Searchable abstracts of presentations at key conferences in endocrinology
Endocrine Abstracts (2022) 81 OC13.6 | DOI: 10.1530/endoabs.81.OC13.6

1University Hospital Carl Gustav Carus at TU Dresden, Dresden, Germany; 2Medical Center Groningen, Groningen, Netherlands; 3Dresden Center of Intelligent Materials, at the TU Dresden, Dresden, Germany; 4National Institutes of Health, Washington DC, United States; 5Institute of Cardiology, Warsaw, Poland; 6University Hospital Zurich, Zurich, Switzerland; 7University Hospital Würzburg, Würzburg, Germany; 8Radboud University Hospital, Nijmegen, Netherlands; 9Spanish National Cancer Research Center, Madrid, Spain


Introduction: Pheochromocytomas and paragangliomas (PPGLs) exhibit an up to 20% malignancy rate. Various clinical, genetic, and pathological features have been proposed as predictors of malignancy. However, until present there are no robust indices to reliably predict metastatic PPGLs.

Aim: The aim of the present study was to prospectively validate the value of methoxytyramine as risk marker of metastatic disease and establish a machine learning (ML) model, based on clinical and biochemical features, to reliably predict malignancy in patients with PPGLs.

Methods: This study included retrospective data of 493 patients for the generation and training of ML models. Data of 295 patients prospectively enrolled in the multicenter international PMT-Study were used for the validation of the predictive value of methoxytyramine and the external validation of the selected ML model. The predefined features for selection analysis were sex, age at initial diagnosis, locations and size of tumor(s), previous history of PPGL, presence of SDHB mutation, plasma normetanephrine, metanephrine and methoxytyramine.

Results: Receiver operating characteristic curves indicated that plasma methoxytyramine using an optimal cutoff of 33 pg/ml provided an accurate biomarker for detecting patients with metastatic PPGLs. After feature selection and the use of 4 different models, ensemble trees model, which comprised 9 features, had the greatest discriminatory ability with an AUC of 0.9938 (95% CI:0.9903-0.9951). The ensemble trees model was validated externally and ranked based on the balanced accuracy with an AUC of approximately 0.9261 (95% CI:0.9293-0.9381).

Conclusion: Our study confirms in a prospective series the value of methoxytyramine as a strong predictor of metastatic PPGLs. Importantly we demonstrate predictive ML models, as the first effective, non-invasive and highly accurate approach to predict malignant disease in patients with PPGLs, providing immediate guidance to clinicians for individualized patient management and follow-up strategies.

Volume 81

European Congress of Endocrinology 2022

Milan, Italy
21 May 2022 - 24 May 2022

European Society of Endocrinology 

Browse other volumes

Article tools

My recent searches

No recent searches.