AI Tool Uses Routine Blood Tests to Predict Immunotherapy Response for Various Cancers

These inhibitors target the immune system, not the cancer directly. They work by releasing the brakes on immune cells, enhancing their ability to attack cancer. However, these drugs are costly and can cause severe side effects, as well as ineffective for most patients. Thus, selecting the right patients is crucial — matching the drugs to those most likely to benefit. While there are existing methods to predict whether tumors will respond to these drugs, they typically require advanced genomic testing, which is not widely accessible globally. Now, a new tool may soon be available to doctors worldwide that could more accurately predict whether individual cancer patients will benefit from immune checkpoint inhibitors, using only routine blood tests and clinical data.

Researchers from Memorial Sloan Kettering Cancer Center (MSK, New York, NY, USA) and the Tisch Cancer Institute at Mount Sinai (Mount Sinai, New York, NY, USA) have developed an AI-based model called SCORPIO. The model is not only more affordable and accessible but also significantly more effective at predicting patient outcomes than the two biomarkers currently approved by the U.S. Food and Drug Administration (FDA), according to a study published in Nature Medicine. The two FDA-approved biomarkers for predicting response to checkpoint inhibitors are tumor mutational burden (which measures mutations in a tumor) and PD-L1 immunohistochemistry (which evaluates the expression of the PD-L1 protein in tumor samples). Both methods require tumor samples. Genomic testing for mutations is costly and not available everywhere, and there is considerable variability in evaluating PD-L1 expression.

SCORPIO, in contrast, uses readily accessible clinical data, including routine blood tests, such as the complete blood count and the comprehensive metabolic profile, which are performed in clinics worldwide. The researchers discovered that SCORPIO outperforms the current clinical tests. This simple, affordable approach could improve access to care, reduce costs, and ensure that patients receive the right treatments. Initially developed by the MSK team with data from MSK patients, SCORPIO was further enhanced in collaboration with Mount Sinai researchers using ensemble machine learning — an AI technique that combines multiple tools to detect patterns in clinical data from blood tests and treatment outcomes. The model was trained using retrospective data from over 2,000 MSK patients treated with checkpoint inhibitors, spanning 17 types of cancer. The model was then tested on data from an additional 2,100 MSK patients to confirm its high accuracy in predicting outcomes.

Next, the team applied the model to nearly 4,500 patients treated with checkpoint inhibitors across 10 different phase 3 clinical trials worldwide. Additional validation was done with data from almost 1,200 patients treated at Mount Sinai. In total, the study encompasses nearly 10,000 patients from 21 different cancer types, making it the largest cancer immunotherapy dataset to date. This extensive testing and validation were carried out not only to develop a predictive model but to create one that is widely applicable to patients and physicians in various locations. The team plans to collaborate with hospitals and cancer centers globally to test the model using more data from diverse clinical settings while optimizing it based on feedback. They are also working on developing a user-friendly interface for clinicians, making it accessible wherever they may be located.