Abstract:
|
In clinical studies, subjects are often followed for varying lengths of time for the occurrence of some event of interest. Standard approaches for comparing the proportion of events observed in two or more treatment groups at the end of follow-up are then typically applied. However, these approaches ignore the fact that the individual observation periods for subjects may vary. If this variation is considerable, survival analysis may be used to evaluate such data. At the end of each subject's observation period, one notes whether or not the subject has experienced the event of interest. Hence the binary data generated this way represent either left-censored (if the event has occurred) or right-censored (if the event has not occurred) observations from the survival distribution governing the exact but unobserved event times. These data can be used to model the hazard function from flexible survival distributions such as the Weibull. Moreover, model fitting is easily achieved using standard software for GLMs. Convenient adjustment for covariates other than treatment is also possible. This approach will be discussed in some detail and an example given to illustrate the method.
|