Study on the Combined Prediction Model of c-KIT and PDGFRa Gene Mutation Status and Clinicopathological Features in 241 Patients with Gastrointestinal Stromal Tumors

Objective Through retrospective analysis of clinical and pathological data and c-KIT / PDGFRa gene mutation status of gastrointestinal stromal tumor (GIST) patients in Gansu Province, the correlation characteristics between the two were clarified, and a joint predictive model based on clinicopathological features and gene mutation status was constructed. This study aims to provide a practical reference for the auxiliary diagnosis and treatment of GIST in primary hospitals, and supplement the molecular epidemiological data of GIST in the northwestern region of China. Methods Clinical data of GIST patients confirmed by pathology at Gansu Provincial Hospital were retrospectively collected, and the mutation status of c-KIT exons 9,11,13,17 and PDGFRa exons 12,18 was detected by Sanger sequencing. Univariate and multivariate analyses were performed to compare the clinicopathological characteristics between the mutant and wild-type groups, as well as among various mutation subtypes. Subsequently, statistically significant variables (age, Ki-67 proliferation index, SDHB expression, primary tumor site, maximum tumor diameter, mitotic figures) were screened using logistic regression and incorporated into the XGBoost model for mutation status prediction. The XGBoost (Extreme Gradient Boosting) gradient boosting tree model was used for classification modeling to predict the mutation types of c-KIT 11, c-KIT 9, c-KIT 17, c-KIT 13, and PDGFRa , with model performance evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC) metrics. Results Among 241 patients, the gene mutation detection rate was 90.5% (218 cases), including c-KIT mutations in 85.9% (207 cases) and PDGFRa mutations in 4.98% (12 cases, 11 cases of exon 18 mutation and 1 case of exon 12 mutation). 0.4% of the cases were excluded due to uncertain detection results, and 9.0% (23 cases) were wild-type. No statistically significant differences were observed in clinicopathological characteristics among major mutation subtypes (all P  > 0.05). Prediction model performance analysis revealed: c-KIT 9 mutation prediction AUC = 0.832 (95% CI: 0.750–0.913, P  < 0.001); c-KIT 11 mutation prediction AUC = 0.727 (95% CI: 0.641–0.814, P  < 0.001); c-KIT 17 mutation prediction AUC = 0.897 (95% CI: 0.836–0.958, P  < 0.001); and PDGFRa mutation prediction AUC = 0.810 (95% CI: 0.733–0.886, P  < 0.001). Conclusion The combined model based on clinicopathological features demonstrated optimal predictive efficacy for c-KIT 9 mutations and moderate predictive value for c-KIT 11 mutations. This model provides a reference for rapid screening of gene mutation types in primary care hospitals in northwest China, and supports individualized targeted therapy decisions for GIST, in line with the development trend of precision medicine. Clinical trial registration: Not applicable.