SUMMARY: Immune checkpoint inhibitors (ICIs) have dramatically transformed the management and prognosis of various malignancies. By enhancing the ability of the immune system to identify and destroy cancer cells, ICIs have provided significant improvements in treatment outcomes across a range of tumor types.
Currently, Tumor Mutational Burden (TMB) and PD-L1 expression are the primary biomarkers approved by the FDA for predicting ICI efficacy. While these biomarkers have shown promise, they exhibit limited accuracy and face significant practical limitations, such as the need for sufficient tumor tissue for TMB analysis and the lack of standardized protocols for PD-L1 ImmunoHistoChemistry (IHC). These challenges underscore the need for a predictive model that is more accessible, cost-effective, and applicable across diverse healthcare settings.
Machine learning is a a branch of artificial intelligence that enables algorithms to learn from data, and identify key patterns and make predictions based on available clinical and biological data. SCORPIO is an AI-based model that was developed to address this clinical need, and provides an accurate and scalable tool for predicting patient responses to ICIs.
Study Design: The present study employed a machine learning approach to predict the efficacy of ICIs across a range of cancer types. By using routine blood tests as well as clinical characteristics, and utilizing data from diverse populations and cancer types, SCORPIO aimed to provide a broadly applicable predictive tool that could be used in real-world clinical settings to guide treatment decisions, with the goal of improving precision medicine. SCORPIO was developed and validated using retrospective data from multiple cohorts, including internal data from two major cancer centers (Memorial Sloan Kettering Cancer Center and Mount Sinai Health System, as well as data from 10 global Phase 3 clinical trials.
Participants: This study encompassed a total of 9,745 patients treated with ICIs across 21 different cancer types. The participants came from diverse backgrounds, including varying ethnicities, socioeconomic statuses, comorbidities, and health literacy levels, which added to the robustness of this model and real-world applicability. The study included data from patients at Memorial Sloan Kettering Cancer Center (2,000 patients), Mount Sinai Health System (1,159 patients), and global clinical trials (4,447 patients), representing the largest dataset in cancer immunotherapy to date.
Data Collection Methods: The SCORPIO model was trained using routine blood test results and basic clinical variables such as age, gender, cancer type, and prior treatments. The model was validated across multiple datasets to ensure that it could accurately predict both overall survival (OS) and Clinical Benefits (Response Rate and stable disease) for patients treated with ICIs. This study utilized machine learning algorithms to identify patterns in this data and make predictions about patient outcomes.
Analysis Techniques: Time-dependent Area Under the receiver operating characteristic Curve (AUC) was used to evaluate the predictive accuracy of SCORPIO for OS across different time points (6, 12, 18, 24, and 30 months). The performance of this model was compared to existing FDA-approved biomarkers, including TMB and PD-L1 immunostaining, to assess its superiority in predicting patient outcomes.
Results: The SCORPIO model demonstrated strong predictive performance in multiple cohorts. In internal testing, SCORPIO achieved median time-dependent AUC values of 0.763 and 0.759 for predicting OS at various time points, significantly outperforming TMB, which achieved AUC values of 0.503 and 0.543. SCORPIO also outperformed TMB for predicting clinical benefit, with AUCs of 0.714 and 0.641, compared to TMB’s 0.546 and 0.573. Additionally, SCORPIO was able to predict clinical outcomes more accurately than PD-L1 IHC.
External validation using data from 10 global phase 3 clinical trials and a real-world cohort from Mount Sinai Health System further supported the predictive power of SCORPIO. Despite the heterogeneity in the Mount Sinai Health System cohort, performance of SCORPIO remained consistent, demonstrating its potential for broad applicability in different patient populations and healthcare settings. While SCORPIO performed consistently well for Overall Survival prediction, its ability to predict Clinical Benefit varied across cancer types and cohorts, indicating that while the model is reliable for survival prognosis, there may still be challenges in accurately predicting Response Rate or stable disease in some contexts.
Conclusion: The SCORPIO model represents a significant advancement in the field of cancer immunotherapy. By leveraging routine blood tests and basic clinical data, SCORPIO provides a highly accurate and accessible method for predicting patient response to ICIs, surpassing the performance of existing FDA-approved biomarkers like TMB and PD-L1 IHC. Its consistent performance across diverse patient populations and cancer types makes it a promising tool for precision oncology.
This model could revolutionize the way physicians identify patients who are most likely to benefit from ICIs, offering a cost-effective, scalable alternative to current biomarkers that rely on complex genomic and immunologic assays. The widespread use of SCORPIO in clinical practice could improve patient outcomes by guiding more personalized treatment decisions, especially in settings where resources for advanced biomarker testing are limited. Future studies will focus on further refining the model to enhance its ability to predict clinical benefit across all cancer types and patient cohorts.
Prediction of checkpoint inhibitor immunotherapy efficacy for cancer using routine blood tests and clinical data. Yoo, S-K, Fitzgerald CW, Cho BA, et al. Nat Med (2025). https://doi.org/10.1038/s41591-024-03398-5
