Hi. I'm looking for suggestions/literature on methods for imputing missing X-values (explanatory variables) in survival data with time-varying covariates. I am not focusing on imputation of events and event times.
Basically, we are modeling time to surgery. Given the inclusion of time-varying covariates, the number of repeated assessments modeled for any particular patient will depend upon her event/censor time. The imputation model should account for intra-person correlation of response across repeated assessments. When performing multiple imputation on repeated measures data with fixed assessments for all participants, I usual fit the imputation model to the 'wide' data set (one record of data per participant) and subsequently reshape the data into 'long' format (one record per person-assessment) for substantive modeling. However, for survival data with time-varying covariates, this method is not an attractive option because it would require imputing X values for occasions that occur after observed events--that would constitute a misspecification of the imputation model (even though such imputed values would be ignored in subsequent modeling). I've searched the literature some, but so far no luck. Any suggestions? Thanks in advance. Steve ------------------------------------------------------------------ Steve Gregorich University of California, San Francisco Department of Medicine 3333 California Street, Suite 335, Box 0856 San Francisco, CA 94143-0856 (FedEx and UPS use zip code 94118) [email protected] ------------------------------------------------------------------ -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20070523/6809f2c6/attachment.htm From jcole <@t> webcmg.com Wed May 23 19:41:40 2007 From: jcole <@t> webcmg.com ([email protected]) Date: Wed May 23 19:41:48 2007 Subject: [Impute] imputing missing X's in survival data with time-varying covariates In-Reply-To: <[email protected]> References: <[email protected]> Message-ID: <[email protected]> Hi Steve, I am mostly avoiding your question, but thought I would offer this up. Have you considered using FIML in Mplus? This should be able to produce the model you want, allow for the complex control of missing data you require, and even allow for some well-placed auxiliary variables in the model to help control the missingness mechanism. Jason ____________________________________ Jason C. Cole, PhD Senior Research Scientist & President Consulting Measurement Group, Inc. Tel: 866 STATS 99 (ex. 5) Fax: 310 539 1983 2390 Crenshaw Blvd., #110 Torrance, CA 90501 E-mail: [email protected] <mailto:[email protected]> web: http://www.webcmg.com <http://www.webcmg.com/> ____________________________________ From: [email protected] [mailto:[email protected]] On Behalf Of Gregorich, Steven Sent: Wednesday, May 23, 2007 3:18 PM To: IMPUTE post Cc: Gregorich, Steven Subject: [Impute] imputing missing X's in survival data with time-varying covariates Hi. I'm looking for suggestions/literature on methods for imputing missing X-values (explanatory variables) in survival data with time-varying covariates. I am not focusing on imputation of events and event times. Basically, we are modeling time to surgery. Given the inclusion of time-varying covariates, the number of repeated assessments modeled for any particular patient will depend upon her event/censor time. The imputation model should account for intra-person correlation of response across repeated assessments. When performing multiple imputation on repeated measures data with fixed assessments for all participants, I usual fit the imputation model to the 'wide' data set (one record of data per participant) and subsequently reshape the data into 'long' format (one record per person-assessment) for substantive modeling. However, for survival data with time-varying covariates, this method is not an attractive option because it would require imputing X values for occasions that occur after observed events--that would constitute a misspecification of the imputation model (even though such imputed values would be ignored in subsequent modeling). I've searched the literature some, but so far no luck. Any suggestions? Thanks in advance. Steve ------------------------------------------------------------------ Steve Gregorich University of California, San Francisco Department of Medicine 3333 California Street, Suite 335, Box 0856 San Francisco, CA 94143-0856 (FedEx and UPS use zip code 94118) [email protected] ------------------------------------------------------------------ -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20070523/3e4f27d8/attachment.htm From gregorich <@t> medicine.ucsf.edu Fri May 25 15:42:48 2007 From: gregorich <@t> medicine.ucsf.edu (Gregorich, Steven) Date: Fri May 25 15:44:00 2007 Subject: [Impute] imputing missing X's in survival data with time-varying covariates Message-ID: <[email protected]> Hi, Jason. Jason said: ------------------------------------------------------------------------ ---------------------------- Hi Steve, I am mostly avoiding your question, but thought I would offer this up. Have you considered using FIML in Mplus? This should be able to produce the model you want, allow for the complex control of missing data you require, and even allow for some well-placed auxiliary variables in the model to help control the missingness mechanism. Jason ------------------------------------------------------------------------ ---------------------- Steve replied: For some reason, I missed your original post and just saw it now when I checked the archives. Thanks for your idea. I have been avoiding Mplus, but this may prompt me to look into it. Also for everyone else I'm reposting my original message--hoping for more responses. ------------------------------------------------------------------------ -------------------------- I'm looking for suggestions/literature on methods for imputing missing X-values (explanatory variables) in survival data with time-varying covariates. I am not focusing on imputation of events and event times. Basically, we are modeling time to surgery. Given the inclusion of time-varying covariates, the number of repeated assessments modeled for any particular patient will depend upon her event/censor time. The imputation model should account for intra-person correlation of response across repeated assessments. When performing multiple imputation on repeated measures data with fixed assessments for all participants, I usual fit the imputation model to the 'wide' data set (one record of data per participant) and subsequently reshape the data into 'long' format (one record per person-assessment) for substantive modeling. However, for survival data with time-varying covariates, this method is not an attractive option because it would require imputing X values for occasions that occur after observed events--that would constitute a misspecification of the imputation model (even though such imputed values would be ignored in subsequent modeling). I've searched the literature some, but so far no luck. Any suggestions? Thanks in advance. Steve ------------------------------------------------------------------------ ---------------------- Steve NEW E-MAIL ADDRESS: [email protected] ------------------------------------------------------------------ Steve Gregorich University of California, San Francisco Department of Medicine 3333 California Street, Suite 335, Box 0856 San Francisco, CA 94143-0856 (FedEx and UPS use zip code 94118) [email protected] ------------------------------------------------------------------ -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20070525/f23272fb/attachment-0001.htm From JUDKIND1 <@t> westat.com Tue May 29 17:01:18 2007 From: JUDKIND1 <@t> westat.com (David Judkins) Date: Tue May 29 17:01:25 2007 Subject: [Impute] imputing missing X's in survival data with time-varying covariates In-Reply-To: <[email protected]> Message-ID: <[email protected]> Damn, that one is complex. Some imputation software that we have been developing would want the full record for each person stretched out sideways in order to model both cross-sectional and longitudinal relationships. But in this case, the length of the record would vary, which would be impermissible. I might think about partially dropping back to a time-invariant set of covariates by doing something simple like carrying forward/backward the last/next reported value of the covariate when it is missing. There were some papers a few years back exploring the properties of these simple procedures on SIPP, the Census Bureau's former Survey of Program Participation, and on its predecessor the ISDP. I think relevant results might be in the articles by Rizzo, Kalton and Brick and by Folsom and Witt in the 1994 ASA SRMS Proceedings. My recollection is that in setting, the simple methods behaved quite well. -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Gregorich, Steven Sent: Wednesday, May 23, 2007 6:18 PM To: IMPUTE post Cc: Gregorich, Steven Subject: [Impute] imputing missing X's in survival data with time-varying covariates Hi. I'm looking for suggestions/literature on methods for imputing missing X-values (explanatory variables) in survival data with time-varying covariates. I am not focusing on imputation of events and event times. Basically, we are modeling time to surgery. Given the inclusion of time-varying covariates, the number of repeated assessments modeled for any particular patient will depend upon her event/censor time. The imputation model should account for intra-person correlation of response across repeated assessments. When performing multiple imputation on repeated measures data with fixed assessments for all participants, I usual fit the imputation model to the 'wide' data set (one record of data per participant) and subsequently reshape the data into 'long' format (one record per person-assessment) for substantive modeling. However, for survival data with time-varying covariates, this method is not