Related Experiment Video
Updated: Jun 30, 2026
Modeling Neural Immune Signaling of Episodic and Chronic Migraine Using Spreading Depression In Vitro
Published on: June 13, 2011
Unraveling immune-inflammation-aging network interactions: an interpretable machine learning model predicts the risk
Pei-Pei Kang1, Shi-Jie Bi2, Yan-Ran Kang3
1Pain Department, The Second Affiliated Hospital of Henan University of Science and Technology, Luoyang, Henan, China.
Background:
Postherpetic neuralgia (PHN) is the most common and intractable complication of herpes zoster (HZ). Early and accurate identification of patients at high risk for PHN is crucial for effective interventions. This study aimed to establish a high-performance and interpretable machine learning prediction model.
Methods:
This was a single-center retrospective cohort study that ultimately included 480 patients hospitalized with a diagnosis of HZ at the First Affiliated Hospital of Shandong First Medical University (January to December 2024). Patients were divided into PHN and non-PHN groups. Multidimensional predictors were collected through the electronic medical record system. Integrated feature screening was performed using the Boruta algorithm, random forest, and LASSO regression. Six machine learning models (including XGBoost) were trained and compared using nested cross-validation, and the optimal model was interpreted via the SHAP framework.
Results:
The incidence of PHN was 23.3%. Eight key predictors were identified: age, neutrophil-to-lymphocyte ratio (NLR), absolute lymphocyte count (ALC), serum albumin (ALB), platelet-to-lymphocyte ratio (PLR), absolute eosinophil count (AEC), serum calcium (Ca) and neutrophil-to-platelet ratio (NPR). Among the six models, XGBoost demonstrated optimal performance with an AUC of 0.919 (95% CI: 0.910-0.927) in nested cross-validation. It also showed a sensitivity of 0.836 and a specificity of 0.831. SHAP analysis suggested relatively linear effects for age, ALB, ALC and AEC, whereas NLR, PLR and Ca exhibited more complex, potentially nonlinear associations with PHN risk. Interaction analysis further indicated extensive synergistic effects among these factors, collectively providing a preliminary outline of a potential risk network grounded in "aging," centered on "immune-inflammation," and modulated by nutritional status. An online calculator based on this model has been deployed.
Conclusions:
This study developed and internally validated a high-performance, interpretable PHN risk prediction model. Its interpretable output suggests that the occurrence of PHN may be associated with an imbalance in the immune-inflammation-aging network, generating new hypotheses for its pathogenesis. The accompanying online tool offers research-grade decision support for individualized clinical risk management, pending external validation.
