PCN219 PREDICTING CHEMOTHERAPY-ASSOCIATED THROMBOCYTOPENIA IN REAL WORLD CLINICAL SETTINGS

Abstract

Chemotherapy-induced thrombocytopenia is a frequent challenge in the management of cancer patients and can limit the ability to maintain effective dosing and treatment duration1. In this study, we assess real-world rates of chemotherapy-associated thrombocytopenia, measure the impact on patient dosing, and explore the potential of machine learning methods, using commonly available clinical variables, to predict development of this common, potentially treatment-limiting side effect. Patients initiating treatment for Ovarian cancer with gemcitabine, platinum-based agents, taxanes or PARP inhibitors were identified in a nationally representative commercial insurance claims database from 1Q2018-2Q2018. ICD codes were used to identify thrombocytopenia subsequent to treatment initiation. Baseline descriptive characteristics and subsequent dosing changes were assessed. An ensemble-model machine learning approach (Logistic Regression, Gaussian Naïve Bayes, Multi-layer Perceptron, K-Neighbors, Decision Tree, Random Forest, Gradient Boosted, XG Boost) was used to develop a predictive algorithm for thrombocytopenia, with a 3:1 training/testing split, and cross-validation scoring with a K-fold of 6. Model performance was assessed by standard metrics such as ROC curve, PPV, and NPV estimates. Feature ranking/impact was validated with Recursive Feature Elimination and extracted from the individual ensemble trees where possible. Among those treated with PARP inhibitors, for whom dosing information was available, more than two-thirds had no evidence of dose reduction. Dose reduction was also associated with a near-doubling of time on therapy. Highest model accuracy was achieved with Gradient Boosted approach, which when compared against a previously presented algorithm, showed a nearly 20% improvement in AUC2. Overall the study identified real-world rates of thrombocytopenia consistent with previous publications3,4,5. Results also indicate opportunities to improve the management of chemotherapy-associated thrombocytopenia through both active management of dose-adjustment as well as potential use of point-of-care predictive algorithms.