Utilizing predictive AI to enhance cancer immunotherapy

Understanding ICIs and its challenges

According to the National Cancer Institute, ICIs are a type of immunotherapy in which medications are used to block immune checkpoints that prevent strong immune responses in the body. Immune checkpoints engage when T-cell proteins on the surface of immune cells bind to partner proteins on other cells, telling the body to stop producing an immune response that could kill healthy cells.

However, when checkpoint proteins bind with partner proteins on tumor cells, the immune checkpoint engages, preventing a strong immune response to kill the cancer cells. Thus, ICIs override the immune checkpoints, allowing the body to provide a strong enough response to kill cancer cells.

"ICIs have revolutionized cancer treatment, offering durable responses for some patients," Chowell said. "However, ICIs do not benefit everyone, and ineffective treatments can lead to unnecessary toxicities, wasted resources and increased healthcare costs."

Variable patient responses to ICIs are one of the primary challenges of using this immunotherapy to treat cancer. Multiple studies have attempted to determine the reason behind these unpredictable responses to ICIs and other forms of immunotherapy. For instance, a study published in 2021 theorized that inherited genetic variations influence the likelihood of benefitting from ICIs.

Despite research in this area, the ability to accurately predict cancer patient responses to ICI remains elusive. Additionally, Chowell said that other biomarkers commonly used to predict patient responses, like tumor mutational burden (TMB) and programmed cell death ligand 1 (PD-L1), are "costly, technically demanding and limited in their reliability across cancer types and healthcare settings."

Thus, there is an urgent need for low-cost, easy-to-implement tools to predict responses to ICIs.

Creating the new AI tool

To create an immunotherapy response prediction tool that is affordable and easy to adopt, the Mount Sinai and Memorial Sloan Kettering research team turned to AI.

SCORPIO represents a major step toward democratizing access to precision oncology.

As described in Nature Medicine, they developed and evaluated the SCORPIO tool, which utilizes routine blood tests and clinical data to predict ICI efficacy and overall survival. The researchers developed the tool by retrospectively collecting data from 2,035 patients across 17 cancer types treated with ICIs between 2014 and 2019 at Memorial Sloan Kettering Cancer Center.

The tool was created using the training set, which included 1,628 patients. Chowell explained that the tool uses machine learning to identify patterns that correlate with treatment outcomes, enabling it to make predictions without requiring specialized assays.

The researchers tested the model on two internal test sets comprising 2,511 patients across 19 cancer types. They found that the SCORPIO tool outperformed TMB for predicting the overall survival of patients undergoing ICI treatment and the clinical benefit of ICI for these patients.

Further, the researchers externally validated the tool using 10 global phase 3 trials that comprised 4,447 patients across six cancer types and a real-world cohort from the Mount Sinai Health System that included 1,159 patients across 18 cancer types. They found that SCORPIO maintained its efficacy in predicting ICI outcomes, surpassing PD-L1 staining.

"The study comprised nearly 10,000 patients across 21 cancer types and [the tool] was validated on both clinical trial and real-world datasets, confirming its reliability, generalizability, and potential to improve precision oncology," Chowell said.

Incorporating the tool into clinical care

Following the internal and external validation of the SCORPIO tool, researchers plan to conduct prospective validation in various clinical environments. Chowell noted that this would enable the research team to confirm its predictive accuracy and real-world applicability, a crucial first step to implementing the tool in clinical care settings.

However, the implementation process would require health systems to consider numerous factors. For instance, Chowell said health system leaders must ensure the tool is seamlessly integrated into existing workflows to minimize disruption. This would include ensuring data compatibility between the tool and EHRs, providing adequate training for clinicians and establishing clear guidelines for using the tool.

From a patient care perspective, health system leaders must communicate the tool's predictions transparently and ethically, ensure equitable access across diverse populations and conduct continuous monitoring to improve the tool's performance, Chowell added.

SCORPIO is not the only AI tool developed and validated for use in cancer immunology, but it does offer some advantages over other tools

Chowell emphasized that because SCORPIO uses easily available and low-cost clinical data like routine blood tests, it offers an alternative to genomic data that could more easily be adopted in resource-limited settings. Additionally, the tool supports treatment personalization, reducing the risks and costs associated with ineffective therapies, and offers insights into immune system dynamics, which could inform future research.

"SCORPIO represents a major step toward democratizing access to precision oncology," Chowell said.

Anuja Vaidya has covered the healthcare industry since 2012. She currently covers the virtual healthcare landscape, including telehealth, remote patient monitoring and digital therapeutics.