Authors
Daniel J George, Elle C Moore, Gerard C Blobe, Nicholas C DeVito, Brent A Hanks, Michael R Harrison, Christopher J Hoimes, Jingquan Jia, Michael A Morse, Parvathy Jayaprakasan, Andrew MacKelfresh, Hillary Mulder, Kyle A Beauchamp, Jackson Michuda, Martin C Stumpe, Eric Perakslis, Timothy Taxter
Background: Patients with cancer of unknown primary (CUP) have a poor prognosis due to delays in diagnostic workup and empirically-selected platinum-based regimens that may not be the most active first line for the primary disease. The Tempus Tumor Origin (TO) test is a machine learning classifier that uses RNA-Seq data to identify the most likely cancer type or subtype from 68 possible diagnoses. Despite the importance of cancer type identification in advising guideline-based treatment, prior studies of molecular classifiers have found unclear clinical impact in the CUP setting.
Methods: We retrospectively analyzed de-identified records from 289 patients who received a diagnosis of uncertain/unknown primary; all had NGS and TO testing ordered by the original treating clinician. Two oncologists separately reviewed patient clinical information—including imaging, pathology, and NGS reports— to determine the course of treatment before they reviewed TO testing and evaluated whether the algorithm’s predicted diagnosis would change treatment. Disagreement was adjudicated by a third reviewer.
Results: The most common cancer types predicted by the TO test were cholangiocarcinoma (n=58), lung adenocarcinoma (n=41), and pancreatic adenocarcinoma (n=22). The TO test results would have altered the treatment plan in 235 out of 289 cases (81%) and led to follow-up site-specific testing in 140 cases (48%). When treatments changed, the drugs that were most often removed as treatments were: 5-FU (n=46), oxaliplatin (n=42), and carboplatin (n=41). Gemcitabine and cisplatin were added as treatment recommendations in 54 and 49 cases, respectively. Additionally, non-chemotherapy changes include checkpoint inhibitors (CPI; added in 26 cases, removed in 4), surgery (added in 4 cases), and radiation (added in 3 cases, removed in 1).
Conclusions: In our study, we demonstrate that the TO test can specify a diagnosis that alters the therapeutic management for up to 81% of patients with unknown primary. These changes include the addition and removal of chemotherapy and CPI, as well as alterations in radiation and surgical treatments, highlighting the potential of molecular classifiers to provide clinical insight into the management of CUP patients.
VIEW THE PUBLICATION
VIEW THE POSTER