Can the combination of positron emission tomography (PET), computed tomography (CT) and clinical findings enhance preoperative lymph node staging in patients with primary lung cancer?
In a new retrospective study, recently published in European Radiology, researchers developed and compared five nomograms to assess their ability to predict lymph node metastasis (LNM) in 455 patients with primary lung cancer. All patients in the cohort had chest CT and PET/CT and underwent endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA), according to the study. The five nomograms included Clinical-CT-PET, which incorporated clinical, CT and PET findings, Clinical-CT, PET, Clinical-PET and CT-PET models.
In the internal testing cohort, the study authors found that the clinical-CT-PET nomogram offered better risk stratification for LNM prediction than the Clinical-CT and PET models with a higher AUC (85 percent vs. 76.3 percent and 82.5 percent respectively), sensitivity (80 percent vs. 66.3 percent and 75 percent respectively) and specificity (86.9 percent vs. 81.4 percent and 86.4 percent respectively).
“(The clinical-CT-PET) model achieved the highest sensitivity and specificity among all models and showed significantly improved diagnostic capability over models based solely on clinical and CT features or LN SUVmax,” wrote lead study author Xiaoyu Han, M.D., who is affiliated with the Department of Imaging Physics at the University of Texas MD Anderson Cancer Center in Houston, and colleagues.
However, the researchers noted no statistically significant difference between the Clinical-CT-PET nomogram and the Clinical-PET and CT-PET models. The Clinical-PET model provided an 84.5 percent AUC, 80 percent sensitivity and 87.4 percent specificity. The CT-PET nomogram offered an 85 percent AUC, 80 percent sensitivity and 86.9 percent specificity for predicting LNM, according to the study authors.
“ … The inclusion of clinical data may not provide substantial additional benefits in predicting lymph node metastasis,” suggested Han and colleagues.
Three Key Takeaways
• Multimodal nomograms improve LNM risk stratification. A nomogram combining clinical, CT, and PET data (Clinical-CT-PET) demonstrated higher overall diagnostic performance for preoperative lymph node metastasis prediction than models based on Clinical-CT or PET alone, with improved AUC, sensitivity, and specificity.
• PET-CT–based models perform comparably with or without clinical data. The Clinical-CT-PET model did not show a statistically significant advantage over Clinical-PET or CT-PET models, suggesting that adding clinical variables may offer limited incremental benefit beyond combined PET and CT imaging.
• Added value for small lymph node assessment. For lymph nodes smaller than 1 cm, the combined Clinical-CT-PET approach significantly outperformed PET alone, supporting a multimodal strategy to mitigate false positives from SUVmax-only assessment and improve staging accuracy in subtle nodal disease.
In a sub-analysis looking at prediction of small lymph node (< 1 cm) metastasis, the study authors found that the clinical-PET-CT model offered over a 7 percent higher AUC (79.7 percent vs. 72.2 percent) and over a 19 percent higher sensitivity (71.4 percent vs. 52 percent) in comparison to the PET model.
“Inflammatory or fibrotic changes may mimic nodal metastasis on PET, contributing to false-positive predictions. This finding indicated that relying on SUVmax alone was insufficient for evaluating small LNs and that a multimodal approach improves diagnostic accuracy by capturing complementary information,” added Han and colleagues.
(Editor’s note: For related content, see “Meta-Analysis Examines Impact of AI in Radiology for Cancer Detection,” “Prospective Study Shows Capability of CXR AI to Detect Early-Stage Lung Cancer in Young Patients and Never Smokers” and “CT Study Shows No Significant Difference in Long-Term Outcomes Between Surveillance and Surgery for Ground-Glass Nodules.”)
Beyond the inherent limitations of a single-center retrospective study, the authors acknowledged the lack of external validation, pre-selection of patients who had EBUS-TBNA and a lack of assessment for serum biomarkers and genetic alterations that may affect the capacity of the LNM prediction models.
