Risk of Malignancy in Thyroid Atypia of Undetermined Significance

Thyroid nodules labeled “atypia of undetermined significance” (AUS), also called follicular lesion of undetermined significance (FLUS), sit in a diagnostic gray zone where the cancer risk is real but far from certain. Under the Bethesda System for Reporting Thyroid Cytopathology, this category (Bethesda III) has historically carried an estimated malignancy risk of roughly 5% to 15%, but newer data and the 2023 update to the Bethesda System have revised those figures upward for many patients. The actual risk for any individual nodule depends heavily on what kind of atypia the pathologist sees, what ultrasound features the nodule shows, and whether molecular testing is performed.

What the Bethesda System Says and Why the Numbers Shifted

The Bethesda System groups thyroid fine-needle aspiration (FNA) results into six categories, each tied to an expected range of cancer risk. AUS/FLUS is Category III, a catch-all for samples that look abnormal enough to worry about but not abnormal enough to call suspicious or malignant. For years, the published risk for this category hovered around 5% to 15%. The third edition of the Bethesda System, published in 2023, updated and refined the risk of malignancy (ROM) for each category based on data accumulated since the previous edition, offering both an average ROM and an expected range for each tier.¹ Many institutions now report actual malignancy rates well above the old 15% ceiling, particularly when surgical follow-up is available for most of the nodules sampled.

Part of the reason for that upward shift is selection bias: nodules that go on to surgery tend to be the ones with more worrisome features, so the malignancy rate among operated cases is naturally higher than the rate you would get if every AUS nodule were removed. But part of it is also real. The AUS category is broad, and lumping all its subtypes together masks important differences.

Nuclear Atypia Versus Architectural Atypia

Not all AUS diagnoses carry the same weight. Pathologists increasingly sub-classify Bethesda III cases based on the type of abnormality they see under the microscope. The two main flavors are nuclear atypia, where individual cell nuclei look abnormal in ways that suggest papillary thyroid carcinoma, and architectural atypia, where the cells are arranged in unusual patterns like microfollicles but the nuclei themselves look relatively normal. Some cases show both, and a smaller subset involves Hürthle cells.

This distinction matters because nuclear atypia consistently carries a higher cancer risk. One study found a malignancy rate of about 37% for nodules with nuclear atypia compared to roughly 15% for those with only architectural atypia.² Another large analysis found a similar gap: about 17% ROM for nuclear atypia cases versus roughly 9% for architectural atypia cases when all patients (not just those who had surgery) were included.³ Cases can be reclassified into finer subcategories, such as architectural atypia, cytologic (nuclear) atypia, a combination of both, or Hürthle cell type.⁴

Not every study finds the same neat hierarchy, though. A retrospective analysis of nearly 200 AUS cases found that Hürthle cell AUS actually had the highest malignancy rate at about 56%, followed closely by cytologic atypia at 55%, with architectural atypia at roughly 51%, and the differences were not statistically significant in that cohort.⁵ This kind of variability between institutions underscores a recurring theme: your pathologist’s practice patterns and case mix matter. The subcategory is a useful signal, but it is not a precise number you can bank on.

How NIFTP Changed the Math

In 2016, the thyroid pathology community reclassified a subset of tumors previously called encapsulated follicular variant of papillary thyroid carcinoma. These were re-labeled as noninvasive follicular thyroid neoplasm with papillary-like nuclear features, or NIFTP, and because they behave in an indolent manner, they were removed from the “malignant” column. This single reclassification meaningfully lowered the reported ROM across several Bethesda categories.

For AUS/FLUS specifically, one study found that excluding NIFTP from malignancies produced a relative decrease of about 15% and an absolute decrease of roughly 5.5 percentage points in the ROM.⁶ Another institution reported an absolute drop of about 7 percentage points after NIFTP cases were reclassified, but even after that adjustment the ROM at their center remained above 35%.⁷ The practical takeaway is that when you read older literature quoting ROM numbers for AUS, check whether those numbers counted NIFTP as cancer. If they did, the “true” malignancy risk by today’s definitions is somewhat lower than reported.

What Ultrasound Adds to the Picture

A Bethesda III result does not exist in a vacuum. The nodule’s appearance on ultrasound provides independent information about cancer risk, and combining the two produces better predictions than either alone. Standardized ultrasound risk stratification systems like ACR TI-RADS and K-TIRADS assign nodules to suspicion categories based on features like composition, echogenicity, shape, margins, and calcifications.

A study examining multiple ultrasound-based risk stratification systems found that an intermediate or high suspicion ultrasound category independently increased the malignancy risk in AUS nodules. Larger nodules (3 cm or above) also independently raised the risk when the ultrasound category was low or intermediate. Combining ultrasound, size, and cytology subcategory into a unified scoring system improved diagnostic accuracy beyond what ultrasound alone could achieve, and nodules in the very lowest risk tier of the combined system had zero malignancies on follow-up.⁸

Other researchers have integrated ultrasound scores with cytologic subclassification and BRAF mutation testing, achieving high specificity in the range of 81% to 87%, though sensitivity remained moderate.⁹ The bottom line for patients is that a reassuring ultrasound makes a Bethesda III result less alarming, while a suspicious-looking nodule on imaging makes it more so.

