Post-operative prognostication of patients diagnosed with Hurthle cell carcinoma: a machine learning approach.

OBJECTIVES

To evaluate the performance of machine learning models in predicting the 5-year overall survival of patients with Hurthle cell carcinoma, and to identify significant prognostic factors influencing survival.

METHODS

A retrospective cohort study was conducted using data from the Surveillance, Epidemiology, and End Results database, encompassing patients treated between 2010 and 2015. Key variables included demographic information (age, sex, race), clinical characteristics (tumor size, T, N, M stages, and overall stage), and survival outcomes. Patients were included if they had complete data, were not censored before 60 months of follow-up, and had undergone thyroid surgery.

RESULTS

The study included 1,143 patients with a mean age of 57.7 years (standard deviation = 15.8). The cohort consisted of 770 females (67.4%) and was predominantly White (83.0%). Tumor classifications were varied, with T2 being most common (37.2%). The majority had no nodal involvement (94.1%) or distant metastasis (97.6%). The support vector model achieved the highest area under receiver characteristics operating curve of 0.8402 (95% CI: 0.7915 to 0.8847), indicating good predictive performance. Sensitivity and specificity were 81.16% and 73.72%, respectively. The Brier score for the model was 0.1223, demonstrating adequate calibration. Higher age and T classification were the most significant predictors of decreased survival, while being female was associated with increased survival.

CONCLUSION

Machine learning models, particularly the support vector model, effectively predicted 5-year overall survival in patients with Hurthle cell carcinoma. The study highlights age and tumor extent as critical prognostic factors.