Abstract
BACKGROUND: The treatment for herpetic-related neuralgia focuses on symptom control by use of antiviral drugs, anticonvulsants, and tricyclic antidepressants. We aimed to explore the clinical characteristics associated with medication responsiveness, and to build a classifier for identification of patients who have risk of inadequate pain management. METHODS: We recruited herpetic-related neuralgia patients during a 3-year period. Patients were stratified into a medication-resistant pain (MRP) group when the pain decrease in the visual analogue scale (VAS) is < 3 points, and otherwise a medication-sensitive pain (MSP) group. Multivariate logistic regression was performed to determine the factors associated with MRP. We fitted four machine learning (ML) models, namely logistic regression, random forest, supporting vector machines (SVM), and naïve Bayes with clinical characteristics gathered at admission to identify patients with MRP. RESULTS: A total of 213 patients were recruited, and 132 (61.97%) patients were diagnosed with MRP. Subacute herpes zoster (HZ) (vs. acute, OR 8.95, 95% CI 3.15-29.48, p = 0.0001), severe lesion (vs. mild lesion, OR 3.84, 95% CI 1.44-10.81, p = 0.0084), depressed mood (unit increase OR 1.10, 95% CI 1.00-1.20, p = 0.0447), and hypertension (hypertension, vs. no hypertension, OR 0.36, 95% CI 0.14-0.87, p = 0.0266) were significantly associated with MRP. Among four ML models, SVM had the highest accuracy (0.917) and receiver operating characteristic-area under the curve (0.918) to discriminate MRP from MSP. Phase of disease is the most important feature when fitting ML models. CONCLUSIONS: Clinical characteristics collected before treatment could be adopted to identify patients with MRP.