American Society of Clinical Oncology, Journal of Clinical Oncology, 6_suppl(35), p. 455-455, 2017
DOI: 10.1200/jco.2017.35.6_suppl.455
Full text: Unavailable
455 Background: Potential liver toxicity associated with pazopanib treatment has provided a challenge for decision-making regarding patient selection and monitoring. The goal of this study was to develop a tool for predicting liver toxicity for patients receiving pazopanib. Methods: Data was pooled from several clinical trials in patients with renal cell carcinoma, soft tissue sarcoma, or ovarian carcinoma; the dataset consisted of 2,051 patients treated with pazopanib in these trials. Liver toxicity was defined as an ALT level > 3 times the upper limit of normal (ULN) and a bilirubin level > 2 times ULN. Predictors of hepatotoxicity included were age, gender, race, hypertension, performance status, indication (tumor type), prior antineoplastic treatment, concomitant treatment with CYP450, P-gp, or BCRP inhibitors, and baseline clinical characteristics of weight, ALT, AST, ALP, and bilirubin. The statistical modeling was performed using logistic regression. Continuous predictors were modeled with restricted cubic splines to permit potentially nonlinear effects. The performance of the model was evaluated using discrimination (measured by concordance index) and calibration (assessed graphically). Results: Calibration plots suggested adequate congruence between predicted probabilities and actual proportions of liver toxicity. The model had a bootstrap corrected concordance index of 0.739. The calibration of the model was excellent for predicted probabilities up to 30% (Table). The prediction model was represented as a nomogram. Conclusions: This appears to be the first nomogram that predicts, with reasonable accuracy, liver toxicity in patients receiving pazopanib based on their baseline clinical characteristics. Although nomograms are limited by the degree of uncertainty that is inherent in the estimates of probability, the findings may aid clinicians in selecting patients for treatment and monitoring. Validation in a separate dataset is needed to confirm the nomogram’s accuracy. [Table: see text]