Development and validation of a machine-learning-based pathomics nomogram for predicting recurrence of localized...

Accurate prediction of the recurrence risk of localized primary gastrointestinal stromal tumors (GISTs) after complete surgical resection is crucial for determining adjuvant treatment and surveillance strategies. Traditional risk-stratification schemes often exhibit heterogeneity and may not provide sufficient prognostic information. Therefore, the authors' objective was to develop a pathomics nomogram that integrates digital pathology and machine-learning algorithms to improve predictive accuracy. The authors enrolled 421 eligible participants (253 in the training cohort, and 168 in the external validation cohort) from four medical centers. Four distinct machine-learning methods were evaluated, and the one that demonstrated optimal performance in the validation cohort was selected to develop the pathomics model. Subsequently, stepwise multivariate Cox regression analysis was performed to construct a machine-learning-based pathomics nomogram (the MLPNom). The MLPNom exhibited superior predictive performance compared with traditional risk criteria (concordance index values: training cohort, 0.892; validation cohort, 0.964). The time-dependent area under the curve values for the MLPNom were notably higher than those for traditional risk criteria (5-year area under the curve values: training cohort, 0.919; validation cohort, 0.959). Calibration curves and Brier scores confirmed the excellent calibration of the MLPNom. Decision curve analysis further underscored the utility of the MLPNom in clinical decision making for GISTs. Furthermore, the MLPNom identified three distinct prognostic subgroups that retained their significance after stratification based on diverse clinicopathologic factors. The MLPNom demonstrates robust discrimination and calibration in predicting recurrence-free survival in localized primary gastric and small intestinal GISTs after complete surgical resection. This may complement traditional risk criteria and aid in selecting patients for adjuvant imatinib therapy.