I thought I should point out the exact problems with John's second
alternative. First, by putting all events at the end of the censoring
interval, it will estimates of P(Y<t) will be less than the true
values, for any t. Second, the added censored observations will
inflate the denominators for risk calculations in prior intervals.
This will greatly shrink the estimates of P(Y<t).
-Steve
On Jan 29, 2009, at 8:13 AM, John Stephenson wrote:
I am attempting to do a parametric survival analysis with interval
censored
data. However, I don't know if Stata can handle interval censoring.
Does
anyone know how to do this using Stata? (I am working with release 9)
Alternatively, if it is not possible (or easy), two alternatives are:
1. find the midpoint of each interval and use that
2. assuming that each item of data is interval censored between
time T1 and
some later time T2, duplicate each item so that the same item
appears both
as a positive outcome at time T2 and as a right censored outcome at
T1.
If anyone has any views on the accuracy and applicability of either of
these methods I'd be grateful to hear them.
*
* 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/
