This cohort study examines the use of human epidermal growth factor receptor 2 immunohistochemistry pan-tumor testing among patients with solid tumors.

A subset of cancers present with unclear or potentially incorrect primary histopathologic diagnoses, including cancers of unknown primary (CUP). We aimed to develop and validate an artificial intelligence (AI) tool, Genomic Probability Score AI (GPSai™), which predicts tumor tissue of origin in CUP and flags potential misdiagnoses for additional workup during routine molecular testing. The GPSai model was trained on whole exome and whole transcriptome data from 201,612 cases submitted for tumor profiling at Caris Life Sciences. Retrospective (N = 21,549) and prospective (N = 76,271) validations were performed. The clinical impact was evaluated over 8 months of live testing and through physician surveys. GPSai demonstrated 95.0% accuracy in non-CUP cases and reported on tumor tissue of origin in 84.0% of CUP and 96.3% of non-CUP cases. During the initial 8 months of implementation, GPSai changed the diagnosis on 704 patients (0.88% of all profiled cases), which were supported by orthogonal evidence including imaging, IHC, mutational signatures, hallmark fusions, or viral reads. Diagnosis changes prompted changes in targeted therapy eligibility based on level 1 clinical evidence in 86.1% of cases (n = 606/704). A majority (89.7%; n = 87/97) of physician responses indicated acceptance of the GPSai results, and 53.6% (n = 52/97) of responses stated that the results prompted a change in treatment plan. GPSai accurately identifies tumor tissue of origin and has the potential for clinical impact in a small but meaningful subset of patients with CUP or pathologically ambiguous tumors. Our results support the integration of this AI tool into routine molecular testing to improve diagnostic accuracy and guide subsequent therapeutic decisions. Significance: Our findings show that GPSai, a deep learning–based tool, can support the identification of primary tumor sites with high accuracy in conjunction with orthogonal evidence. Its integration into routine tumor profiling furthermore allows simultaneous biomarker identification. Analysis of real-world implementation of GPSai shows that it enhances diagnostic accuracy, including resolution of CUP cases, and prompts clinically relevant therapeutic recommendation changes without requiring additional specimen.

Background/Objectives: Immune checkpoint inhibitors (ICIs) are frontline treatment for advanced Merkel Cell Carcinoma (MCC), regardless of viral status. Frontline ICIs provide durable benefit to only half of patients, highlighting a need for alternative therapies. In this study, the objective is to leverage whole exome sequencing (WES) and transcriptome sequencing (WTS) to distinguish genomic alterations associated with ICI response. Investigate differential genomic alterations between virus-positive (VP) and virus-negative (VN)-MCC to identify novel therapeutic targets. Methods: A total of 95 MCC cases underwent WES and WTS. Utilizing computational pipelines applied to WES, we identified viral status and tumor mutational burden (TMB). RNA-seq data was used to characterize the immune microenvironment. Results: Of 95 MCC cases, 57 (60%) were VP-MCC and 38 (40%) were VN-MCC. Median TMB was higher in VN-MCC (27.5 vs. 1 Muts/Mb). Mutations in TP53, RB1, NOTCH1, KMTD2, KMT2C, and PIK3CA were primarily found in VN-MCC. MAPK Pathway Activity Score, NK cell infiltration, and the immune checkpoint gene CD276 in VN-MCC tumors were upregulated. No overall survival (OS) difference was identified between VP and VN-MCC, even after ICIs. Conclusions: MCC oncogenesis and treatment response transcend viral status. While mutational analysis confirms previous findings, assessment of the transcriptome and tumor microenvironment suggests alternate therapeutic targets.

GPSai model scoring. A representative example of scoring assigned by GPSai is shown. Subcategories (e.g., lung adenocarcinoma, lung squamous cell carcinoma) must sum to unity of the major category score (e.g., non-small cell lung carcinoma).

Retrospective validation of GPSai. Model performance per major cancer category is shown for primary (A) and metastatic (B) tumors. Confusion matrices show numbers of each predicted tumor type in the boxes, with numbers of correctly predicted tumors (matching submitted diagnosis) in each category shown in the diagonal. PPV for each category is indicated by the bars on top, and sensitivity by the bars to the right. Prospective validation results are shown in Supplementary Fig. S8 and Supplementary Table S3. CNS, central nervous system.

GPSai model performance in prospective validation.