Constructing a machine learning model for predicting early postoperative recurrence of pancreatic head cancer based on a novel inflammatory factor composite index

Early postoperative recurrence of pancreatic head cancer (PHC) severely affects prognosis. We developed and validated a machine learning (ML)-based model incorporating a novel inflammatory composite index to predict early recurrence in PHC patients. We retrospectively analyzed 526 PHC patients who underwent pancreaticoduodenectomy at Fudan University Shanghai Cancer Center (2021–2022). Patients were randomly divided into training (70%) and test (30%) sets. Preoperative clinical, laboratory (ALI, PNI, SIRI), and pathological data were collected. Ten machine learning models were developed and evaluated using AUC, DCA, calibration, and precision-recall curves. The Random Forest model showed the best performance and was interpreted with SHAP. Of the 526 patients, 164 (31.2%) developed recurrence or metastasis within one year. Multivariate logistic regression identified ALI, CA199, tumor differentiation, capsule integrity, and nerve invasion as independent risk factors. The RF model demonstrated excellent performance, with an AUC of 0.992 in the training set and 0.783 in the test set. SHAP analysis highlighted CA199, ALI, tumor differentiation, capsule status, and nerve invasion as key predictors. We developed and validated an RF-based predictive model incorporating a novel inflammatory index for assessing early recurrence risk in PHC patients, which may aid individualized postoperative management.