Date: Mon, 14 Sep 2009 15:49:03 -0400
Reply-To: Peter Flom <firstname.lastname@example.org>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject: Re: URGENT - Kaplan-Meier curves,
95% CI and statistically significant difference
Content-Type: text/plain; charset=UTF-8
"cat.." <cat.b41@GMAIL.COM> wrote
>> I have limited experience in time-to-event data analysis. I requested,
>> in a statistical analysis plan for an observational study, to plot theKaplan-Meiercurves for 2 groups of patients on a single plot,
>> together with the 95% CIs at specific timepoints.
>> When 95% CIs do not overlapp at a given timepoint, is it correct to
>> draw the conclusion that disc rates are significantly different at
>> this timepoint ?
>> Reversely, when they overlapp, is it correct to infer that the disc
>> rates at that timepoint are NOT significantly different ?
>> Thanks for your feedback. This is quite urgent as I need to give my
>> feedback early this coming week.
>Has anyone an idea ?
Yes to the first question, no to the second question.
That is, if the 95% CI do not overlap, then p will be below .05; but p can be below .05 even if the CIs do overlap.
This was recently discussed, in some depth, on the MEDSTATS list; here:
or, in its original form
or google MEDSTATS, p, CI, overlap
the topic of the thread was "subgroup analysis"
Peter L. Flom, PhD
Website: www DOT peterflomconsulting DOT com