Molecular Testing and Its Limitations

Molecular testing has become a central tool for managing indeterminate thyroid nodules. The main commercial platforms, ThyroSeq, Afirma (including its gene sequencing classifier), and ThyGeNEXT/ThyraMIR, analyze genetic or gene-expression signatures from the FNA sample to help predict whether a nodule is likely benign or malignant. Their primary clinical value is as “rule-out” tests: a negative (benign) molecular result lowers the residual cancer risk enough that many patients can avoid surgery.

A recent study comparing all three platforms head-to-head in over 1,250 indeterminate nodules found broadly similar performance, with ThyroSeq showing the highest sensitivity at about 88%, followed by Afirma at roughly 73% and ThyGeNEXT/ThyraMIR at about 67%. Negative predictive values ranged from 66% to 80%.¹⁰ A meta-analysis pooling data from dozens of studies found that the newer iterations of these platforms, particularly ThyroSeq v3, performed best overall, with the gene sequencing classifier version of Afirma showing the strongest ability to rule out malignancy.¹¹

The performance of molecular testing is not uniform across AUS subtypes, though. One study found that the Afirma genomic sequencing classifier achieved 100% negative predictive value across all AUS subcategories, but positive predictive value varied widely, from 5% to 56%, depending on the pattern of atypia.¹² A “suspicious” molecular result does not mean cancer is certain; it means the risk is high enough that surgery is recommended. A “benign” result does not mean cancer is impossible; it means the residual risk is low enough that surveillance is usually appropriate.

These tests also come with practical constraints. They require adequate FNA material, add cost, and are not universally available. A cost-effectiveness analysis found that molecular testing was considerably more cost-effective than proceeding directly to diagnostic lobectomy for indeterminate nodules.¹³ A separate analysis from an Asian population found that routine molecular testing could become cost-saving if test prices dropped below a certain threshold, and that without molecular testing, rates of unnecessary surgery were high.¹⁴

What Specific Mutations Tell You

Beyond panel-based molecular tests, the detection of individual oncogenic mutations in AUS nodules can sharpen risk estimates. BRAF V600E is the most clinically significant. This mutation is strongly associated with papillary thyroid carcinoma and, when found, essentially confirms malignancy. A study of indeterminate and suspicious thyroid aspirates detected BRAF V600E in about 10% of all cases, and the mutation correlated with lymph node spread and higher disease stage.¹⁵ TERT promoter mutations, though rare in AUS cases specifically, flag aggressive behavior when present. In that same study, all TERT-mutated cases turned out to be an aggressive variant of papillary carcinoma.

The clinical implication is straightforward: if your AUS nodule tests positive for BRAF V600E, the conversation shifts firmly toward surgery. If targeted mutation testing comes back negative, the picture is less clear, because many thyroid cancers do not carry BRAF mutations and would be missed by this single-gene approach. That is why broader panel tests exist.

Repeat Biopsy and What a Second AUS Result Means

One standard approach after an initial AUS result is to repeat the FNA. The hope is that a second sample will yield a more definitive diagnosis. In many cases, this works. A recent study found that repeat FNA resulted in a benign diagnosis in about 49% of cases and another AUS result in roughly 36%.¹⁶

When a second FNA again returns AUS, the situation becomes more concerning. One study of patients whose initial AUS was followed by another AUS on repeat biopsy found a 64% malignancy rate on final surgical pathology, far higher than the risk after a single AUS result.¹⁷ That study’s authors argued that surgery is a reasonable step after two consecutive AUS results. A multicenter study tracked patients through up to three FNAs and found rising malignancy rates at each round among those who eventually had surgery: about 39% after the first FNA, 48% after a second, and 69% after a third.¹⁸ These numbers likely reflect selection bias (the most worrisome nodules keep getting biopsied), but the trend is consistent: persistence of an indeterminate result is itself a risk factor.

Core Needle Biopsy as an Alternative

When repeat FNA keeps returning indeterminate results, core needle biopsy (CNB) offers another option. CNB extracts a small cylinder of tissue rather than loose cells, giving the pathologist a better look at the nodule’s architecture. A study comparing the two approaches found that CNB produced fewer inconclusive diagnoses than repeat FNA in AUS cases, with AUS readings of about 24% for CNB versus roughly 40% for repeat FNA. Sensitivity for detecting malignancy was also higher with CNB.¹⁹ Another study found that CNB was especially advantageous for larger nodules (over 2 cm) with intermediate ultrasound suspicion, though for other nodule types the two techniques performed comparably.²⁰

CNB is more widely used in Asian countries than in the West, where FNA remains the dominant approach. It is a slightly more invasive procedure and requires operator skill, but for patients stuck in repeat-FNA limbo, it can break the diagnostic stalemate.

