A machine-learning risk prediction model constructed from routine clinical laboratory test indicators can effectively identify individuals at heightened risk of developing nasopharyngeal carcinoma, according to a peer-reviewed research study published in the open-access European Journal of Medical Research on October 22, 2026.
The study authors underscored that early screening for nasopharyngeal carcinoma is of vital clinical importance for substantially improving the survival rates of diagnosed patients. In regions with a high incidence of nasopharyngeal carcinoma, Epstein-Barr virus testing currently serves as the predominant primary screening modality. Nonetheless, conventional Epstein-Barr virus screening protocols present substantial practical hurdles, characterized by a relatively high false-positive rate and high economic costs, which together create notable barriers to widespread, cost-effective screening across broad populations in medium- and high-risk zones.
To address the clinical limitations associated with existing diagnostic approaches, the researchers aimed to use machine-learning methods to build a predictive model for nasopharyngeal carcinoma based entirely on routine laboratory test results. The primary objective of this modeling strategy was to facilitate personalized screening protocols tailored to populations located in medium- and high-risk geographic regions.
The empirical investigation analyzed clinical data gathered from a total of 9,957 participants across two major medical centers. The primary cohort was established at Wuzhou Red Cross Hospital, comprising 3,660 patients with nasopharyngeal carcinoma and 3,576 control individuals. The dataset from Wuzhou Red Cross Hospital was subsequently partitioned into an internal training set and an internal test set for model training and initial evaluation. For external validation, an independent cohort was recruited from The People’s Hospital of Guangxi Zhuang Autonomous Region, consisting of 1,446 patients with nasopharyngeal carcinoma and 1,275 controls, serving to test the generalizability of the algorithms in an external patient population.
During the diagnostic evaluation process, the research team examined 22 candidate predictor variables obtained from routine patient examinations. This initial variable pool consisted of 19 routine laboratory measurements along with three derived ratios: the neutrophil-to-white blood cell ratio (NEUT/WBC), the neutrophil-to-lymphocyte ratio (NEUT/LYMPH), and the albumin-to-alkaline phosphatase ratio (ALB/ALP). To perform rigorous predictor selection and remove uninformative or redundant parameters, the investigators applied least absolute shrinkage and selection operator (LASSO) regression, which successfully screened the candidate parameters down to six core predictive variables: gender, age, hemoglobin (HGB), lymphocyte count (LYMPH), calcium (Ca), and magnesium (Mg).
Following feature selection, the research team developed eight distinct machine-learning algorithms using the core predictors: Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Naive Bayes (NB), K-Nearest Neighbor (KNN), and Multi-Layer Perceptron (MLP). To comprehensively determine and compare the classification capabilities of each algorithmic architecture, the researchers evaluated model performance across four key statistical metrics: the area under the curve (AUC), accuracy (ACC), recall, and F1 scores.
Within the training set derived from Wuzhou Red Cross Hospital, the random forest architecture achieved an area under the curve of 0.99, with a 95% confidence interval of 0.98 to 0.99. In the internal test cohort, the random forest model maintained strong discriminative ability, recording an area under the curve of 0.93 with a 95% confidence interval of 0.92 to 0.94. Furthermore, when assessed on the independent external-validation set from The People’s Hospital of Guangxi Zhuang Autonomous Region, the random forest model maintained robust classification performance, yielding an area under the curve of 0.83.
To assess model interpretability and determine the relative influence of each predictive indicator, the authors used Shapley values derived through SHAP analysis. The SHAP analysis revealed that lymphocyte count (LYMPH) contributed most significantly to the model’s risk predictions, followed in descending order of contribution by gender, age, calcium (Ca), magnesium (Mg), and hemoglobin (HGB). The authors concluded that constructing a prediction model based on routine physical-examination indicators using machine learning may help identify individuals at increased risk of nasopharyngeal carcinoma and support early detection and targeted intervention in medium- and high-risk areas.
Subject of Research: Machine learning prediction of nasopharyngeal carcinoma using routine laboratory indicators
Article Title: Building a prediction model for nasopharyngeal carcinoma based on machine learning using laboratory test indicators: a study targeting populations in high-risk areas
Article References: Ye, Y., Lu, J., Zhou, C., Lin, S., Pan, L., & Pan, S. (2026). Building a prediction model for nasopharyngeal carcinoma based on machine learning using laboratory test indicators: a study targeting populations in high-risk areas. European Journal of Medical Research. https://doi.org/10.1186/s40001-026-05369-7
Image Credits: AI Generated
DOI: 10.1186/s40001-026-05369-7
Keywords: nasopharyngeal carcinoma, machine learning, random forest, laboratory indicators, cancer screening, LASSO regression, SHAP analysis, disease risk prediction, Building, prediction, model, nasopharyngeal
News Source: Nathaniel Bowman. (November 10, 2026). Machine Learning Model Predicts Nasopharyngeal Carcinoma Using Routine Laboratory Tests. Scienmag.
