Development and validation of a nomogram and preoperative early-recurrence score for pancreatic ductal adenocarcinoma

Pancreatic ductal adenocarcinoma (PDAC) frequently recurs within 12 months despite curative-intent resection, indicating biological aggressiveness not fully captured by current staging. This study aimed to develop and validate a Transparent Reporting of a multivariate prediction model for Individual Prognosis or Diagnosis (TRIPOD)-compliant preoperative Early-Recurrence Score (ERS) for ≤ 12-month recurrence. We conducted a retrospective prognostic-model study at a tertiary referral center. Of 213 patients assessed for upfront resection (2018–2022), 127 underwent pancreatectomy, with 68 (53.5%) experiencing recurrence within 12 months. All 127 patients were retained in the development cohort for model training and internal validation. A temporal validation cohort ( n = 32) comprised consecutive patients treated July–December 2022 and was used to assess exploratory temporal stability. Early recurrence was defined as radiologic or histopathologic evidence of relapse within 12 months post-surgery. A penalized multivariable logistic model with 10-fold cross-validation and bootstrap optimism correction was fitted to derive the ERS and a nomogram; complementary Cox and Fine–Gray time-to-event models were fitted as sensitivity analyses. Calibration and discrimination (AUC) were assessed in both the development and temporal validation cohorts. Competing-risk analyses treated death without recurrence as a competing event. Three preoperative features independently predicted ≤ 12-month recurrence: CA19-9 (per doubling, aOR 1.62), CT-defined necrosis ≥ 30% (aOR 2.43), and peritumoral invasion (aOR 2.08) (all P < 0.011). The ERS demonstrated good and reproducible discrimination: the optimism-corrected AUC in the development cohort was 0.80 (95% CI, 0.72–0.87), and the AUC in the temporal validation cohort was 0.78 (95% CI, 0.64–0.89). Calibration was near ideal across cohorts (slopes 0.88–0.93). Compared to AJCC stage, ERS showed numerically higher discrimination estimates in the development and temporal cohorts (ΔAUC + 0.10), but neither comparison reached statistical significance ( P = 0.08 and P = 0.09, respectively). These head-to-head findings are hypothesis-generating and require multicenter external validation before any inference of clinical superiority can be drawn. Competing-risk analyses confirmed recurrence as the dominant failure mode, with increasing cumulative incidence across ERS strata (low, intermediate, high risk). Decision-curve analyses showed favorable net benefit for ERS-guided management compared with treat-none and AJCC stage across threshold probabilities (0.05–0.20); the treat-all strategy showed competitive net benefit at lower thresholds, reflecting the high baseline event rate. This parsimonious preoperative ERS, integrating log2-transformed CA19-9, CT-defined tumor necrosis, and peritumoral invasion, provides calibrated and transportable estimates of ≤ 12-month PDAC recurrence. It may offer incremental discrimination beyond AJCC stage and shows preliminary evidence of clinical utility across practical decision thresholds. However, these findings are hypothesis-generating from a single-center, small-sample cohort; definitive assessment of clinical utility requires prospective multicenter external validation.