Predicting thyroid nodule malignancy at several prevalence values with a combined Bethesda-molecular test

Abstract

Investigation of thyroid nodules using fine-needle aspiration cytology (FNAC) gives indeterminate results in up to 30% of samples using the Bethesda System for Reporting Thyroid Cytopathology (TBSRTC). We present a combined Bethesda-molecular predictor of nodule malignancy to improve the accuracy of the preoperative diagnosis of thyroid nodules. To detect a molecular signature of thyroid nodule malignancy, a molecular test was performed on FNACs from 128 thyroid nodules from prospectively included patients, collected in a tertiary center. The test relied on a transcriptomic array of 20 genes selected from a previous study. An optimal set of seven genes was identified using a logistic regression model. Comparison between the combined predictor (TBSRTC + molecular) and TBSRTC alone used the area under the ROC curve (AUC). Performance of the combined predictor was calculated according to various malignancy prevalence values and benefit-to-harm ratios (B/Hr) (favoring sensitivity or specificity). In our population (36% malignancy prevalence) and with a B/Hr of 1, the combined predictor achieved 95% specificity and 76% sensitivity. The AUC was 93.5%; higher than that of TBSRTC (P = 0.004). Among indeterminate nodules (30% malignancy prevalence), sensitivity and specificity were 52.2% and 96.2%, respectively, with a B/Hr of 1, or 95.7% and 64.2% with a B/Hr of 4 (favoring sensitivity), allowing avoidance of 64% of unnecessary surgeries at the cost of only one false-positive result. In conclusion, this predictor could improve the detection of thyroid nodule malignancy, taking into account malignancy prevalence and B/Hr, and reduce the number of unnecessary thyroidectomies.