Frailty and Intraoperative Hypoxemia Risk in Elderly ERCP Patients

This dual-cohort study investigates the predictive value of preoperative frailty for intraoperative hypoxemia in patients aged 65 years and older undergoing elective endoscopic retrograde cholangiopancreatography (ERCP) under monitored anesthesia care at China-Japan Friendship Hospital, with a total target enrollment of 701 patients (313 retrospective and 388 prospective). Intraoperative hypoxemia, defined as pulse oxygen saturation (SpO₂) \< 90% for 5 seconds or longer from anesthesia induction to procedure completion, serves as the primary outcome, while secondary outcomes include severe hypoxemia, postoperative hypoxemia, time to oxygen discontinuation, hemodynamic instability, anesthesia- and procedure-related adverse events, respiratory depression, and apnea. Frailty will be assessed primarily using the Frailty Index (FI) as the core instrument, with additional secondary screening performed using the Modified Frailty Index-11 (mFI-11), Edmonton Frail Scale (EFS), Clinical Frailty Scale (CFS), FRAIL questionnaire, and Fried frailty phenotype. Eligible patients are aged 65 years or older with ASA class II-IV status, scheduled for elective ERCP with a planned procedure duration of less than 60 minutes and complete intraoperative SpO₂ monitoring data; those requiring emergency surgery, severe anemia, pre-existing severe respiratory or hemodynamic instability, or impaired consciousness will be excluded. The retrospective phase will use existing electronic health records to identify risk factors and construct a preliminary prediction model, with propensity score matching, univariate and multivariable logistic regression, bootstrap internal validation, and random forest for feature importance ranking. The prospective phase will prospectively collect standardized data, validate the association between frailty and hypoxemia using adjusted logistic regression, refine the prediction model, compare the performance of different frailty scales using Delong's test, and evaluate machine learning models with SHapley Additive exPlanations (SHAP) values for interpretability. The final model will be tested in an internal validation cohort for discriminative ability, calibration, and clinical utility. All statistical analyses will be performed using R and SPSS software at a two-sided significance level of α = 0.05, with multiple imputation for missing data and Holm-Bonferroni correction for multiple comparisons.