Refining malignancy risk stratification in cervical lymphadenopathy: development of the LASSO-LAP score (novel...

Accurate preoperative differentiation of cervical lymphadenopathy remains clinically challenging. While systemic immune-inflammatory indices provide prognostic value, their simultaneous diagnostic evaluation is often hindered by multicollinearity. This study aimed to integrate comprehensive hematological and nutritional markers with established clinico-radiological predictors using a penalized machine learning approach to refine our previously developed malignancy risk score.

A retrospective cohort of 420 patients undergoing excisional biopsy for cervical lymphadenopathy was analyzed. Fourteen candidate predictors were initially evaluated for baseline characteristics. However, to prevent mathematical tautology arising from indices derived from identical complete blood count (CBC) parameters, eight structurally redundant indices were excluded. Subsequently, the Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression was employed on the optimized pool of six predictors for final variable selection and regularization. A weighted multiparametric clinical scoring system was subsequently developed, and its diagnostic performance was assessed via Receiver Operating Characteristic (ROC) analysis.

There were 420 individuals in the final development cohort (240 non-malignant and 180 malignant). Six independent variables without multicollinearity were found using LASSO regression: age, male gender, roundness index, conglomeration, Level V involvement, and neutrophil-to-lymphocyte ratio (NLR). With an internal validation area under the curve (AUC) of 0.859 (95% confidence interval (CI): 0.822-0.895), the resulting 11-point risk score showed high discriminating ability. The model correctly classified patients into low (11.1%), moderate (35.7%), and high (85.0%) malignancy risk categories in an independent temporal validation cohort of 57 patients, maintaining strong predictive accuracy (AUC: 0.867). There was no statistically significant diagnostic difference between this simplified, NLR-based model and a more complicated, previously developed multi-index model, according to a matched DeLong test (p = 0.566).

The 11-point clinical scoring system provides a robust and practical method of preoperative malignancy risk stratification for cervical lymphadenopathy. Demonstrating that a streamlined NLR-integrated model achieves the same level of diagnostic accuracy as complex composite indices, this tool avoids mathematical tautology and provides accessible, evidence-based guidance on the optimal timing of biopsies in routine practice.