Association of preoperative frailty, inflammatory markers and CT radiomics with complications after esophagectomy in older adults: a five-year retrospective analysis and online dynamic nomogram development

Objective Older adults with esophageal cancer have poor physiological tolerance and a high risk of postoperative complications. This study examined whether preoperative frailty, hematologic inflammatory indices and CT radiomic features independently predict these complications and built a validated web-based nomogram for personalized risk prediction. Methods This five-year retrospective study included 526 patients aged at least 60 years who underwent radical esophagectomy between 2019 and 2023. Frailty was assessed with the Fried phenotype, five inflammatory indices were calculated from preoperative blood counts, and CT radiomic features were extracted from contrast-enhanced chest CT. Because the five indices share the same four cell counts, their collinearity was diagnosed and the indices were compared through sequential selection, single-index and alternative-model discrimination, reclassification metrics, and order-reversed and bootstrap validation. Independent predictors were identified by two-stage multivariable logistic regression, ranked with SHapley Additive exPlanations (SHAP), and assembled into a nomogram; discrimination, calibration and net clinical benefit were evaluated in training and validation cohorts. Results Complications occurred in 187 of 526 patients (35.55%). The five inflammatory indices were strongly collinear; across all sensitivity analyses the systemic inflammation response index and systemic immune-inflammation index were retained independently of entry order, whereas the neutrophil-, platelet- and monocyte-to-lymphocyte ratios added no incremental information. Frailty, the two retained composite indices and three CT radiomic features (lesion texture heterogeneity, gray-level mean value and shape irregularity) were independent predictors, with frailty carrying the greatest SHAP weight. The nomogram achieved areas under the receiver operating characteristic curve of 0.824 (95% confidence interval 0.780-0.868) and 0.807 (0.746-0.868) in the training and validation cohorts, with good calibration and net clinical benefit, and it was implemented as a web tool. Conclusion Preoperative frailty, inflammatory indices and CT radiomic features are independent predictors of postoperative complications, with distinct contributions revealed by SHAP. The validated web nomogram reliably stratifies risk and supports individualized perioperative management.