Missing value when modeling categorical data

Dear NMUsers, I am modeling ordered-categorical PD data versus time, but since the clinical trial is ongoing, I don't currently have complete data set for each subject. In other words, for some subjects, I have 1 year's PD data, but for some others, I can only collect PD data for 1 month. For the 1 month's case, it is like missing data for the rest of time. But I need to consider time course of the proportion of subjects who got a certain PD score. In my case, if I use the general logistic regression model to fit the relationship between time and proportions of event, would there be any problem? If so, how can I avoid it? Or need I consider censoring like survival analysis? Any suggestion would be appreciated very much. Thanks in advance, Tianli ***************************************************************** Tianli Wang University of Minnesota

Dear NMUsers, I am modeling ordered-categorical PD data versus time, but since the clinical trial is ongoing, I don't currently have complete data set for each subject. In other words, for some subjects, I have 1 year's PD data, but for some others, I can only collect PD data for 1 month. For the 1 month's case, it is like missing data for the rest of time. But I need to consider time course of the proportion of subjects who got a certain PD score. In my case, if I use the general logistic regression model to fit the relationship between time and proportions of event, would there be any problem? If so, how can I avoid it? Or need I consider censoring like survival analysis? Any suggestion would be appreciated very much. Thanks in advance, Tianli ***************************************************************** Tianli Wang University of Minnesota

Tianli, Your data sounds like it could be described by a survival analysis. A time to event model will give you the survivor function i.e. the prob of not having had the event as a function of time. The event in your case is defined by the 'certain PD score'. The censoring will of course be taken care of by a typical survival analysis. Nick [email protected] wrote: > Dear NMUsers, > > I am modeling ordered-categorical PD data versus time, but since the clinical trial is ongoing, I don't currently have complete data set for each subject. In other words, for some subjects, I have 1 year's PD data, but for some others, I can only collect PD data for 1 month. For the 1 month's case, it is like missing data for the rest of time. But I need to consider time course of the proportion of subjects who got a certain PD score. In my case, if I use the general logistic regression model to fit the relationship between time and proportions of event, would there be any problem? If so, how can I avoid it? Or need I consider censoring like survival analysis? > > Any suggestion would be appreciated very much. > > Thanks in advance, > Tianli > ***************************************************************** > Tianli Wang > University of Minnesota -- Nick Holford, Professor Clinical Pharmacology Dept Pharmacology & Clinical Pharmacology University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand [email protected] tel:+64(9)923-6730 fax:+64(9)373-7090 mobile: +64 21 46 23 53 http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford

Tianli, Let me check whether I understood your data correctly: You have a trial where subjects enroll over time. You collected data up to some calendar date. Since subjects were enrolled over time, you have unequal amount of data for each subject. However, period of observation does not depend on the observed PD scores. If this is a correct description, then your data are missing completely at random (probability of an observation being missing does not depend on observed or unobserved measurements). In this case, you may analyze them ignoring missingness: just build your model with all the available data. ACoP conference that just has ended had a good section on missing data ( http://www.go-acop.org/acop2009/program ). You may look at the presentations there, or just on some literature. Thanks Leonid -------------------------------------- Leonid Gibiansky, Ph.D. President, QuantPharm LLC web: www.quantpharm.com e-mail: LGibiansky at quantpharm.com tel: (301) 767 5566 [email protected] wrote: > Dear NMUsers, > > I am modeling ordered-categorical PD data versus time, but since the clinical trial is ongoing, I don't currently have complete data set for each subject. In other words, for some subjects, I have 1 year's PD data, but for some others, I can only collect PD data for 1 month. For the 1 month's case, it is like missing data for the rest of time. But I need to consider time course of the proportion of subjects who got a certain PD score. In my case, if I use the general logistic regression model to fit the relationship between time and proportions of event, would there be any problem? If so, how can I avoid it? Or need I consider censoring like survival analysis? > > Any suggestion would be appreciated very much. > > Thanks in advance, > Tianli > ***************************************************************** > Tianli Wang > University of Minnesota

Tianli, Please look carefully at this: http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford/teaching/pharmacometrics/_docs/modelling_likelihoods_using_NONMEM_VI.pdf It shows you how to do time to event models which also include PKPD variables which change as a function of dose and time. Nick [email protected] wrote: > Dear Nick, > > I think I didn't make it clear in my first email that my model is a PK-PD linked model, which means, the probability of a certain event is a function of drug exposure which can describe as a function of time and some other parameters like CL, V, etc. But it seems to me that the hazard function only takes time into account. I am not familiar with survival analysis so this might be a very basic question: How can I combine the PK/PD model with the time-to-event model? > > Thanks in advance, > Tianli >