Development and temporal validation of a preoperative prediction model for postoperative Richmond Agitation-Sedation...

Accurate preoperative prediction of postoperative neurobehavioral instability remains an important unmet need in perioperative medicine. Existing prediction models often rely on intraoperative variables, specialized biomarkers, or limited validation, reducing their usefulness during routine preoperative assessment. We developed and temporally validated a multivariable prediction model for postoperative Richmond Agitation-Sedation Scale alterations (RASSa) using routinely available preoperative variables. We conducted a retrospective observational cohort study including consecutive adults undergoing elective orthopedic surgery (EOS) at a tertiary referral hospital between 2018 and 2024. The model was developed in a chronological derivation cohort using routinely collected demographic and laboratory variables and evaluated by tenfold cross-validation, 2000 bootstrap resamples, and independent temporal validation. Model performance was assessed through discrimination, calibration, prediction error, precision-recall analysis, and decision curve analysis. Among 46,804 screened procedures, 41,010 patients met the eligibility criteria. The development cohort comprised 34,271 patients and the temporal validation cohort 6739 patients. RASSa occurred in 377 patients. The final model included age, sex, preoperative hemoglobin, neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, lymphocyte-to-monocyte ratio, systemic immune-inflammation index, systemic inflammation response index, and a two-component representation of the basophil-to-lymphocyte ratio. Internal validation showed minimal optimism, whereas temporal validation demonstrated good discrimination (AUROC 0.892), satisfactory calibration (intercept 0.132; slope 0.861), low prediction error (Brier score 0.013), and consistent utility. A prediction model based exclusively on routinely available preoperative variables provided individualized estimation of postoperative RASSa after EOS. It demonstrated stable temporal performance and may support perioperative risk stratification. External validation is required before routine implementation.