Utility of artificial intelligence-based genomic classification of occult malignancies in clinical practice.

Abstract

3065 Background: Occult primary malignancies (OPM) are a significant clinical problem, representing 3 to 5% of advanced cancers. Patients with OPM have a worse prognosis and fewer treatment options than when the primary site of origin of cancer can be confidently identified. Novel artificial intelligence-based genomic classification of occult malignancies has been validated for clinical use by previous studies, but reports of clinical utility have not yet been published. We report our experience of consecutive cases where a commercial AI-based genomic classifier was used in routine practice at a high-volume community cancer center. Methods: Malignancies with uncertain pathological classification or site of origin were subjected to molecular profiling and AI-based genomic classification (GPSai) capable of recognizing a set of 115 distinct primary tumor sites and histology classes. Molecular profiling consisted of whole exome and whole transcriptome sequencing of FFPE tumor samples along with other predictive biomarkers assays performed by a commercial laboratory (Caris Life Sciences). The identification of targetable mutations was reported, but variants of unknown significance were excluded when reporting genomic alterations. Results: From 1,316 patients between October 2021-November 2022, we analyzed 70 (5.3%) patients with OPM, including 44 (62.9%) females and 26 (37.1%) males, with 34 (48.6%) being 65 or older. GPSai classification was ordered for 73 samples from 70 patients, which resulted in the classification of 33 (45.2%) tumors, inconclusive results in 36 (49.3%), and insufficient tissue quantity in 4 (5.5%). Of the 33 cases with a GPSai classification, the putative pathology diagnosis was concordant in 26 (78.8%) and discordant in 7 (21.2%) cases. We investigated discordance between the GPSai and pathological diagnoses in discordant cases . The pathological classification was favored over GPSai in all 7 discordant cases. Of the 36 cases that could not be confidently assigned a classification, 14 (38.9%) possessed potentially actionable alterations . 23 of the 36 had no presumptive site of origin. Of these, 10 of 23 (43.5%) were PD-L1 positive, 4 of 23 (17.4%) had a high TMB, and 1 of 23 (4.4%) had MMR deficiency. Additionally, based on molecular profile and examination of RNA expression levels, 2 cases were able to be classified with subsequent IHC confirmation. Based on the results from genomic testing, 55.7% of patients (39 of 70) had FDA-approved treatment options. Conclusions: Over a third of patients with OPM were able to be assigned a cancer diagnosis concordant with diagnoses by pathologists. Of those which could not be classified, targetable biomarkers were commonly found. AI-based genomic classification of occult primaries is useful in clinical practice to help discern the site of origin and identify treatment options.