When Hashimoto’s Thyroiditis Muddies the Waters

Hashimoto’s thyroiditis, the most common autoimmune thyroid condition, creates a particular headache for thyroid FNA interpretation. The chronic inflammation and lymphocytic infiltration it produces can mimic the cellular changes that pathologists look for when calling a sample atypical. A study found that patients with Hashimoto’s had a significantly higher percentage of indeterminate FNA results, including AUS, and that overall diagnostic performance of FNA was lower in this population.²¹ If you have Hashimoto’s and receive an AUS result, it is worth knowing that the indeterminate call may partly reflect the autoimmune background rather than a truly suspicious nodule. This does not eliminate the need for follow-up, but it is relevant context for shared decision-making with your doctor.

AUS in Children and Adolescents

Thyroid nodules are less common in children than in adults, but when they do occur, they carry a higher overall rate of malignancy. This pattern extends to AUS. A single-institution pediatric study found that AUS nodules with nuclear atypia had a malignancy rate of 52%, compared to 15% for other AUS subtypes, a statistically significant difference.²² A broader pediatric study found malignancy in 50% of AUS cases that went to surgery, with nuclear atypia being the most frequently noted subtype and half of those cases turning out to be papillary thyroid carcinoma.²³ The higher baseline malignancy rates in children mean that the management threshold may be lower; many pediatric endocrinologists and surgeons lean more quickly toward surgery or at least molecular testing rather than prolonged observation.

Active Surveillance for Small Indeterminate Nodules

Not everyone with a Bethesda III nodule needs immediate intervention. For small, low-suspicion nodules, active surveillance, meaning serial ultrasounds to watch for growth or changing features, is increasingly discussed as a viable option. A prospective study followed patients with small indeterminate thyroid nodules under active surveillance and found that only about 14% showed tumor enlargement over a median follow-up of roughly 42 months, with no cases of lymph node or distant metastases during observation. Four patients eventually had surgery, and only one of those turned out to have cancer, a follicular variant of papillary carcinoma that remained disease-free after treatment.²⁴

This is a small study, and active surveillance for indeterminate nodules is far less established than it is for confirmed low-risk microcarcinomas. But it hints that for selected patients, particularly those with small nodules, favorable ultrasound features, and a benign molecular test result, watchful waiting can be safe. The conversation between patient and physician about comfort with uncertainty becomes central here.

Emerging Tools on the Horizon

Researchers are exploring several newer approaches to improve diagnosis of indeterminate thyroid nodules. MicroRNA-based assays analyze small non-coding RNA molecules in FNA samples. A multicentre validation study of one such assay reported a negative predictive value of 91% and sensitivity of 85%, with performance improving further when only cases with high pathologist agreement on final diagnosis were considered.²⁵ An earlier proof-of-concept study applied a microRNA panel specifically to AUS cases, achieving about 73% diagnostic accuracy, though the panel struggled with follicular neoplasias.²⁶

Artificial intelligence is also entering the field. Deep learning models are being developed to assist with Bethesda category classification and to integrate cytology images with molecular and clinical data.²⁷ One recent model combining clinical data, ultrasound findings, and cytopathological features through machine learning achieved 82% accuracy and 97% specificity for predicting malignancy in Bethesda III nodules, performing best at identifying benign cases that could safely skip surgery.²⁸ These tools are not ready to replace existing workflows, but they represent the direction the field is heading: layering multiple data streams to narrow the gray zone.

Why Practice Varies So Much Around the World

If you compare how AUS nodules are managed in Seoul, Houston, and São Paulo, you will find strikingly different approaches. Some of this is cultural, but much of it reflects genuine differences in disease patterns, healthcare economics, and available technology. Asian countries, for instance, have notably different thyroid nodule practices compared to Western countries, which led to the formation of the Asian Thyroid Working Group to study and document these disparities.²⁹ Core needle biopsy is routine in South Korea but uncommon in the United States. Molecular testing is widely available in North America and parts of Europe but cost-prohibitive or logistically impractical in many other regions. Where molecular testing is unavailable, more patients with AUS go directly to surgery, and in settings where thyroid cancer screening is aggressive, the AUS category itself gets used more liberally.

These differences mean that a “Bethesda III” from one institution is not perfectly interchangeable with the same label from another. Institutional malignancy rates for AUS vary enormously, from under 10% to over 50%, depending on how liberally the category is applied, how often NIFTP is encountered, what population is being served, and whether surgical follow-up is available. This is not a failure of the system so much as an inherent limitation of a category designed to capture diagnostic uncertainty. Asking your doctor what the AUS malignancy rate is at your specific institution, if they track it, gives you far more useful information than any number from a journal article.