Hello,
I am a new subscriber to this list after a colleague suggested I do so, as
the SPSS list was not addressing my concern.
Earlier this week I posted to the SPSS List a request for assistance with
addressing some criticisms levied against Kaplan-Meier survival analysis.
As I mentioned, I received no responses to my request so I'd like to re-
post the request here.
I have frequently used K-M survival analyses (using SPSS) to analyze time
to a critical event, and provide estimates the incidence of that event.
However, while K-M survival is a staple in the statistician's arsenal,
especially within the medical field, I would describe my understanding of
it as general.
However, I read recently that K-M survival analysis artificially inflates
the estimates of the actuarial incidence of an event, as compared to the
cumulative incidence model. As I understand it, "actuarial incidents" are
based on censoring patients who die or otherwise leave the analysis due to
causes other than from the disease.
Critics suggest that the K-M estimates are not appropriate if competing
risks are to be considered. K-M ignores the presence of competing risks,
whereas the cumulative incidence model does not.
So, K-M can underestimate the benefit of a particular intervention or
therapy because K-M artificially inflates the estimates, confidence
intervals, and SE at later time points. This they claim results from the
decreasing N of cases at risk at later time points.
The cumulative incidence model, on the other hand, does take into account
competing risks. So, it supposedly provides more accurate estimates of the
percentage of patients who will actually sustain an event.
With a general level of understanding relative to this procedure, I
unfortunately do not feel so well versed as to know whether the criticisms
above are valid ones (having only just recently encountered them). Might
any on this list be able to provide me with an overview of these two
approaches, or perhaps cite a reference which may help compare and contrast
them, and help explain when to use one in favor of the other?
I very much appreciate your help and guidance. Thank you in advance,
John Norton
|