Title: A Prediction Model for Thyroid Cancer with Integration of Patient Demographics, History, Ultrasound Characteristics and Bethesda Classification
Tharakan T, Ow T, Shankar V, Schiff B, Smith R, Mehta V
Objectives: Create a predictive model for malignancy of a thyroid nodule which integrates demographic information, clinical history, ultrasound features, and Bethesda cytology classification.
Methods: We reviewed the records of 876 patients who underwent thyroidectomy at Montefiore Medical Center in the Bronx, NY between 2012 and 2017. A model to predict malignancy on final surgical pathology was fitted using multiple logistic regression. Predictors considered were patient age, gender, race/ethnicity, history of thyroid cancer or disease, history of nonthyroid cancer, radiation history, family history of thyroid cancer, compressive symptoms, hoarseness, thyroid nodule appearance on ultrasound (maximum diameter, taller than wide shape, composition, echogenicity, margin irregularity, presence of calcifications), and Bethesda class.
Results: The predictive model had a sensitivity of 72%, specificity 82%, and accuracy 78%. The area under the ROC curve (AUC) for this model was 0.86 (95% CI 0.82–0.90) versus 0.77 (95% CI 0.72 – 0.82) for a model using Bethesda class alone. This suggests that incorporation of the clinical and imaging features to Bethesda class improves the predictive power of the model from “fair” to “good”.
Conclusion: A prediction model integrating clinical and imaging features with Bethesda classification can help clinicians make decisions about the management of thyroid nodules with greater precision than Bethesda class alone.