ABSTRACT
The sensitivity of the various preclinical models used to predict toxicity in humans varies. In this review, we analyze the various models used to predict safety in drug discovery and development with a view to understanding their translational value in humans. Twenty recently approved anticancer drugs were studied and the available nonclinical and clinical adverse effects information available from the US Food and Drug Administration (FDA) was analyzed. We found that gastrointestinal and hematopoietic toxicities in humans were predicted to the maximum extent by nonclinical models, whereas respiratory, integumentary, renal, nervous, hepatic, and cardiovascular were moderately predicted. Ocular, endocrine, and musculoskeletal toxicities were poorly predicted while lymphatic and immune toxicities were difficult to predict.