Predicting pediatric postoperative pain trajectories : an intraoperative HRV-based analytics approach

Postoperative pain in children is common and difficult to predict from preoperative information alone. This thesis evaluated whether intraoperative heart rate variability (HRV), a non-invasive marker of autonomic nervous system activity, improves prediction of postoperative pain beyond routinely collected perioperative clinical data. Using linked data from the Pediatric Pain Prediction study and the BC Children's Hospital Vital Signs Data Bank, HRV features were extracted from four intraoperative windows (pre-incision, post-incision, closing, and emergence) and combined with perioperative clinical variables to predict two outcomes: high postoperative opioid consumption in the post-anesthesia care unit (Aim 1) and pain outcomes at postoperative day 30, including worst pain, average pain, and pain interference (Aim 2). Ridge logistic regression and random forest models were trained with repeated cross-validation and SMOTE-ENN resampling to address class imbalance. Performance was compared across HRV-only, perioperative-only, and combined feature sets. Combined models achieved the highest discrimination for PACU opioid consumption (AUC 0.82) and worst pain at postoperative day 30 (AUC 0.90). However, the incremental benefit of adding HRV to perioperative data was modest and did not reach statistical significance by DeLong's test for any of the four outcomes evaluated. Perioperative features, particularly surgical and demographic variables, dominated feature importance rankings across models. A sensitivity analysis removing raw heart rate features showed that HRV's contribution was not attributable to heart rate level alone, suggesting a modest, heart-rate-independent autonomic signal. These findings indicate that intraoperative HRV carries some predictive information for postoperative pediatric pain outcomes, but its practical value beyond existing perioperative data remains limited.