Abstract

Shelf life estimation procedures, following ICH guidelines, use multiple batch regression with fixed batch effects. This guidance specifically mandates estimates based on at least 3 batches. Technically, the fixed-batch model limits inference to the batches actually observed, whereas ICH requires resulting estimates to apply to all future batches stored under similar conditions. This creates a conflict between the model used and the inference space the model is intended to address. Quinlan, et al. (2013) and Schwenke (2010) studied the small sample behavior of this procedure. Both studies revealed large sampling variation associated with the ICH procedure, producing a substantial proportion of extremely low and extremely high estimates. Quinlan, et. al (2013) also considered alternative approaches including mixed models with random batch effects. While this eliminated the conflict between model and intended inference space, there were still problems with the mixed model approaches Quinlan considered. We present a Bayesian augmented mixed model approach to shelf life estimation that takes advantage of the theoretical benefits of the mixed model and uses prior information about variance components to improve accuracy of shelf life estimation procedure.

Keywords

BLUP, shelf life estimation

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Share

COinS
 
Jan 1st, 1:07 AM

Best Linear Unbiased Prediction: an Illustration Based on, but Not Limited to, Shelf Life Estimation

Shelf life estimation procedures, following ICH guidelines, use multiple batch regression with fixed batch effects. This guidance specifically mandates estimates based on at least 3 batches. Technically, the fixed-batch model limits inference to the batches actually observed, whereas ICH requires resulting estimates to apply to all future batches stored under similar conditions. This creates a conflict between the model used and the inference space the model is intended to address. Quinlan, et al. (2013) and Schwenke (2010) studied the small sample behavior of this procedure. Both studies revealed large sampling variation associated with the ICH procedure, producing a substantial proportion of extremely low and extremely high estimates. Quinlan, et. al (2013) also considered alternative approaches including mixed models with random batch effects. While this eliminated the conflict between model and intended inference space, there were still problems with the mixed model approaches Quinlan considered. We present a Bayesian augmented mixed model approach to shelf life estimation that takes advantage of the theoretical benefits of the mixed model and uses prior information about variance components to improve accuracy of shelf life estimation procedure.