Integrating machine learning and pathway modelling to explore factors associated with chronic post-surgical pain and quality of life: a secondary observational analysis of the ENIGMA-II trial

Background Chronic post-surgical pain (CPSP) is a common long-term complication with multifactorial contributors, and improved risk stratification remains a clinical priority. This study aimed to identify factors associated with CPSP and related outcomes and to explore hypothesised multivariable pathways using a hybrid analytical approach. Methods We performed a secondary observational analysis of 2999 participants from the Evaluation of Nitrous Oxide in the Gas Mixture for Anaesthesia (ENIGMA)-II CPSP substudy with 12-month postoperative assessment. Grouped lasso (GLASSO) was used for penalised feature selection across prespecified predictor groups, and partial least squares structural equation modelling (PLS-SEM) was applied within a theory-informed framework to explore associations among measured domains and outcomes. Analyses were exploratory and did not establish causal effects. Results GLASSO models showed moderate discrimination/explanatory power (CPSP area under the curve 0.69, 95% confidence interval 0.64–0.74; EQ-5D R 2 =0.34; pain severity R 2 =0.29; pain interference R 2 =0.24). In PLS-SEM for CPSP, presurgical pain showed the strongest association (B=0.348), with smaller associations for demographics (B=0.169) and surgical aspects (B=0.091) and lifestyle patterns showed a small inverse association (B=−0.051). For EQ-5D, lifestyle was negatively associated (B=−0.248), and NSAID-related treatment patterns showed a small association (B=0.079). No clear associations were observed between dexamethasone or NSAID treatment patterns and CPSP. Conclusions A GLASSO–PLS-SEM workflow provided an exploratory approach to identifying key features and mapping hypothesised multivariable associations in CPSP-related outcomes. External validation and richer psychosocial/structural measures are needed before clinical translation. Trial registration ClinicalTrials.gov NCT00430989.