Re: criteria for implementing BSV-V2 and BSV-Q
Dear Carolien,
Compare predictions (PRED and IPRED) between the models without and with BSV on the parameters of the second compartment. It could be that extra ETA accomodates few outlying points. 17 points of OF is not much (unless your data set is small). You may be better off with a stable model with fewer random effects unless you see a real improvement of the model in diagnostic plots (including VPC).
Regards,
Katya
Ekaterina Gibiansky, Ph.D.
CEO&CSO, QuantPharm LLC
Web: www.quantpharm.com
Email: EGibiansky at quantpharm.com
Quoted reply history
On 3/19/2014 12:53 PM, H.C.A.M. Hazendonk wrote:
> Dear all,
>
> In the present project a two compartmental structural model adequately describes our data. Estimated parameters are Cl, V1, Q and V2, between-subject varibility is estimated for CL and V1 - estimates are precise. In the further development of the structural model introduction of BSV on either V2 and Q produced a significant drop in objective function of -17 points. Estimates were however large (>100%) with moderate precision (40-50%) and large shrinkage (40-50%). We are aware that interpreting the goodness of fit plots with these shrinkage values is not reliable (Savic & Karlsson 2009). We are wondering what criteria to use for implementing BSV-V2 and BSV-Q in the structural model.
>
> Kind regards,
>
> Carolien Hazendonk
>
> Drs. H.C.A.M. Hazendonk, M.D. PhD-student
> Department of (Pediatric)Hematology,
>
> Erasmus University Medical Center Rotterdam
>
> E-mail address: [email protected]