RESUMO
BACKGROUND: Standard risk assessment algorithms for gastrointestinal stromal tumor (GIST) are based on anatomic and histopathological variables with arbitrarily defined subcategories. Our goal was to improve risk assessment for GIST through retrospective analysis of patient data. METHODS: The National Cancer Database (NCDB) was queried for patients with GIST; the final cohort consisted of 19,030 cases. Main outcomes were metastasis at presentation and overall survival. A test dataset was used to reevaluate risk stratification parameters in multivariate regression models. A novel risk assessment system was applied to the validation dataset and compared to other currently used risk assessment schemes. RESULTS: Analysis of observed prevalence of metastases at presentation suggested 7 cm and mitotic rates > 10 per 5 mm2 as optimal threshold values. A proposed risk stratification score showed statistical superiority compared to the National Comprehensive Cancer Network, American Joint Committee on Cancer, and modified National Institute of Health classifications in predicting probability of presentation with metastasis at diagnosis and 4-year overall survival after accounting for important covariables including patient age and comorbidities, year of diagnosis, and surgical/systemic therapeutic regimen. CONCLUSIONS: Reexamination of prognostic factors for GIST demonstrated that current threshold values for tumor size and mitotic rate are suboptimal. A risk stratification score based on revised categorization of these factors outperformed currently used risk assessment algorithms.
