PD56-03\u2003INDEPENDENT VALIDATION OF TWO MODELS TO PREDICT NECROSIS/FIBROSIS IN POST CHEMOTHERAPY RESIDUAL RETROPERITONEAL MASSES OF PATIENTS WITH ADVANCED TESTICULAR CANCER

Abstract

INTRODUCTION AND OBJECTIVES: Metastatic non-seminomatous testicular germ cell tumor patients with residual retroperitoneal lesions >1cm are treated with postchemotherapy retroperitoneal lymph node dissection (pcRPLND) according to guideline recommendations. However, up to 50% of patients are overtreated since the histology shows residual necrosis/fibrosis only and patients might have been subjected to active surveillance. The aim of our study is to validate and evaluate the two currently best performing prediction models (Vergouwe and Leao) for postchemotherapy residual mass histology. METHODS: We performed a retrospective analysis including 402 patients who underwent a pcRPLND from 2008 to 2015. This cohort was used to validate the two prediction models by Vergouwe and Leao. Clinical usefulness to predict histology was tested using thresholds as recommended by the original publications. RESULTS: Using our validation cohort, the Vergouwe model reached a significantly better AUC compared to the Leao model (0.760 [CI 0.713-0.807] vs. 0.692 [0.640-0.744], p = 0.002) in the prediction of benign histology (necrosis/fibrosis). However, adjusting our data to the models had no significant improvement. At a threshold of >70%, the Leao model revealed that pcRPLND would be avoided in 10.2% of patients with benign disease with an error rate of 3.8% for viable tumor, while the Vergouwe model showed an error rate of 10.1% for viable tumor and 2.9% to for teratoma to avoid pcRPLND in 27.4% of all patients with benign disease. CONCLUSIONS: Our study represents the largest external validation study of the currently best performing prediction models to predict benign histology with a contemporary patient cohort. According to the present study, the discriminatory accuracy of both models is not sufficient to safely select patients for surveillance strategy instead of pcRPLND. Therefore, further studies including new biomarkers such as miRNA 371-3p are needed to optimize the clinical accuracy of potential prediction models and to minimize pcRPLND overtreatment. Source of Funding: none