Researchers developed GUARD-LGG, a machine learning model designed to predict upfront radiotherapy response in patients with lower-grade glioma.
The GUARD-LGG model was evaluated across five independent validation cohorts totaling 187 patients as well as a multi-center prospective cohort of 328 patients.
The GUARD-LGG model achieved area under the curve values of 0.946 in internal validation, 0.929 and 0.882 in external validation cohorts, and 0.913 in a prospective cohort.
The GUARD-LGG model was trained on a primary dataset containing 963 lower-grade glioma patients.
Biological analysis in the study showed that predicted non-responsiveness to radiotherapy was significantly associated with high expression of radiotherapy resistance pathways.
Patients predicted by GUARD-LGG to be responsive to radiotherapy exhibited significantly longer progression-free survival.
Radiotherapy remains a core treatment for lower-grade glioma, but significant variation in patient response underscores the need for upfront predictive tools.
The GUARD-LGG model identified key predictive features including tumor sub-region volume, T2-weighted heterogeneity, and T1-weighted contrast enhancement changes.
The GUARD-LGG study published in npj Digital Medicine on October 20, 2026 was led by corresponding authors Haris I. Sair of Johns Hopkins University and Yinyan Wang of Beijing Tiantan Hospital.
The GUARD-LGG research involved researchers across international institutions including Beijing Tiantan Hospital, Johns Hopkins University, and the University of Sydney.