Childhood lymphadenopathy: ultrasonographic predictors of malignancy in a retrospective cohort of 500 patients

TL;DR

This study identifies ultrasound features as strong predictors of malignancy in children with swollen lymph nodes, potentially reducing unnecessary biopsies.

Contribution

The study demonstrates that ultrasonographic features are more reliable than traditional parameters like size or location for predicting malignancy in pediatric lymphadenopathy.

Findings

Ultrasonographic assessment was the only independent predictor of malignancy, with suspicious nodes having ~56-fold higher odds of malignancy.

The model showed excellent discrimination with an AUC of 0.92 for predicting malignancy.

Traditional parameters like node size and region were not significant in multivariable analysis.

Abstract

Pediatric lymphadenopathy is common and usually benign, yet selecting children who need biopsy remains challenging. We evaluated clinical and ultrasonographic (US) predictors of malignancy in a large pediatric cohort.

We retrospectively analyzed 500 children (0-18 years) evaluated for regional lymphadenopathy at a tertiary pediatric hematology-oncology clinic (June 2024-March 2025). Demographics, node location/size/count, US features, viral tests, and biopsy/bone-marrow results were extracted. Univariable tests and multivariable logistic regression identified independent predictors of malignancy. Model performance was assessed with Hosmer-Lemeshow and ROC AUC.

Median age was 6 years; cervical nodes predominated (92.2%), and multiple nodes were frequent (93.6%). Biopsy was performed in 47 children; malignancy was found in 23. Final diagnoses in the cohort were reactive (87.4%), infectious (6.8%), hematologic malignancy (4.4%), and solid tumor metastasis (1.4%). On univariable analysis, node size ≥2 cm, multiplicity, and region were associated with malignancy (p < 0.05). In multivariable analysis, only US assessment remained independently predictive: nodes categorized as suspicious on US had ∼56-fold higher odds of malignancy (adjusted OR ≈ 55.6; 95% CI 14.3-200; p < 0.001), whereas size and region were not significant. The model showed good calibration and excellent discrimination (AUC = 0.92; 95% CI 0.88-0.96).

US features outperformed traditional parameters (size/site) for predicting malignancy. Incorporating US-driven criteria into pathways may reduce unnecessary biopsies while preserving timely cancer detection.

Figures