Though immunotherapies have made strong inroads in the fight against cancer, knowing which drugs will work the best in which patients—and with the first dose—has remained a challenge.
Researchers at the National Institutes of Health and Memorial Sloan Kettering Cancer Center said that their artificial intelligence-powered tool, built to peruse common and accessible patient information, may help accurately predict who would most likely benefit from the class of treatments.
The AI model incorporates five clinical features gathered from a routine workup and blood testing: starting with the patient’s age, cancer type and list of previous therapies, it adds their blood albumin levels and neutrophil-to-lymphocyte ratio, for measures of organ function and inflammation. The machine learning program, named LORIS, also incorporates DNA sequencing data and the tumor’s mutational burden.
Taken together, researchers at MSKCC and NIH’s National Cancer Institute said their AI approach was able to identify patients more likely to have objective responses to immune checkpoint inhibitors, as well as forecast short- and long-term survival among different cancers.
The results of their proof-of-concept study, which analyzed data gathered from more than 3,700 patients and 18 types of solid tumors, were published in the journal Nature.
According to the NIH, two biomarkers are currently approved by the FDA to help guide checkpoint inhibitor therapies, including measures of tumor mutational burden and PD-L1 status. However, each has shown limitations in the clinic.
While a version of LORIS is available through the NIH website, the researchers said larger prospective studies will be required to assess the AI model’s performance in clinical settings.