Development and internal validation of a conditional failure-to-rescue prediction model after surgery for metastatic...

The aim of this study was to develop and internally validate an optimal patient-level prediction model for conditional failure to rescue (cFTR) after surgery for spinal metastases.

The authors performed a retrospective analysis of the American College of Surgeons National Surgical Quality Improvement Program database from 2018 to 2023. The primary endpoint was cFTR, defined as 30-day postoperative mortality among patients who experienced at least 1 major complication following surgery for metastatic spinal tumors. Uni- and multivariable logistic regression models were constructed to predict cFTR. Model performance was assessed using discrimination (area under the receiver operating characteristic curve [AUC]), calibration, and information criteria (Akaike and Bayesian). Penalized regression (least absolute shrinkage and selection operator, and elastic net) was performed to evaluate robustness. Internal validation was conducted using 5-fold cross-validation and bootstrap resampling. Reporting of model development followed the TRIPOD statement.

Among 338 patients who experienced at least 1 major postoperative complication, 71 (21%) died within 30 days. Emergency surgery was found to be the strongest predictor of cFTR. In the final 6-predictor clinical model, emergency surgery, current smoking, higher preoperative white blood cell count, lower serum albumin, and functional dependence were associated with increased odds of cFTR. The model demonstrated good discrimination (AUC = 0.740) and acceptable calibration, with minimal optimism on internal validation (cross-validated AUC = 0.713, bootstrap mean AUC = 0.743). Penalized regression models demonstrated slightly higher discrimination but offered limited incremental benefit over the clinical mode. Secondary analyses showed that nonsurvivors experienced a higher burden of complications and disproportionately severe cardiopulmonary and infectious events.

Among patients with metastatic spinal tumors who experienced major postoperative complications, cFTR remained common and was largely driven by patient-level vulnerability. Conditioning on the occurrence of complications highlighted emergency presentation, impaired physiological reserve, systemic inflammation, and functional dependence as key determinants of mortality. This internally validated model provides a transparent framework for risk stratification after complications and identified potentially modifiable targets for perioperative optimization. External validation is needed to establish generalizability and clinical utility.