LAUSR

Background The risk of recurrence in localised, primary gastrointestinal stromal tumour (GIST) classified as high-risk after complete resection varies significantly. Thus, we aimed to develop a nomogram to predict the recurrence of high-risk GIST after surgery to aid patient selection. Methods We retrospectively evaluated patients (n = 424) with high-risk GIST who underwent curative resection as the initial treatment at two high-volume medical centres, between January 2005 and September 2019. The least absolute shrinkage and selection operator (LASSO) regression model was utilised to select potentially relevant features. Multivariate Cox proportional hazards analysis was used to develop a novel nomogram. Findings The nomogram comprised age, fibrinogen levels, prognostic nutritional index (PNI), platelet-lymphocyte ratio (PLR), mitotic counts and tumour size, which provided favourable calibration and discrimination in the training dataset with an AUC of 0•749 and a C-index of 0•742 (95%CI:0•689–0•804). Further, it showed acceptable discrimination in the validation cohort, with an AUC of 0•778 and C-index of 0•735 (95%CI:0•634–0•846). The time-dependant receiver operating characteristic (ROC) curves performed well throughout the observation period. Additionally, the nomogram could classify high-risk GISTs into ‘very high-risk’ and ‘general high-risk’ groups with a hazard ratio (HR) of 5•190 (95%CI: 3•202–8•414) and 5•438 (95%CI: 2•236–13•229) for the training and validation datasets, respectively. Interpretation The nomogram independently predicted post-operative recurrence-free survival (RFS) in high-risk GIST and showed favourable discrimination and calibration values. It may be a useful clinical tool for identifying ‘very high-risk’ GIST, by allowing treatment strategy optimisation in these patients. Funding National Natural Science Foundation of China (No. 81702386 and 81874184)