"Connect the dots" approach for a time-varying covariate

Dear NMusers, I wish to model a biomarker that is controlled by a time-varying variable. The temporal pattern of this time-varying variable is irregular which makes a parametric description of its profile somewhat difficult. I am hoping to use a "connect the dots" approach for this exercise i.e. linear interpolation for the time-varying variable in between the observation. I believe there are suggestions on the users net to modify the dataset to complete slopes and intercepts for each time interval. I was wondering if there is a simpler way to compute the linear interpolation on the fly within the control stream. Finally, the complicating issue is that the biomarker of interest needs an initialization of it's compartment since it doesn't start at zero. I would greatly appreciate if someone has a code and example dataset for such an exercise. Thanking the group in advance...Mahesh

I would interpolate the covariate outside of NONMEM. I think doing it within NONMEM unnecessarily complicates matters. You could use a cubic spline within each subject to get the value of the missing variable or your could use PROC MI in SAS to impute the missing value using the other time values as predictors in the imputation process. pete bonate Peter L. Bonate, PhD, FCP Genzyme Corporation Senior Director Clinical Pharmacology and Pharmacokinetics 4545 Horizon Hill Blvd San Antonio, TX 78229 USA [EMAIL PROTECTED] phone: 210-949-8662 fax: 210-949-8219 crackberry: 210-315-2713

Mahesh This is the interpolation code for use in the $DES block, assuming that you have values of a covariate at all observation records. If not, you need to control the code by appropriate IF-statements or interpolate (for observation times) at the time of data set creation: IF(NEWIND.NE.2) THEN ; SET UP FOR NEW INDIVIDUAL LCP=0 LTM=TIME ENDIF INT = LCP ; INTERCEPT = LAST covariate value TINT = LTM ; TIME SCALE ORIGIN = LAST TIME DEN = TIME-TINT SLOP = 0 IF(DEN.NE.0) SLOP = (CONC-INT)/DEN ; CONC is the value of the covariate in your case LCP=CONC ; RECORD CURRENT TIME & conc LTM=TIME ; FOR NEXT CALL $DES NPUT=INT+SLOP*(T-TINT) .... ;INPUT IS INTERPOLATED covariate, can be used even in $DES block Interpolation is better than imputation if you use differential equations (and need values at any time, not only at observation times). For example, with Nonmem interpolation, you can use interpolated PK and effect compartment model (to integrate, you need values at all times) Initialization can be done either using a unit dose to the biomarker compartment and assigning bioavailability (Nonmem V and VI) or (with Nonmem VI) using A_0(1)= etc. statements Leonid -------------------------------------- Leonid Gibiansky, Ph.D. President, QuantPharm LLC web: www.quantpharm.com e-mail: LGibiansky at quantpharm.com tel: (301) 767 5566 Samtani, Mahesh [PRDUS] wrote: > Dear NMusers, > > I wish to model a biomarker that is controlled by a time-varying variable. The temporal pattern of this time-varying variable is irregular which makes a parametric description of its profile somewhat difficult. I am hoping to use a “connect the dots” approach for this exercise i.e. linear interpolation for the time-varying variable in between the observation. I believe there are suggestions on the users net to modify the dataset to complete slopes and intercepts for each time interval. I was wondering if there is a simpler way to compute the linear interpolation on the fly within the control stream. Finally, the complicating issue is that the biomarker of interest needs an initialization of it’s compartment since it doesn’t start at zero. > > I would greatly appreciate if someone has a code and example dataset for such an exercise. > > Thanking the group in advance…Mahesh