Re: Seeking suggestions for Power analysis in Survival Analysis

Posted by Rich Ulrich on Jan 01, 2020; 7:48pm
URL: http://spssx-discussion.165.s1.nabble.com/Seeking-suggestions-for-Power-analysis-in-Survival-Analysis-tp5738731p5738734.html

Questions. 

You don't specify any hypothesis.  Is this a two-group

comparison?  "time to new treatment" is an outcome, but

that just gives you a score.

Is the "time" measured by looking at change-in-treatments

1-5 as measured from "1"?  - Is there a date-measure, or is

the score simply the difference, one to four, from subtraction

of 2-1, 3-1, 4-1 ,5-1? 

How often will there be multiple outcomes possible for one

patient?  Or, should this be a survivorship analysis?

If your hypothesis does come down to a 1 d.f. test, you can

keep in mind that other power analyses for rather similar

analysis models will give similar results.  Consider the other

information that you will eventually want to report.

If there are two groups, and mostly just one outcome, the

most simple version of an analysis would be the t-test between

groups, comparing time-to-outcome for the first outcome.  I'm

not saying that that should be your analysis, but the power

considerations are intrinsically similar.  Multiple outcomes give

improvement in power.

That is, the simple power analysis on the t-test gives you a /maximum/,

estimate of sample size needed for a certain amount of power, where

the extra "new-treatment" scores per-person should result in slightly

better power (fewer N required) for the more sophisticated analysis.

Or, if there are /many/ changes in treatment, perhaps you could

consider the t-tests between two groups, where you compare "the

number of treatment changes"  for a simplified power analysis. That's

another thing you will want to report.

--

Rich Ulrich

---

From: SPSSX(r) Discussion <[hidden email]> on behalf of William Dudley <[hidden email]>
Sent: Tuesday, December 31, 2019 12:51 PM
To: [hidden email] <[hidden email]>
Subject: Seeking suggestions for Power analysis in Survival Analysis

 

Apologies in advance - this is off topic

I need to estimate sample size for a COX proportional hazard analysis with repeated measures.  The data will come from patients who are treated on multiple occasions (the repeated measure) and we will be looking at time to new  treatment (such as IV to oral for instance).  We anticipate that patients may have on average 5 treatment episodes.

I can estimate sample size with PASS but I don't know how to adjust for the repeated measures and wonder if there are alternative sample size estimation routines that I might employ.

Thanks in advance

Bill

--

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD