Development and validation of CT-based nomograms for predicting recurrence and prognosis in gastrointestinal stromal tumor

Authors:
Publisher website
Google Scholar
Abstract
Background This study analyzed preoperative CT features and postoperative pathological factors to predict recurrence after curative resection in primary gastrointestinal stromal tumor (GIST) and evaluate the prognostic validity of adjuvant therapy. Methods We retrospectively examined 718 patients with pathologically confirmed primary GISTs (mean age 57 ± 12.1 years; range, 14–86 years; 368 males, 350 females), who underwent R0 resection across two hospitals (internal set: n = 624; external set: n = 94). The internal set was randomly divided into training (n = 437) and validation cohorts (n = 187). Predictive efficacy of our nomogram was compared with the modified National Institutes of Health (MNIH) criterion using Kaplan–Meier analysis and multivariable Cox regression for recurrence-free survival (RFS). Results For recurrence prediction, AUC values for the CT model, combined nomogram (integrating CT features and pathological features), and MNIH criterion were 0.867/0.786/0.917 (training/validation/external test) and 0.890/0.810/0.953 versus 0.791/0.739/0.894, respectively. The combined nomogram demonstrated superior efficacy to the MNIH criterion across most probability thresholds (P < 0.001). The prognostic nomogram stratified patients into ‘high-recurrence’ and ‘low-recurrence’ groups with significantly divergent RFS. In the training cohort, the 1-, 3-, 5-, and 7-year RFS rates of these two groups were 95.9%, 85%, 73.5%, and 62.2% versus 100%, 99%, 95.5%, and 95.5%, respectively (P < 0.001). In the validation and external test cohorts, the nomograms were still significantly efficient (P < 0.001). Conclusions The combined nomogram provides improved noninvasive prediction of GIST recurrence and prognosis compared to the MNIH criterion. Trial registration Approved by the Institutional Review Board of West China Hospital, Sichuan University (No. 2020-249).

This publication has 28 references indexed in Scilit: