Abstract B150: Identifying tolerable schedules for targeted anti-cancer agents by applying ordered logistic regression modeling to empirical pathology scores in mice

Abstract

Abstract Even targeted oncology agents may have significant toxicity, either target-mediated or otherwise. Some targets are simultaneously drivers of cancer yet essential for normal turnover in certain tissues. We sought to develop a predictive model for gut toxicity, to enable model-based selection of the optimal dose and schedule in mice. Mice were treated under several schedules (twice daily, once every one, two or three days), at various doses (6.25-100 mg/kg) of the inhibitor AZ17, and for various durations. Target inhibition is expected to disregulate normal turnover of the gut lining. Epithelial histopathology findings collected from individual animals at various timepoints after the final dose were manually scored on an empirical five-point scale. An ordered logistic regression model was fit to all animals individually across all regimens simultaneously, to describe the probability of observing toxicity at a particular level on the five-point scale as a function of dose and time. It has recently been applied to clinical data, but this is a novel approach in the pre-clinical space. This model-based approach allows integration of data collected at various timepoints and from different studies. It differs from the standard PK/PD approach in that it attempts to predict not the toxicity score itself (which can only be an integer), but the probability of observing each score on the scale. It allows for recovery once dosing stops. The predicted probabilities will thus vary with dose, schedule, duration of treatment and time since last dose. We illustrate with a study of 30 mice, showing that the model was able to capture a number of essential features of any such model: (i) the steep nature of the dose-response relationship; (ii) robustness to outliers (e.g., when a single animal experienced severe tox at a generally tolerated dose); (iii) robustness to non-exposure-driven tox (e.g., finding in vehicle-treated animal). Since staining/reading of pathology slides is not resource intensive, this approach is suitable even for lead-optimization projects. Such data could be collected at the end of a standard tolerability study, in which body-weight is the typical endpoint. It could also be collected from tumor-bearing animals, either a pharmacodynamic or efficacy study. Both applications conform with the 3R's imperative to ‘reduce, refine, replace’ use of animals. Since the model, by design, makes no assumptions about mechanism, it could easily be extended to assess synergistic tox when combined with chemotherapy. Citation Format: Francis Gibbons, Shenghua Wen, Prasad Nadella. Identifying tolerable schedules for targeted anti-cancer agents by applying ordered logistic regression modeling to empirical pathology scores in mice. [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference: Molecular Targets and Cancer Therapeutics; 2015 Nov 5-9; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2015;14(12 Suppl 2):Abstract nr B150.