Parameter identifiability and Extended Multiple Studies Analyses (EMSA) of a compartmental model for vitamin A (VA) kinetics: fixing fractional transfer coefficients [L(I,J)s] for absorption

In VA compartmental models in humans, variability in absorption of the oral dose impacts numerical identifiability. Using previous data on serum VA kinetics in 11 U.S subjects given [ 2 H 8 ]retinyl acetate (Tang et al., Am J Clin Nutr, 2003) and a related 6‐compartment whole‐body model (Cifelli et al., J Nutr, 2008), we investigated whether fixing two of the absorption parameters would improve identifiability. Based on mean values in past models, we fixed L(2,1) = 50, where L(2,1) represents initial processing of the ingested dose and L(5,2) = 0.5, where L(5,2) represents direct secretion of retinol bound to retinol‐binding protein from enterocytes into plasma. For each subject, data were fit to the model using the Windows version of the Simulation, Analysis and Modeling software and then a population model was generated using EMSA. Statistical certainty was evaluated from the model sum of squares for each subject; numerical identifiability of estimates was evaluated based on fractional standard deviations (FSD) and correlation coefficients ( r ) after EMSA. All parameters had FSDs that were < 0.5 and none of the parameters had an r above 0.8. We conclude that the relationship L(5,2) = 0.01[L(2,1)] is associated with a model fit that has adequate statistical certainty and that fixing two L(I,J)s related to absorption improves parameter identifiability and reduces the number of adjustable parameters to just five. Grant Funding Source : USDA‐58‐1950‐7‐707