Background: Metastasis of cutaneous squamous cell carcinoma (cSCC) is uncommon. Current staging methods are reported to have sub-optimal performances in metastasis prediction. Accurate identification of patients with tumors at high risk of metastasis would have a significant impact on management.
Objective: To develop a robust and validated gene expression profile signature for predicting primary cSCC metastatic risk using an unbiased whole transcriptome discovery-driven approach.
Methods: Archival formalin-fixed paraffin-embedded primary cSCC with perilesional normal tissue from 237 immunocompetent patients (151 nonmetastasizing and 86 metastasizing) were collected retrospectively from four centers. TempO-seq was used to probe the whole transcriptome and machine learning algorithms were applied to derive predictive signatures, with a 3:1 split for training and testing datasets.
Results: A 20-gene prognostic model was developed and validated, with an accuracy of 86.0%, sensitivity of 85.7%, specificity of 86.1%, and positive predictive value of 78.3% in the testing set, providing more stable, accurate prediction than pathological staging systems. A linear predictor was also developed, significantly correlating with metastatic risk.
Limitations: This was a retrospective 4-center study and larger prospective multicenter studies are now required.
Conclusion: The 20-gene signature prediction is accurate, with the potential to be incorporated into clinical workflows for cSCC.
Keywords: cutaneous squamous cell carcinoma; machine learning; metastasis; prognosis; risk stratification; transcriptomics.
Copyright © 2023 American Academy of Dermatology, Inc. Published by Elsevier Inc. All rights reserved.