AI background, pill
Credit: Just_Super / iStock / Getty Images Plus

In a proof-of-concept study published in Nature Cancer, scientists have shown that their artificial intelligence tool improves the predictive likelihood of a patient responding to immune checkpoint inhibitors. Known as the Logistic Regression-Based Immunotherapy-Response Score (LORIS), the tool relies on six common data points and computes a score predicting whether a patient will respond to immunotherapy.

To build LORIS, scientists at the NIH and Memorial Sloan Kettering Cancer Center constructed and evaluated using data from multiple independent data sets that included 2,881 patients treated with immune checkpoint inhibitors across 18 solid tumor types. They also curated a cohort comprising 841 participants not treated with immunotherapy from 15 solid tumor types.

The test was developed to augment the two available biomarkers that indicate susceptibility to immune checkpoint inhibitors—PD-L1 and tumor mutational burden.

The team made two separate tools—one was a pan-cancer tool, and the other was specific to non-small cell lung cancer (NSCLC). LORIS makes predictions based on five clinical features that are routinely collected from patients: a patient’s age, cancer type, history of systemic therapy, blood albumin level, and blood neutrophil-to-lymphocyte ratio, a marker of inflammation. The model also considers tumor mutational burden (TMB), assessed through sequencing panels.

Their study found that LORIS outperformed previous signatures in predicting immune checkpoint blockade response and identifying responsive patients even with low tumor mutational burden or PD-L1 expression.

“Remarkably, despite using a limited set of features, LORIS matched or exceeded the performance of more complex computational methods, even under stringent testing conditions,” the authors write.

Across several data sets, the pan-cancer LORIS had a 15–68% increase in AUC—a measure of a model’s correct predictability—over the TMB biomarker. The NSCLC-specific LORIS showed 4–17% and 5–23% increases in AUC over the PD-L1 and TMB biomarkers, respectively.

The scientists emphasized three key findings from their study from a translational perspective.

“Firstly, LORIS predicts not only patient immune checkpoint blockade objective response but also short-term and long-term survival benefit following immunotherapy better than existing methods,” they write. “More importantly, our model successfully identifies low-TMB or low-PD-L1 TPS patients who can still benefit from immunotherapy. Lastly, LORIS scores patients by their response probabilities to immunotherapy in a much more consistent manner, leading to more accurate identification of likely responders and more effective exclusion of likely non-responders.” Taken together, LORIS could be a reliable tool for improving clinical decision-making practices in precision medicine.

While a version of LORIS is available through the NIH website, the researchers noted that larger prospective studies are needed to further evaluate the AI model in clinical settings.

Also of Interest