Bayesian hierarchical model predicts biopharmaceutical stability indicators and shelf life with application to multivalent human papillomavirus vaccine

Predictive stability is demonstrated as a powerful method for assessing the shelf-life of biopharmaceutical products, such as therapeutic proteins and vaccines. A Bayesian hierarchical multi-level stability model is illustrated for the Human Papillomavirus (HPV) 9-valent recombinant sub-unit vaccine GARDASIL®9. Ensuring speedy manufacturing and ample supply to satisfy the need of patients globally is pivotal, particularly for expanding vaccine access to underserved populations. Product heat stability and cold-chain supply play a major role in deployment of vaccines particularly to lower income countries, while lengthy real-time stability and shelf-life supporting studies are resource-intensive and time-consuming. Hence, an accelerated model-informed stability approach is devised. The product in this case study contains 9 molecular types (antigens) which each target different viral genotypes of the same class of the virus. The molecular types are mixed together as a co-formulation within a container (vial or syringe). The stability behavior of the platform vaccine was well-characterized experimentally and a single stability-limiting attribute was identified (potency). A Bayesian hierarchical stability model was developed utilizing long-term drug product storage data through shelf life at 5 °C as well as shorter-term accelerated stability data at 25 °C and 37 °C for 30 product batches. The model was able to comprehensively assess the stability of all molecular types in the vaccine as well as covariates like container type within a singular unified model framework. Moreover, method superiority was demonstrated for this application over multiple approaches such as linear and mixed effects models. This work elucidates that biopharmaceutical product platform knowledge from previous lots of a biopharmaceutical product in conjunction with batch-specific data from early stability timepoints can be employed to support long-term assessment for shelf-life of the stability and shelf-life indicating attribute(s). These findings, applied to two types vaccines including a multivalent vaccine, hold utility towards enabling accelerated patient access of future complex vaccines and biotherapeutic modalities. The results provide a novel framework for estimating a model for stability data in the context of evolving regulatory guidance.