AI model optimizes first-line FOLFIRINOX/NALIRIFOX or gemcitabine/nab-paclitaxel selection for advanced pancreatic cancer, improving survival.

Pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal malignancies, often diagnosed at advanced stages due to a lack of early symptoms. While first-line regimens like FOLFIRINOX/NALIRIFOX offer improved outcomes, they are often reserved for fitter patients, with elderly or comorbid individuals frequently receiving gemcitabine/nab-paclitaxel (gem/nab-p). A critical gap exists in comprehensively optimizing treatment selection between these powerful but distinct regimens, leading to suboptimal outcomes for many patients who might benefit from a more personalized approach.

Researchers developed a propensity score matched, transcriptomic-based AI model using 2202 molecularly-profiled PDAC specimens. This model was designed to provide clinically relevant treatment recommendations and prognostic information, specifically for first-line selection between FOLFIRINOX/NALIRIFOX and gemcitabine/nab-paclitaxel. The model's efficacy was then evaluated in a testing dataset, where patient outcomes were compared based on whether they received the treatment recommended by the AI versus an alternative therapy.

In the testing dataset, patients predicted by the AI model to have superior outcomes on first-line FOLFIRINOX indeed experienced significantly longer time-to-next-treatment (TTNT) and overall survival (OS) when treated with FOLFIRINOX. Specifically, the hazard ratios (HRs) were 0.55 for TTNT and 0.48 for OS, both with high statistical significance (p < 0.001). Patients recommended for gemcitabine/nab-paclitaxel treatment generally had similar outcomes regardless of which regimen they received, but a distinct subset showed improved outcomes specifically on gemcitabine/nab-paclitaxel.

Approximately half of the patients in the testing dataset had received a first-line therapy opposite to the AI model's recommendation, highlighting a significant opportunity for improved decision-making. This suggests that current treatment selection practices often miss opportunities to optimize individual patient outcomes, which the AI model could rectify.