Development and internal validation of explainable machine learning models for predicting failure of first ambulation...

Early ambulation is a notable component of postoperative recovery after cesarean delivery, but some women fail to complete first ambulation within the early postoperative period because of pain, dizziness, orthostatic discomfort, residual lower limb numbness, nausea or vomiting, hypotension or poor activity tolerance. Early identification of women at high risk of ambulation failure may support timely and individualized nursing interventions. The present study aimed to develop and internally validate explainable machine learning models for predicting failure of first ambulation within 12 h after cesarean delivery using routinely collected obstetric perioperative nursing assessment data. The present single-center retrospective cohort study included relatively stable women who underwent cesarean delivery between January 2022 and December 2025 and were considered clinically eligible for early mobilization. The primary outcome was failure of first ambulation within 12 h after surgery, defined as no documented successful ambulation within 12 h or discontinuation of an ambulation attempt because of nursing-documented intolerance. Candidate predictors included demographic, obstetric, pregnancy-related, laboratory, anesthetic, surgical and early postoperative nursing assessment variables. Postoperative symptoms and vital signs were assessed within predefined early postoperative intervals, whereas first turning time was treated as a dynamically updated indicator that became available when the first spontaneous or assisted turning event was documented. The model was therefore interpreted as a dynamic early postoperative risk-assessment approach rather than a prediction model applied exclusively at a fixed 4-h time point. The final analytic dataset contained no item-level missing values; therefore, no imputation was performed. The cohort was randomly divided into training and internal validation cohorts at a ratio of 7:3. Logistic regression, random forest, support vector machine, Extreme Gradient Boosting (XGBoost) and LightGBM models were developed and compared. Model performance was assessed using discrimination, classification metrics, calibration, Brier score and decision curve analysis. SHapley Additive exPlanations (SHAP) analysis was used to interpret the final model. A total of 1,486 women were included, and 326 women experienced failure of first ambulation within 12 h, with an incidence of 21.9%. The training and validation cohorts included 1,040 and 446 women, respectively. In the validation cohort, LightGBM achieved an area under the receiver operating characteristic curve (AUC) of 0.809 and a Brier score of 0.132, whereas XGBoost achieved an AUC of 0.805 and a Brier score of 0.134. The paired DeLong test showed no statistically significant difference between the validation AUCs of the two models (P=0.647). XGBoost achieved higher sensitivity, negative predictive value and F1 score than LightGBM (0.714 vs. 0.653, 0.906 vs. 0.892 and 0.567 vs. 0.561, respectively). Because these metrics were particularly relevant to identifying high-risk women and minimizing missed high-risk cases, XGBoost was selected as the final model. Calibration curves showed acceptable agreement between predicted and observed risks, and decision curve analysis demonstrated net benefit across relevant threshold probabilities. The 10 highest-ranked predictors according to mean absolute SHAP value were postoperative pain NRS score at 2 h, emergency cesarean delivery, time to first documented turning, lower limb numbness within 0-4 h, systolic blood pressure at 2 h, body mass index, preoperative hemoglobin, age, intraoperative fluid volume and dizziness or orthostatic discomfort within 0-4 h. In conclusion, in the present single-center retrospective cohort of relatively stable women considered clinically eligible for early mobilization, the explainable XGBoost model showed promising internal performance for predicting failure of first ambulation within 12 h after cesarean delivery. The model should not be extrapolated to women transferred to intensive care, women with severe postpartum hemorrhage, women with explicit medical restrictions on ambulation or women with severe postoperative complications. These findings support further evaluation but do not establish clinical utility, transportability or readiness for implementation. Temporal and external validation, followed by prospective assessment of workflow feasibility and clinical impact, is required before clinical use can be considered.