since we don't observe the on-going cases.
This is probably not called "left truncation", but I can't find a better term describing the problem.
Is there way, we can make an adjustment to the coefficient to correct the bias in Stata?
Or is there any paper that addresses this issue?
Thank you all.
I just saw Antoine and Alan's replies after I completed the email.
Elaine
-----Original Message-----
From: Carlo Lazzaro [mailto:[email protected]]
Sent: Friday, October 23, 2009 3:57 AM
To: [email protected]
Cc: Liu, Elaine
Subject: R: Problem with Left Truncation
Dear Elaine,
Please find beneath the following point-to-point comments about your query:
<We are doing survival analysis, but unlike other dataset, our dataset only includes observations that have failed.>
I would not be concerned about all failure=1; how long patient takes to failure (failure time (tn)- risk onset (t0)) it's the relevant issue.
<Once it fails, the dataset would provide detail information on the date one starts to be at risk, when it fails, some other individual
characteristics(X') at the entry time.>
My suspect is that you are dealing with a retrospective survival analysis (ie, your dataset moves from death to risk onset).
If you have both t(0) and t(n), what's the matter? You have simply to -
stset- your data before performing survival analysis.
<Our goal is to estimate the impact of X on the probability of survival>.
Hence, the choice is between semiparametric (Cox regression) -stcox- and parametric -streg- survival models, provided that your dataset fulfills some requirements (eg. proportional hazard assumption in Cox model).
For further details on survival analysis topics, I will recommend you to take a thorough look at:
Klein JP, Moeschberger ML. Survival Analysis. Techniques for Censored and Truncated Data. Second Edition. Berlin: Springer, 2003.
Cleves MA, Gould WG, Gutierrez R. An Introduction To Survival Analysis Using Stata. Revised edition. College Station: StataPress, 2004;
Mario Cleves, William Gould, Roberto Gutierrez, and Yulia Marchenko (2008) "An Introduction to Survival Analysis using Stata". College Station: Stata Press.
[ST] Stata manual. Survival analysis and epidemiological table. Release 9
Two other relevant contributors of the Statalist - Maarten Buis
(http://home.fsw.vu.nl/m.buis/) and Stephen Jenkins
(http://www.iser.essex.ac.uk/teaching/degree/stephenj/ec968/index.php.)
published really interesting papers as well as teaching-notes on the topics you are interested in.
HTH and Kind Regards,
Carlo
-----Messaggio originale-----
Da: [email protected]
[mailto:[email protected]] Per conto di Liu, Elaine
Inviato: giovedì 22 ottobre 2009 20.26
A: [email protected]
Oggetto: st: Problem with Left Truncation
Dear Statalist readers,
I have a question regarding the use of survival analysis with a problem similar to left truncation.
We are doing survival analysis, but unlike other dataset, our dataset only includes observations that have failed.
Once it fails, the dataset would provide detail information on the date one starts to be at risk, when it fails, some other individual
characteristics(X') at the entry time.
Our goal is to estimate the impact of X on the probability of survival.
I think it's a common problem in medicine (for example if you are estimating the probability some event causes death but you only observe people after they died)
I have checked several posts in the archive and the textbook solution to left truncation, but they don't seem to address the problem.
This is my first time posting in this community. Let me know if more information is needed.
Thank you very much.
Elaine
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
