A multi-dimensional model for predicting opioid response in chronic pain

This study aimed to develop and validate a multi-dimensional prediction model integrating clinical, genetic, epigenetic, and biochemical data to individually predict the risk of inadequate analgesic response to opioid therapy in chronic non-cancer pain patients. The goal was to enable early identification of low responders and optimize therapeutic decision-making. A total of 360 chronic pain patients on stable opioid therapy were retrospectively included between January 2022 and December 2024. Patients were randomly allocated into training ( n =252) and validation ( n =108) sets in a 7:3 ratio. Univariate analysis and Least Absolute Shrinkage and Selection Operator (LASSO) regression were used for variable screening and core feature selection. Four machine learning models—Logistic Regression, Gradient Boosting Machine, Random Forest, and Convolutional Neural Network—were constructed. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. Interpretability was assessed using SHapley Additive exPlanations (SHAP) values. The overall inadequate response rate was 40.00% (144/360), with rates of 40.08% (101/252) in the training set and 39.81% (43/108) in the validation set. LASSO regression selected six predictors: age, baseline pain intensity, OPRM1 A118G genotype, OPRM1 promoter methylation rate, tryptophan/kynurenine ratio, and opioid-induced constipation. In the training set, the Logistic Regression model achieved an AUC of 0.799 (95% CI 0.735–0.863), and at a risk threshold of 0.30, it provided a net benefit of 0.41. In the independent validation set ( n =108), the Logistic Regression model yielded an AUC of 0.706 (95% CI 0.582–0.831), while the Convolutional Neural Network model achieved an AUC of 0.766 (95% CI 0.653–0.878); the DeLong test indicated no statistically significant difference between the models ( P >0.05). Given its highest AUC in the training set (0.799), favorable calibration, greater net benefit on decision curve analysis at clinically relevant thresholds, and superior clinical interpretability, Logistic Regression was selected as the final model. SHAP analysis identified baseline pain intensity as the most critical predictor. A multi-dimensional prediction model for opioid analgesic response was developed and internally validated. The model integrates six key clinical and biological factors and shows acceptable discrimination and net clinical benefit in a single-center cohort. However, prospective multi-center validation is needed before this tool can be recommended for routine clinical use to support personalized opioid therapy decisions in chronic pain patients.