an attractive option because it would require imputing X values for occasions that occur after observed events--that would constitute a misspecification of the imputation model (even though such imputed values would be ignored in subsequent modeling). I've searched the literature some, but so far no luck. Any suggestions? Thanks in advance. Steve ------------------------------------------------------------------ Steve Gregorich University of California, San Francisco Department of Medicine 3333 California Street, Suite 335, Box 0856 San Francisco, CA 94143-0856 (FedEx and UPS use zip code 94118) [email protected] ------------------------------------------------------------------ -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20070529/c897c9df/attachment.htm From gregorich <@t> medicine.ucsf.edu Tue May 29 18:00:51 2007 From: gregorich <@t> medicine.ucsf.edu (Gregorich, Steven) Date: Tue May 29 18:01:39 2007 Subject: [Impute] imputing missing X's in survival data with time-varying covariates In-Reply-To: <[email protected]> References: <[email protected]> <[email protected]> Message-ID: <[email protected]> Thanks for your note, David. We are interested in modeling time-varying covariates: we want to model both between-person (baseline values) and within-person (change since baseline) effects of explanatory variables. Perhaps, as suggested here by Jason, fitting a discrete-time survival model with Mplus via EM is my best option. I've yet to read Bengt Muthen's 2005 JEBS article describing his parameterization of that model. Steve NEW E-MAIL ADDRESS: [email protected] ------------------------------------------------------------------ Steve Gregorich University of California, San Francisco Department of Medicine 3333 California Street, Suite 335, Box 0856 San Francisco, CA 94143-0856 (FedEx and UPS use zip code 94118) [email protected] ------------------------------------------------------------------ ________________________________ From: David Judkins [mailto:[email protected]] Sent: Tuesday, May 29, 2007 3:01 PM To: Gregorich, Steven; IMPUTE post Subject: RE: [Impute] imputing missing X's in survival data with time-varying covariates Damn, that one is complex. Some imputation software that we have been developing would want the full record for each person stretched out sideways in order to model both cross-sectional and longitudinal relationships. But in this case, the length of the record would vary, which would be impermissible. I might think about partially dropping back to a time-invariant set of covariates by doing something simple like carrying forward/backward the last/next reported value of the covariate when it is missing. There were some papers a few years back exploring the properties of these simple procedures on SIPP, the Census Bureau's former Survey of Program Participation, and on its predecessor the ISDP. I think relevant results might be in the articles by Rizzo, Kalton and Brick and by Folsom and Witt in the 1994 ASA SRMS Proceedings. My recollection is that in setting, the simple methods behaved quite well. -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Gregorich, Steven Sent: Wednesday, May 23, 2007 6:18 PM To: IMPUTE post Cc: Gregorich, Steven Subject: [Impute] imputing missing X's in survival data with time-varying covariates Hi. I'm looking for suggestions/literature on methods for imputing missing X-values (explanatory variables) in survival data with time-varying covariates. I am not focusing on imputation of events and event times. Basically, we are modeling time to surgery. Given the inclusion of time-varying covariates, the number of repeated assessments modeled for any particular patient will depend upon her event/censor time. The imputation model should account for intra-person correlation of response across repeated assessments. When performing multiple imputation on repeated measures data with fixed assessments for all participants, I usual fit the imputation model to the 'wide' data set (one record of data per participant) and subsequently reshape the data into 'long' format (one record per person-assessment) for substantive modeling. However, for survival data with time-varying covariates, this method is not an attractive option because it would require imputing X values for occasions that occur after observed events--that would constitute a misspecification of the imputation model (even though such imputed values would be ignored in subsequent modeling). I've searched the literature some, but so far no luck. Any suggestions? Thanks in advance. Steve ------------------------------------------------------------------ Steve Gregorich University of California, San Francisco Department of Medicine 3333 California Street, Suite 335, Box 0856 San Francisco, CA 94143-0856 (FedEx and UPS use zip code 94118) [email protected] ------------------------------------------------------------------ -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20070529/10496e71/attachment-0001.htm
