A SIMPLE PREDICTIVE SCORING MODEL FOR MALIGNANT THYROID NODULES: A RETROSPECTIVE STUDY FROM 10 447 SURGICAL CASES

Objective: We aimed to identify clinical, biochemical and ultrasonographic predictors for malignant thyroid nodules (TNs) in patients with confirmed pathological diagnosis to develop a predictive scoring model. Methods: We reviewed the medical records of patients who had undergone surgery for TNs (n = 10447) at Quit Hospital of Shandong University between January 2005 and April 2016. Multivariate logistic regression was performed to determine independent risk factors, and a predictive scoring model was created based on the regression equation. The score cut-off values were calibrated using receiver operating characteristic (ROC) curve analysis. Results: Independent predictors of malignancy were extremes of age (< 40 years or >= 80 years), duration of disease < 2 yrs, family history of thyroid cancer, history of breast diseases, TSH >= 1.25 mu IU/ml, nodule size < 2 cm, macmcalcifications, micmcalcificalions, solid composition, hypoechogenicity, irregular shapes, poorly defined margins, increased intranodular vascularity, taller-thanwide shapes, and abnormal lymph nodes. A predictive scoring model demonstrated an AUC of 0.88 (95% CI 0.53-0.62) at a cut-off point of 10 scores. Conclusions: The study verified predictors of malignant TNs. A predictive scoring model was found to be a useful tool in diagnostic work-up of TNs.