Prediction Models for Acute and Chronic Postoperative Pain in Patients with Cancer: A Systematic Review and...

To evaluate prediction models for acute postoperative pain (AOPP) and chronic postsurgical pain (CPSP) after cancer surgery, prioritizing pain phenotype and cancer type.

Four databases were searched through February 2026. Risk of bias was assessed with PROBAST. AUCs were synthesized on the logit scale using multilevel random-effects models with estimates nested within studies and CR2/Satterthwaite cluster-robust inference. AOPP and CPSP were analyzed separately; cancer-specific estimates were nested within each phenotype, and mixed-pain estimates were secondary. Bivariate random-effects models synthesized sensitivity, specificity, and SROC curves.

Twenty-nine studies reported 53 models. For AOPP, pooled AUCs were 0.83 (95% CI 0.76-0.88; 95% prediction interval [PI] 0.54-0.95) in training and 0.80 (0.76-0.83; PI 0.66-0.89) in validation. For CPSP, corresponding AUCs were 0.79 (0.75-0.83; PI 0.62-0.90) and 0.75 (0.70-0.80; PI 0.50-0.91). Validation sensitivity/specificity were 0.75/0.80 for AOPP and 0.72/0.72 for CPSP. In cancer-specific validation analyses, AOPP gastrointestinal-cancer models yielded AUC 0.80; CPSP estimates were 0.72 for breast, 0.82 for lung, and 0.80 for other cancers, although the lung estimate included only three studies. Heterogeneity remained substantial. Eight studies were at low risk of bias, seven reported external validation, and calibration parameters were insufficient for pooling.

Current models show moderate average discrimination, but wide prediction intervals, high risk of bias, sparse independent validation, and incomplete calibration reporting preclude routine use. AOPP research should prioritize early perioperative prediction and actionable analgesic thresholds; CPSP research requires standardized long-term outcomes and multicenter external validation.