Probabilities in the cytologic diagnosis of thyroid gland lesions.

Abstract

Thyroid gland fine-needle aspiration (FNA) is widely accepted as the pivotal diagnostic procedure to separate patients who have a thyroid gland nodule into those who may need to have the nodule removed and into those who probably do not. Thyroid gland FNA also serves as a diagnostic test to subcategorize lesions that necessitate specific followup protocols. The conception of thyroid gland FNA as a screening test is shown in ❚Table 1❚, which roughly separates thyroid gland diagnoses into several categories. The majority of neoplastic diagnoses (with the exception of malignant lymphoma, metastasis, and, sometimes, anaplastic carcinoma) usually result in an operative procedure. This conception is probabilistic, although most clinicians usually do not use a formal Bayesian method (eg, likelihood ratios) to incorporate the probabilities associated with FNA diagnoses into their clinical decision making. However, clinicians use “implicit” probabilities associated with the FNA diagnosis along with the clinical findings and history (and preferences) to determine the management plan. This generally means that the probability a nodule is neoplastic is decreased after an FNA diagnosis of benign (eg, benign thyroid nodule) and increased following an FNA diagnosis of neoplastic. Clinicians learn these outcomes through experience and then behave in a manner in which some thyroid nodules are “left in” and others “taken out.” Qualified diagnoses, such as atypical or “suspicious,” generally increase the post-FNA probability of neoplasia.1,2 Of course, the clinical findings and history also affect the clinical management (eg, if the FNA diagnosis is benign thyroid nodule and the clinical findings are more worrisome, the nodule may be removed). Certain FNA diagnoses also affect the type of operative procedure performed (eg, an FNA diagnosis of papillary carcinoma may result in a total thyroidectomy, whereas a diagnosis of follicular neoplasm may result in a lobectomy). Problems arise with FNA diagnoses when the postFNA probabilities of neoplasia (or specific neoplasias) are not as high or as low as we would like them to be. Three problematic FNA diagnoses are: (1) suggestive of follicular neoplasm, (2) follicular neoplasm (or its corollary, Hurthle cell neoplasm), and (3) suggestive of papillary carcinoma. Although histologic follow-up shows that each one of these lesions is more likely to be neoplastic (and, therefore, increase the post-FNA probability of neoplasia), the specific probability of neoplasia or benignity may be insufficient compared with what practitioners desire. In other words, it would be advantageous to find some way to be more certain of the outcome or, practically speaking, place fewer benign lesions in each of these categories. Methods to achieve this goal include determining better cytologic descriptors for classification, ancillary studies (eg, DNA flow cytometry), and additional techniques (eg, core biopsy or large-needle aspiration biopsy [LNB]). In the current issue of the Journal, Carpi et al3 examine the usefulness of LNB for the triage of patients with problematic FNA diagnoses. Their study examines several of the aforementioned diagnostic dilemmas, and before discussing their study, I would like to discuss these problem areas. In an elegant fashion, DeMay4 portrays the problems 1 and 2 in The Art & Science of Cytopathology. Much of thyroid gland FNA diagnosis is concerned with the correct classification of the follicular lesion, which, broadly defined, includes nodular hyperplasia (benign thyroid nodule) and follicular neoplasms (follicular adenomas and carcinomas). DeMay4 shows that the majority of follicular lesions may be classified correctly by using 2 inversely related cytologic features: the amount of colloid and the cellularity. In nodular hyperplasia, the amount of colloid is high, and the cellularity is low. In the follicular neoplasm, the amount of colloid is low, and the cellularity is high. Secondary features (eg, microfollicles) may further aid in the diagnosis. Problem 1 arises when a lesion does not fit neatly into this binary separation and has some features suggestive but not diagnostic of a follicular neoplasm. These lesions are those in which the cellularity and colloid are high (additional passes probably