Accessible AI Solution Improves Immunotherapy Success Rates

Leveraging routine blood tests and advanced AI, SCORPIO sets a new benchmark in predicting immunotherapy outcomes, promising affordable and equitable cancer care worldwide.

Research: Prediction of checkpoint inhibitor immunotherapy efficacy for cancer using routine blood tests and clinical data. Image Credit: Connect world / Shutterstock

Doctors worldwide may soon have access to a new tool that could better predict whether individual cancer patients will benefit from immune checkpoint inhibitors - a type of immunotherapy - using only routine blood tests and clinical data.

A team of researchers from Memorial Sloan Kettering Cancer Center (MSK) and the Tisch Cancer Institute at Mount Sinai developed the artificial intelligence-based model SCORPIO.

The model is not only cheaper and more accessible, but it also achieves significantly higher predictive accuracy than the two current biomarkers approved by the U.S. Food and Drug Administration (FDA), according to findings published January 6 in the journal Nature Medicine.

"Immune checkpoint inhibitors are a very powerful tool against cancer, but they don't yet work for most patients," says study co-senior author Luc Morris, MD, a surgeon and research lab director at MSK. "These drugs are expensive, and they can come with serious side effects."

Dr. Morris says the key is patient selection—matching the drugs with patients most likely to benefit from them.

"There are some existing tools that predict whether tumors will respond to these drugs, but they tend to rely on advanced genomic testing that is not widely available around the world," he adds. "We wanted to develop a model that can help guide treatment decisions using widely available data, such as routine blood tests."

SCORPIO demonstrated robust performance in large datasets, achieving a median area under the curve (AUC) of 0.763 to 0.759 for predicting overall survival at 6, 12, 18, 24, and 30 months, far exceeding the AUC values of 0.503 to 0.543 for tumor mutational burden. It also outperformed PD-L1 immunohistochemistry in predicting clinical benefit.

Collaborating to Make Checkpoint Inhibitor Therapy Work for More Cancer Patients

Checkpoint inhibitors target the immune system rather than the cancer itself. These drugs work by taking the brakes off immune cells, allowing them to better fight cancer. MSK clinicians and scientists played a key role in bringing the new class of drugs to patients.

Dr. Morris and Diego Chowell, PhD, an Assistant Professor of Immunology and Immunotherapy, Oncological Sciences, and Artificial Intelligence and Human Health at the Icahn School of Medicine at Mount Sinai and a former postdoctoral fellow at MSK, jointly oversaw the new study.

Q&A with Dr. Morris

We spoke with Dr. Morris about the team's prediction model and the next steps for the research:

Why did you develop this new model to predict checkpoint inhibitor response? It was clear there was room for improvement.

There are currently two FDA-approved biomarkers for predicting response to checkpoint inhibitors: tumor mutational burden (the number of mutations in a tumor) and PD-L1 immunohistochemistry (evaluating the expression of the programmed death-ligand 1 protein in tumor samples).

Both require tumor samples to be collected. Meanwhile, genomic testing to assess mutations is expensive and not available everywhere, and the evaluation of PD-L1 expression is highly variable.

Instead, our model relies on readily available clinical data, including routine blood tests performed in clinics around the world—the complete blood count and the comprehensive metabolic profile. This simplicity ensures more equitable access to care, significantly reduces costs, and delivers personalized treatment by predicting outcomes with precision.

SCORPIO’s performance varied across cancer types but consistently outperformed existing tools. For example, in bladder cancer, it achieved a median AUC of 0.809 in real-world cohorts, surpassing the 0.782 AUC observed in clinical trials.

How was the model developed? Our team initially developed SCORPIO by collecting data from MSK patients because of the length and depth of experience oncologists here have treated patients with these drugs. Collaborating with the team at Mount Sinai, we used a type of artificial intelligence called 'ensemble machine learning,' which combines several tools to look for patterns in clinical data from blood tests and treatment outcomes. The model was developed using a rich resource of retrospective data from more than 2,000 patients from MSK who had been treated with checkpoint inhibitors, representing 17 different types of cancer. The model was then tested using data from 2,100 additional MSK patients to verify that it was able to predict outcomes with high accuracy.

Next, we applied the model to nearly 4,500 patients treated with checkpoint inhibitors in 10 different phase 3 clinical trials worldwide.

Further validation was done with additional data from nearly 1,200 patients treated at Mount Sinai.

The study includes nearly 10,000 patients across 21 different cancer types, representing the largest dataset in cancer immunotherapy to date.

We also conducted detailed interpretability analyses using SHapley Additive exPlanations (SHAP) to understand how features like albumin levels and ECOG performance status contributed to predictions.

We conducted this extensive testing and validation because our goal was to develop a predictive model that would be widely applicable to patients and physicians in different locations.

What are the next steps? We plan to collaborate with hospitals and cancer centers around the world to test the model with additional data from a wider variety of clinical settings. The feedback we receive will help us to continue to optimize the model.

Additionally, work is underway to develop an interface that is readily accessible by clinicians, regardless of where they're located.