Hello all. I have a small question. Assume a one arm clinical trial, with the primary endpoint being safety related: 1 = A subject had AE (at least one), 0 = the subject did not have any AE. In addition assume that the sample size is small, say 30 subjects, as this is a preliminary pilot study. The time for the primary endpoint is up to 4 weeks, i.e. if subject had AE's within the first 4 weeks, he is a failure. Now, assume that a particular subject drops out after 3 weeks. How would you handle this, assuming that your analysis should be done on the ITT analysis set ? Clearly, if this subject already had an AE prior dropping out, he is a failure. But what if he didn't, and we do not know if he would have had AE after dropping out. Thank you for your opinions.
Avi, unfortunately in ITT analysis you need to carry over the value from 2 or 3 weeks to the forth (final week)> So if there were no accidents in the 2nd week then you need to carry over zero. After doing ITT it is common to do a sub-analysis of completers. Then to compare the attributes of completers to dropouts. If therer are no differences between the two, you analysis of completers will carry more weight.
The main concept behind an intent to treat analysis is that when conducting inference on the effect of an intervention on an outcome, all subjects who were randomized to be treated should contribute towards estimation of the treatment effect, whether they were compliant order not. In the usual case, this means, if outcome Y is regressed on treatment indicator X, then you should code X=0 for all subjects in the control arm, and most importantly, X=1 for all subjects who were enrolled in the intervention arm that are followed to an outcome, Y as opposed to an analysis that assigns X=1 only for subjects that complied. This is the source of confusion. In your case, what I think the reviewer (?) means is the following. I think if you have the outcome variable, Y=0/1 (AE) measured only in individual's that were treated and followed to AE, end of study or dropped out, and you know the timing of AE's then you compute the rate of AE's by taking the count of AE's divided by the sum of all the follow up times, including people who had AE's, people who finished the study without AE's and people who were lost to follow-up. The best you can do is compare this with the rate of the AE in a comparable age standardized population. A typical measure is the standardized AE ratio which would be the ratio of these two rates.
AE rate in your study:
AER = #{AE's}/ (tot person years at risk)
theta =log(AER)
v = var(theta) = 1/#{AES's}
95%CI AER = exp( theta - 1.96 v^0.5) to exp( theta + 1.96 v^0.5)
From published tables try the CDC or the NCHS, you can determine the total age specific rate for all AE's that you're monitoring and then age adjust them to your population... With a sample of 30, a kernel density estimate of your age distribution may have to do. Once you have the AE rate in the age adjusted general population, AER_aagp, then form a standardized adverse event rate ratio, SAERR, as the ratio of the AE rate in your pop'n to the AER_aagp.
Calculate it's 95% CI by logging the AE rate in the age adjusted general pop'n,
phi = log(AER_aagp) and then you have
95% CI SAERR =exp( theta - phi - 1.96 v^0.5) to exp( theta - phi + 1.96 v^0.5)
Notice that we assume that the AER_aagp is error free
Avi, your problem looks like a standard situation for the Kaplan-Meier model (i.e. survival analysis with right-censored data). From the KM model you will obtain the (possibly time-dependent) AE occurrence rate, and by integrating it over the total time-span (4 weeks) you will get the estimated ratio of patients with AE by the end of 4 weeks. And this will take the drop-out patients into account.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Probability
- Probability Theory