Risk stratification for reoperation outcomes in recurrent or persistent papillary thyroid carcinoma: development and validation of a machine learning model.

backgroundsReoperation is a key therapeutic strategy for recurrent or persistent papillary thyroid carcinoma (PTC), but its outcomes remain highly heterogeneous. Effective early risk stratification is essential to guide clinical decision-making and improve patient prognosis. This study aims to develop and validate interpretable machine learning (ML) models for the early prediction of reoperation response in patients with recurrent or persistent PTC.

methodsThis retrospective study included 670 patients with PTC who underwent reoperation at Wuhan Union Hospital between January 2012 and January 2024. Multiple ML algorithms, including random forest (RF), were used to construct predictive models. Model performance was evaluated by the area under the receiver operating characteristic curve (AUC), and SHapley Additive exPlanations method were applied to interpret feature contributions.

resultsAmong the twelve ML models developed, the RF model demonstrated the highest predictive accuracy. In the training cohort, the RF model achieved an AUC of 0.923 (95% CI, 0.899-0.947); in the validation cohort, the AUC was 0.865 (95% CI, 0.816-0.914). Thirteen variables were retained in the final model, including age, body mass index, Hashimoto's thyroiditis, total tumor size, primary positive lymph node, intraglandular dissemination, gross extrathyroidal extension, primary radioactive iodine treatment, levels of thyroglobulin, the systemic immune-inflammation index, reoperation interval, reoperative lymph node ratio, and recurrent lesion size.

conclusionsThis study developed and validated an interpretable ML model capable of accurately predicting reoperation outcomes in patients with recurrent or persistent PTC. The model may assist clinicians in identifying high-risk individuals and tailoring personalized treatment strategies. CLINICAL TRIAL NUMBER: Not applicable.