In the mid-1990s, researchers showed that releasing a natural brake, or checkpoint, on the immune system could help cells recognize and attack tumors.1 This finding, which eventually won scientists the Nobel prize in 2018, paved the way for immune checkpoint inhibitors (ICIs) to treat several forms of cancer.
“The advent of checkpoint inhibitor drugs has been revolutionary for many cancer types,” said Luc Morris, a surgeon and cancer researcher at Memorial Sloan Kettering Cancer Center (MSKCC). “But fundamentally, we all realize the limitation that these drugs do not achieve a tumor response in every patient.” A combination of several factors—patient genetics, tumor microenvironment, and tumor genomes—influence the efficacy of ICIs.2
These therapies are expensive and can cause severe side effects, explained Diego Chowell, a computational immunologist at Icahn School of Medicine at Mount Sinai. “So, it's very important to [solve] the problem of predicting who will and who will not respond to this particular treatment,” he noted.
While some other tools exist, Chowell, Morris, and their team have developed an artificial intelligence (AI)-based model that could better predict whether people with cancer will benefit from ICI therapy using routine blood tests and clinical data.3 Their research, published in Nature Medicine, describes a widely accessible tool to advance precision oncology medicine for patients.
The researchers started by collecting basic blood test data including complete blood count and metabolic panel from about 1,600 patients across several cancer types treated with ICIs between 2014 and 2019 at MSKCC. They used machine learning algorithms to extract patterns between blood test data and treatment outcomes. They then trained a model, that they dubbed SCORPIO, on these data to calculate the probability of survival after ICI treatment.
Chowell, Morris, and their team tested their trained model on data of 2,100 patients treated in the same hospital. This testing revealed that the model’s predictions largely matched the outcomes that patients experienced. Next, the researchers tested SCORPIO with data from a diverse set of patients at Mount Sinai Health System and from 10 global Phase 3 clinical trials consisting of more than 4,400 patients. These tests indicated that SCORPIO could predict outcomes significantly better than the two current biomarkers approved by the US Food and Drug Administration.
The researchers compared SCORPIO with other machine learning models that predict ICI efficacy in patients using advanced pathology, radiology, and genomic data. SCORPIO performed as well as these models, even in the absence of such advanced data.
“[These] results look pretty promising,” said Rahul Siddharthan, a computational biologist at The Institute of Mathematical Sciences who was not involved with the study. “Even if [SCORPIO] is as good as existing models, it has the advantage of being low cost because you're just using data that is there in the hospital database already, and you're not doing some expensive assay to further [conduct] some biopsy, tumor sample analysis or sequencing analysis.”
“There is a future in exploring these [AI-based] methods,” said Siddharthan. Other cases whether patients have to undergo expensive treatment could benefit from similar approaches, he noted. “This is the future of personalized medicine, but it has to be done cautiously.”
Both Chowell and Morris envision that the model can be applied in healthcare settings to help physicians determine whether ICIs are the best treatment option for a patient. However, Chowell noted that one of the limitations of their model is that they tested it retrospectively, and that it could benefit from being tested in an ongoing clinical trial.
“This is the first version,” said Morris. The model needs further refinement, and they need to include additional data sets from more diverse healthcare settings around the world, he noted. “If the true dream is global access, then there's still work to do.”
- Leach DR, et al. Enhancement of antitumor immunity by CTLA-4 blockade. Science. 1996;271(5256):1734-1736.
- Havel JJ, et al. The evolving landscape of biomarkers for checkpoint inhibitor immunotherapy. Nat Rev Cancer. 2019;19(3):133-150.
- Yoo SK, et al. Prediction of checkpoint inhibitor immunotherapy efficacy for cancer using routine blood tests and clinical data. Nat Med. 2025:1-12
