Prognostic modeling of overall survival in metastatic pancreatic cancer: an inflammation-based tool validated in PANTHEIA-SEOM cohort

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract Purpose To develop and internally validate the PANTHEIA-SIRI prognostic model, which integrates log-transformed systemic inflammation response index (SIRI) with clinical predictors, to estimate overall survival (OS) in metastatic pancreatic ductal adenocarcinoma (mPDAC) treated with first-line chemotherapy. Methods We used data from the multicenter PANTHEIA-SEOM registry. OS was defined from chemotherapy start. The model was fitted as a Weibull accelerated failure time model in the survival-analysis population with multiple imputation. Predictors were log-transformed baseline SIRI, modeled with restricted cubic splines, ECOG, tumor burden, chemotherapy regimen, and anorexia-cachexia syndrome. Internal validation used a separate, non-overlapping cohort from the same registry; the centers contributing to each cohort are listed in a supplementary annex. TRIPOD was followed. Discrimination was assessed with Harrell´s C-index and calibration with IPCW Brier scores and IPA. Results The derivation cohort comprised 672 patients with SIRI data (593 analyzed for survival) across 22 Spanish hospitals (2015–2025); 80.1% had died after a median OS of 9.9 months. The imputation-pooled derivation C-index was 0.654 (95% CI, 0.627–0.681); optimism-corrected, 0.629. Internal validation used 62 separate patients from the same registry; 96.8% had died after a median OS of 9.2 months. The validation C-index was 0.603 (95% CI, 0.518–0.687). Calibration was adequate at 6 and 12 months. Conclusions The PANTHEIA-SIRI model provides individualized OS estimates in mPDAC with routine clinical predictors. Its open-access calculator ( https://pantheia-siri.shinyapps.io/calc/ ) may support prognostic communication, treatment-intensity selection, and supportive-care planning. Routine clinical implementation will require further validation in larger, fully independent cohorts.