Utilizing Electronic Health Records (EHR) and Tumor Panel Sequencing to Demystify Prognosis of Cancer of Unknown Primary (CUP) patients

Abstract

When a standardized diagnostic test fails to locate the primary site of a metastatic cancer, it is diagnosed as a cancer of unknown primary (CUP). CUPs account for 3-5% of all cancers but do not have established targeted therapies, leading to typically dismal outcomes. Here, we develop OncoNPC, a machine learning classifier of CUP, trained on targeted next generation sequencing data from 34,567 tumors across 22 primary cancer types collected as part of routine clinical care at three institutions under AACR Project GENIE initiative. OncoNPC achieved a weighted F1 score of 0.94 for high confidence predictions on known cancer types (65% of held-out samples). To evaluate its clinical utility, we applied OncoNPC to 971 CUP tumor samples from patients treated at the Dana-Farber Cancer Institute (DFCI). OncoNPC CUP subtypes exhibited significantly different survival outcomes, and identified potentially actionable molecular alterations in 23% of tumors. Importantly, patients with CUP, who received first palliative intent treatments concordant with their OncoNPC predicted sites, showed significantly better outcomes (Hazard Ratio 0.348, 95% C.I. 0.210 - 0.570, p-value 2.32e-5) after accounting for potential measured confounders. As validation, we showed that OncoNPC CUP subtypes exhibited significantly higher polygenic germline risk for the predicted cancer type. OncoNPC thus provides evidence of distinct CUP subtypes and offers the potential for clinical decision support for managing patients with CUP.