External Validation and Feedback-Driven Improvement of the Diagnostic Model for Pediatric Cervical Lymphadenopathy.
In an independent cohort of 255 children with cervical lymphadenopathy, the authors externally validated and simplified a diagnostic model from 12 to 9 variables, reporting 94% sensitivity and 91% specificity for identifying children at risk of high-grade lymphoma.
Open original publication →What the AI sees
In an independent cohort of 255 children with cervical lymphadenopathy, the authors externally validated and simplified a diagnostic model from 12 to 9 variables, reporting 94% sensitivity and 91% specificity for identifying children at risk of high-grade lymphoma.
Research significance
Evidence: the revised model retained high diagnostic performance while reducing predictor redundancy; inference: if prospectively validated in representative referral settings, it could accelerate lymphoma referral and diagnosis while reducing unnecessary invasive investigations, but this record does not demonstrate improved treatment or patient outcomes.
Source abstract
Background/Objectives: Early identification of high-grade lymphoma in children presenting with cervical lymphadenopathy is essential to ensure timely referral while avoiding unnecessary invasive procedures. We have previously developed a diagnostic model to support referral decisions. The aim of this study was to validate this model in an independent cohort and improve it based on clinical feedback and new data. Methods: We included 255 pediatric patients with cervical lymphadenopathy, of whom 62.4% were diagnosed with malignancy. External validity was assessed by applying the original scoring model to the new dataset and calculating the sensitivity, specificity, and area under the curve (AUC). Thereafter, the model was updated by addressing overlapping predictors, and evaluating additional variables that were identified through univariate analysis. Results: The original model demonstrated stable performance in the new cohort, with sensitivity and specificity remaining within the previously reported 95% confidence intervals. No new clinically relevant predictors emerged beyond those already included in the original model. Model refinement resulted in a simplified model with 9 variables (reduced from 12), which improved clinical usability by reducing redundancy among predictors and facilitating implementation in routine practice. The revised model showed stable diagnostic accuracy with a sensitivity of 94% and specificity of 91% in the new dataset. Conclusions: Our updated diagnostic nine-factor model maintains high diagnostic performance while having a reduced number of variables. These findings support its use as a practical tool for early identification of children at risk of high-grade lymphoma.