Diagnostic value of peripheral blood leukocyte parameters, NE-WX, LY-Y, MO-WY, MO-Y, MO-Z, and NE-SFL, in lung cancer.

Introduction: We aimed to assess the diagnostic potential of six leukocyte parameters (NE-WX, LY-Y, MO-WY, MO-Y, MO-Z, and NE-SFL) for lung cancer by analysing 423 patients, 39 benign cases, and 346 healthy controls. Methods: The results of these six leukocyte parameters were compared among groups using the Kruskal-Wallis H test. Multivariate logistic regression was used to analyze correlations between NE-WX, LY-Y, MO-WY, MO-Y, MO-Z, NE-SFL, and lung cancer diagnosis. Patients were randomly grouped into training and testing sets, and nine machine-learning models (support vector machine, gradient boosting machine, artificial neural network, random forest [RF], extreme gradient boosting, K-nearest neighbor, adaptive boosting, light gradient boosting machine, and catboost) were built to compare the diagnostic performances. The DeLong test was adopted to compare the area under the receiver operating characteristic curve (AUC) values among nine machine learning models. Results: These six leukocyte parameter levels differed significantly between the lung cancer and control groups. Multivariate logistic regression revealed a strong predictive value of NE-WX, LY-Y, MO-WY, MO-Y, MO-Z, and NE-SFL for lung cancer. NE-WX and NE-SFL were independent risk predictors. Receiver operating characteristic analysis revealed that MO-Y was the best diagnostic single indicator for lung cancer (AUC = 0.787, specificity=82.0%). The combination of LY-Y, MO-Y, MO-Z, and NE-SFL demonstrated superior diagnostic efficacy (AUC = 0.897, sensitivity=70.6%, specificity=78.9%, Youden index=0.495) compared with any single indicator. The RF model exhibited high consistency between predicted and actual values, demonstrating robust and reliable predictive performance in lung cancer diagnosis. MO-Y and MO-Z were the top predictive biomarkers with the strongest predictive value. Conclusions: NE-WX, LY-Y, MO-WY, MO-Y, MO-Z, and NE-SFL may aid in diagnosis of lung cancer. The RF model demonstrates strong potential for lung cancer detection.