Development and validation of a nomogram for predicting postoperative complications after pancreaticoduodenectomy: a...

Postoperative complications occur in 30-50% of patients undergoing pancreaticoduodenectomy and critically influence both recovery and oncological outcomes. Existing prediction models target pancreatic fistula in isolation, leaving the full spectrum of postoperative morbidity unaddressed. We developed and internally validated a nomogram to predict any clinically significant complication within 30 days of pancreaticoduodenectomy, integrating anatomical, pathological, surgical, and systemic perioperative variables.

This single-centre retrospective cohort study enrolled consecutive adults undergoing pancreaticoduodenectomy between 2020 and 2024. The primary endpoint was any Clavien-Dindo grade II-IV complication within 30 days. Predictors were identified through a pre-specified hybrid scoring system combining univariate logistic regression, multivariable logistic regression, and LASSO penalised regression with 10-fold cross-validation; variables scoring above 3 were retained. Surgical approach was included as a pre-specified forced predictor. Internal validation employed 1000-iteration bootstrap resampling and 100-iteration repeated stratified cross-validation. Clinical utility was quantified by decision curve analysis.

Of 265 patients (mean age 62.0 years; 55.5% male), 97 (36.6%) developed Clavien-Dindo grade II-IV complications. The nomogram incorporated six data-driven predictors - pancreatic texture, histological diagnosis, preoperative albumin, CRP, anatomical site, and NRS-2002 nutritional score - with surgical approach as a forced inclusion. Pancreatic texture was the dominant predictor (OR 3.49, 95% CI 1.71-7.11; P<0.001); preoperative hypoalbuminaemia independently conferred excess risk (OR 0.93 per g/L; P = 0.016). The model achieved an apparent AUC of 0.747 (bootstrap 95% CI 0.704-0.825), an optimism-corrected AUC of 0.704, and satisfactory calibration (Hosmer-Lemeshow P = 0.686). Repeated cross-validation yielded a mean validation AUC of 0.707 ± 0.063. Decision curve analysis confirmed net benefit across threshold probabilities of 8-87%.

This internally validated nomogram - incorporating pancreatic texture, histological diagnosis, anatomical site, CRP, surgical approach, preoperative albumin, and NRS-2002 - provides clinically actionable perioperative risk stratification for significant complications following pancreaticoduodenectomy, pending prospective external validation. The convergent data-driven selection of two complementary nutritional predictors positions preoperative nutritional optimisation as a modifiable target for complication reduction in this high-risk population.