Re: Time-to-event

From: LSheiner <lewis@c255.ucsf.edu> Subject: Re: Time-to-event Date: Mon, 30 Nov 1998 09:24:44 -0800 I can respond to some of your questions ... tod michel wrote: > > Dear NONMEM users, > > I have some questions about the analysis of time-to-event data with > Weibull distribution using NONMEM version IV and the code described at > the PKPD server: > pkpd.icon.palo-alto.med.va.gov/nonmem.dir/NON_CONTINUOUS.DIR/ > > 1- What are the key assumptions that must be met when using the Weibull > distribution for time-to-event? In other words, can this model be used, > at least as an approximation, for all this kind of data type? (I > recognize the restriction given by LB Sheiner that covariates have to be > time-independent). The same assump6ions as when using the Weibull for survival analysis in a more classical context, wqith teh single additiontaht the random effect is distributed normally. You might findit useful to read about survival analysis in some textbook, for example, Kalbfleisch & Prentice, "The statistical analysis of failure time data", Wiley, 1980. > > 2- Are there restrictions about the value of Gamma (the exponent of > (T/Tmedian))? Has it to be greater than 1? No, restrictions - as the exponent changes, the shape of the hazard changes. > > 3- Is the subroutine CCONTR identical to that used for random logistic > regression (given in the same directory /NON_CONTINUOUS.DIR/)? You should no longer be using CCONTR or CONTR. NONMEM V includes the LIKELIHOOD or -2LOGLIKELIHOOD option on the $ESTIM record. Your $PRED should supply the likeihood (probability) of the observation, or -2log likelihood, respectively, and NONMEM takes care of the rest.