Malignancy Rate in Thyroid Nodules Classified as Bethesda Category III (AUS/FLUS): Is There a Correct Answer?

Thyroid fine-needle aspiration (FNA) is an effective tool in the evaluation of thyroid nodules and helps guide the clinician in determining the need for surgery and the extent of the initial surgical procedure. The Bethesda System for Reporting Thyroid Cytopathology was created to permit effective communication among clinicians, to facilitate research, and to allow easy and reliable sharing of data. It consists of six categories. Category III, atypia of undetermined significance/follicular lesion of undetermined significance (AUS/FLUS), includes those thyroid cytologies that are not easily classifiable. Its inclusion as a separate category was a subject of considerable debate. Nine different cytologic patterns are described that can be included in this category (1). This heterogeneity and ambiguity reflects that the distinction between category III and the other categories (either benign or more suspicious for tumor or malignancy) is not well defined and a category III cytology in one institution may be a category II, IV, or V nodule in another. Many patients with category III cytology do not undergo surgery and are observed with repeat FNA and imaging. The authors of the Bethesda Classification state: “[nodules] that are resected represent a selected population of patients with repeated AUS results or patients with worrisome clinical or sonographic findings. In this selected population, 20–25% of the patients with AUS prove to have cancer after surgery, but this is undoubtedly an overestimate of the risk for all AUS interpretations. The risk of malignancy is certainly lower and probably closer to 5–15%” (1).

In this issue of Thyroid, Allen Ho and colleagues at the Memorial Sloan-Kettering Cancer Center (MSKCC) report that in that institution, the malignancy rate in patients who underwent surgery for category III nodules was 37.8% (2). The data are carefully and clearly presented. Potential confounding factors such as treatment, referral, and verification bias are thoughtfully addressed, but evidence suggesting that these biases account for the unusually high percentage of malignancy could not be identified. The authors do state, however, that if all the nodules that were observed were benign, the risk of malignancy in the entire cohort would be 26.6%, still substantially greater than expected. This series is unusual in that 65% of patients presenting with category III nodules underwent initial surgery, a far greater percentage than would normally be anticipated. This is attributed to institutional practice rather than a high percentage of nodules that were considered suspicious for malignancy for reasons unrelated to cytology.

The authors nevertheless caution that their conclusions, while accurate for their own institution, may not be applicable to other institutions. This is certainly true and the unique nature of MSKCC as a tertiary cancer referral center makes a referral bias almost inevitable. In fact, the data presented support this hypothesis. In patients referred with an outside cytology finding (subsequently reviewed at MSKCC) the incidence of malignancy (including patients who did not undergo surgery) was 29.6%. In patients in whom the initial FNA was performed at MSKCC, the incidence of malignancy was only 21.8%. Other frequently referenced series report lower incidences of malignancy in category III nodules undergoing surgery, for example 14% at the University of Pittsburgh (UPMC) (3), and 24% in the Gene Expression Classifier Validation Study (4). Whether this reflects the biases mentioned above, or simply differences in opinion among experienced cytopathologists, cannot be determined.

What can be learned from this study? Should we recommend surgery for all category III nodules? I think not. What is crucial is that clinicians need to understand that cytopathology, in spite of all the attempts to standardize nomenclature, remains subjective, and that patient populations differ among institutions. It is crucial to know the incidence of malignancy in each of the Bethesda categories in one's own institution in order to make reasonable treatment decisions. MSKCC is no more or less correct than UPMC, nor do their data prove that the Bethesda Classification needs to be revised or that surgical strategies should be changed.

Molecular testing is being used with increasing frequency to predict the risk of malignancy in thyroid nodules and to determine whether or not surgery is necessary. While this seems much less subjective than cytopathologic evaluation, it requires an important caveat. The accuracy of these tests, particularly the Gene Expression Classifier, is highly dependent on the prevalence of malignancy in the population studied. The negative predictive value of this test, and hence its usefulness in selecting patients who can be observed without surgery, decreases as the prevalence of malignancy increases (5). In a population of patients such as that reported from MSKCC, with a cancer prevalence of 38%, the risk of malignancy may be unacceptably high even with a benign Gene Expression Classifier. We need to look to the future, not for better cytologic classification schemes, but for more accurate molecular and genetic studies, and a better understanding of which nodules, even if malignant, require treatment.