Prediction model for pheochromocytoma/paraganglioma using nCounter assay.

ABSTRACT

BACKGROUND: World Health Organization defined pheochromocytomas/paragangliomas (PPGL) as malignant tumors in 2017 because the existing classification system could not reflect locally aggressive behavior sufficiently. However, predicting the likelihood of metastasis remains a crucial part of the treatment strategy. METHODS: From one tertiary care hospital and one secondary hospital, 97 PPGL cases were selected. Medical records of PPGL cases with the presence of formalin-fixed and paraffin-embedded (FFPE) tissue of primary lesion were reviewed. For FFPE tissues, a nCounter assay was conducted to determine differently expressed genes between metastatic and non-metastatic PPGL groups. Performances of prediction models for the likelihood of metastasis were calculated. RESULTS: Of a total of 97 PPGL cases, 39, 20, and 38 were classified as benign, malignant, and validation, respectively. In the nCounter assay, CDK1, TYMS, and TOP2A genes showed significant differences in expression. Tumor size was positively correlated with CDK1 expression level. The Lasso regression model showed supreme performance of sensitivity 91.7% and specificity 95.5% when those significant factors were considered. CONCLUSION: Machine learning of multi-modal classifiers can be used to create a prediction model for metastasis of PPGL with high sensitivity and specificity using nCounter assay. Moreover, CDK1 inhibitors could be considered for developing drug treatment.