Development and validation of a machine learning model for predicting postherpetic neuralgia risk

Area of Science:

• Computational medicine
• Epidemiology
• Machine learning in healthcare

Background:

• Postherpetic neuralgia (PHN) is a significant complication of herpes zoster (HZ).
• Early identification of patients at high risk for PHN is crucial for effective intervention.
• Developing predictive models can aid in timely clinical management.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting PHN risk following HZ onset.
• To identify key clinical factors associated with PHN development.
• To create a tool for early risk stratification of HZ patients.

Main Methods:

• Retrospective analysis of two prospective cohorts (n=627 training, n=219 validation).
• Feature selection using LASSO regression and Boruta algorithm.
• Development and evaluation of ten ML models using AUC, accuracy, sensitivity, F1 score, calibration, and clinical utility.
• Model interpretation via SHapley Additive exPlanations (SHAP).

Main Results:

• PHN incidence was 19.0% in the training cohort and 22.8% in the validation cohort.
• Five key risk factors identified: age, timing of antiviral therapy, acute pain severity, prodromal pain, and diabetes.
• XGBoost model demonstrated superior performance with AUCs of 0.826 (training) and 0.840 (validation).
• SHAP analysis confirmed age as the most significant predictor.

Conclusions:

• The XGBoost model effectively predicts PHN risk using five accessible clinical factors.
• This ML tool facilitates early risk stratification for HZ patients.
• The model supports personalized management strategies for preventing PHN.