Simple logistic regression matches complex ML for predicting PD-L1 high expression on FDG PET/CT in resectable NSCLC

In an external validation set, an FDG PET/CT-based logistic regression model predicted PD-L1 high expression (TPS ≥50%) in resectable NSCLC with an AUC of 0.83, comparable to SVM (0.86) and random forest (0.85) (all adjusted p=1.000). The interpretable nomogram offers a non-inva…

• Retrospective dual-center study of 269 stage IB-IIIB NSCLC patients (216 training, 53 external validation) using 22C3 IHC as reference.
• DeLong's test showed no significant AUC advantage of complex ML over logistic regression; an easy-to-use nomogram with SUVmax, smoking, histology, T stage, and grade was built.
• Limited by small external validation set and retrospective design; prospective validation is needed.

RadPigeon summaries are original and for information only. They are not clinical advice.