Dynamic prediction model for early postoperative pain based on generalized estimating equations: a multi-timepoint...

To identify the risk factors associated with early postoperative pain (Visual Analogue Scale [VAS] ≥3) one hour after extubation in surgical patients in the intensive care unit (ICU), and to establish a dynamic prediction model using generalized estimating equations (GEE) to support precise analgesic management.

This retrospective longitudinal study was conducted in postoperative ICU patients of the West China Hospital (n=373). Patients were randomly divided into training (70%) and testing sets (30%) for model development and internal validation. An external validation cohort from The People's Hospital of Rugao (n=124) was used to assess generalizability. At 30 minutes, one hour, and two hours post-extubation, clinical, perioperative, and extubation variables were collected. Multivariable GEE modeling was performed based on significant factors identified from univariate analysis. Model performance was evaluated using AUC, accuracy, sensitivity, specificity, Hosmer-Lemeshow test, Brier score, and decision curve analysis (DCA). The discriminative performance of the model was compared with that of the Pain Catastrophizing Scale (PCS) using the DeLong test.

Independent risk factors for early postoperative pain included a higher Critical-Care Pain Observation Tool score at extubation, BMI, smoking history, older age, higher Acute Physiology and Chronic Health Evaluation II (APACHE-II) score, and intraoperative sedative use. Post-extubation analgesic pump use and time elapsed after extubation were protective factors. The model exhibited good discrimination with an AUC of 0.820 in the training set and 0.785 in the testing set. Stable performance was observed during external validation (AUC: 0.772). DCA demonstrated a significant net clinical benefit across a wide range of threshold probabilities. The model performed significantly better than the PCS (p<0.05).

Factors determining early post-extubation pain include patient-related, disease-related, sedation-related, and behavior-related factors. The GEE-based dynamic model provides robust discriminative ability and clinical utility for early identification of high-risk patients, supporting individualized analgesic interventions in the ICU.