A recent study published in Cancer Cell highlights a novel approach utilizing artificial intelligence (AI) to improve patient stratification for urothelial carcinoma, enhancing predictions for responses to PD-1 and PD-L1 checkpoint inhibitors.
Romain Banchereau, PhD, a senior scientist at Genentech and lead author, stated, "This study represents a large integration of molecular and clinical data in randomized trials, paving the way for tailored treatment based on specific molecular subtypes."
While checkpoint inhibitors have transformed cancer treatment, many eligible patients do not benefit from these therapies. Previous attempts to improve patient selection through PD-L1 biomarkers have had limited success.
The new method classifies urothelial carcinoma into four subtypes based on tumor microenvironment profiles, significantly enhancing prediction accuracy for treatment responses compared to traditional PD-L1 biomarker assessments.
Researchers analyzed data from four late-stage clinical trials involving 2,803 patients treated with the PD-L1 inhibitor atezolizumab. They employed machine learning techniques to identify distinct tumor subtypes, offering a rapid and effective patient stratification process.
Banchereau noted that "high throughput AI-based imaging biomarkers can potentially be integrated into routine clinical practice," improving diagnostic efficiency and scalability.
The identified subtypes—luminal desert, stromal, immune, and basal—demonstrated varying responses to treatment, with immune and basal subtypes showing significant survival benefits from atezolizumab.
The study advocates for tailored treatment strategies, suggesting that immune subtype tumors may benefit from combined therapies, while basal subtype tumors may respond better to PD-L1 inhibitors alone or in combination with other treatments.
Banchereau emphasized the potential of digital pathology and deep learning models to enhance understanding of tumor biology and accelerate patient subtyping in clinical settings.