This retrospective study was approved by the institutional review board of our medical center, and informed consent was waived. The study was HIPAA compliant.

Demographic analysis of the 62 patients revealed 56 female patients (median age, 16.5 years; interquartile range [IQR], 15–18 years) and six male patients (median age, 12.5 years; IQR, 9–17 years). Of the total 74 nodules, 54 (73.0%) were benign and 20 (27.0%) were malignant. The median nodule size was 1.90 cm (IQR, 1.30–2.80 cm).

Of the 54 benign nodules, tissue diagnosis was confirmed by FNAB cytologic analysis in 39 (38 Bethesda class II and one Bethesda class III) and by surgical histopathologic analysis in 15 (two follicular adenomas, two benign thyroid nodules of colloid type, two lymphocytic thyroiditis, and nine multinodular goiters). Tissue diagnosis of all 20 malignant nodules was confirmed on the basis of surgical histopathologic analysis (14 papillary thyroid carcinoma, three follicular variant-papillary carcinoma, one papillary microcarcinoma, one follicular carcinoma, and one follicular neoplasm with Hürthle cell features).

For each radiologist, intraobserver agreement was substantial for TI-RADS category (κ = 0.69–0.77; p < 0.001). Intraobserver agreement was substantial to almost perfect for all individual parameters (κ = 0.66–0.94; p < 0.001), except for echogenicity, which was moderate to substantial (κ = 0.46–0.70; p < 0.001). In contrast, interobserver agreement, although substantial for composition and shape (κ = 0.68–0.79; p < 0.001), was only moderate for TI-RADS category (κ = 0.37; p = 0.002) (Table 1). However, a cross-tabulation calculation performed for TI-RADS category between the two observers showed that 19 of 41 (46.0%) disagreements were between adjacent categories (e.g., ACR TI-RADS category 3 vs 4, or category 4 vs 5).

The median ACR TI-RADS category was 4 (IQR, 4–5). ACR TI-RADS category categorized by pathologic status is shown in Table 2 with distribution as follows: four nodules (5.4%) with TI-RADS category 1 (three benign and one malignant); four (5.4%) with TI-RADS category 2 (four benign and zero malignant), six (8.1%) with TI-RADS category 3 (six benign and zero malignant), 24 (32.4%) with TI-RADS category 4 (22 benign and two malignant), and 36 (48.7%) with TI-RADS category 5 (19 benign and 17 malignant). All malignant nodules (19/20; 95.0%) were rated as TI-RADS category 4 or 5 except for one false-negative of a malignant 0.8-cm left thyroid nodule in a 17-year-old girl that was erroneously classified as TIRADS category 1 (benign) (Fig. 5B); this patient also had a concomitant 2.5-cm right malignant papillary carcinoma nodule that was assigned a TI-RADS category 5 rating (Fig. 5A).

On univariable analysis, for every 1-unit increase on the ordinal TI-RADS scale, the likelihood of malignancy increased 2.63 times (95% CI, 1.08–6.41; p = 0.03). For every 1-cm increase in size, nodules were also 69% more likely to be malignant (odds ratio, 1.69; 95% CI, 1.04–2.73; p = 0.03). On multivariable analysis, after adjusting for nodule size, ACR TI-RADS category, although not statistically significant, remained marginally associated with malignancy (adjusted odds ratio, 2.27; 95% CI, 0.93–5.54; p = 0.07), but size was no longer associated. Using the ROC curve generated from ACR TI-RADS category, the AUC estimate was 0.75 (95% CI, 0.64–0.86), which confirmed the accuracy of the predictive modeling. An optimal cut point of ACR TI-RADS category 5 was selected because it maximized both sensitivity (85%) and specificity (65%). The corresponding false-positive rate of 53% at this cut point did result in suppression of the positive predictive value, which was 47%. The negative predictive value was 92% (Table 3). Overall, nodules with an ACR TI-RADS category of 5 were 10.44 times (95% CI, 2.71–40.21; p < 0.001) more likely to be malignant.

The distribution of individual ultrasound features is also shown in Table 2. Some features were well differentiated by pathologic status. For the echogenicity feature, most (90.0%) malignant nodules were hypoechoic or very hypoechoic, whereas most (61.1%) benign nodules were hyper- or isoechoic. For the margin feature, most (83.3%) benign nodules were smooth or ill-defined, whereas most (75.0%) malignant nodules were lobulated or irregularly marginated. Other features, however, were not well differentiated by pathologic analysis. These included punc-tate echogenic foci, either alone or in combination with macrocalcifications, which were found in the majority of both malignant and benign nodules (85.0% and 64.8%, respectively). Finally, taller-than-wide shape, a suspicious feature, was found in a small percentage of both malignant (5.0%) and benign (9.3%) nodules, including in a 14-year-old boy with six thyroid nodules, all showing benign multinodular goiter on surgical pathologic analysis (Fig. 6).

The results of the current study reveal that the ACR TI-RADS discriminates well between malignant and benign thyroid nodules in a pediatric population and is particularly effective in identifying positive cases. The TI-RADS category assigned to a nodule was proportional to the risk of malignancy when we used a univariable regression analysis. This strong association between ACR TIRADS category and malignancy was further enhanced by a more rigorous multivariable regression analysis and ultimately was confirmed by the AUC estimate of 0.75. These results corroborate recent reports that have validated the ACR TI-RADS in large adult populations [14–17, 22]. We therefore suggest that this classification can be helpful as a decision-making tool in the management of pediatric thyroid nodules.

Apart from the ACR TI-RADS, other thyroid nodule ultrasound risk stratification guidelines have also recently been found to perform well for children. One study assessed the 2015 ATA Management Guidelines for Children and found that the composite pattern of ultrasound features accurately predicted malignancy, with the odds of malignancy significantly higher in nodules determined to have a high level of suspicion by ultrasound [18]. Other studies found overall good diagnostic performance in correctly classifying pediatric nodules using the ATA system and a different TIRADS developed at a single institution [3, 12]. One reason we sought in the current study to specifically appraise the TI-RADS developed by the ACR was the advantage it may have over other TIRADS classifications with regard to accessibility, applicability, and general reputation and acceptance in practice settings in the United States.

Because of the differing biologic behavior of thyroid cancer in the pediatric versus the adult population, adjustments in the way protocols are applied, particularly with regard to nodule size, may prove prudent. ACR TI-RADS does not consider size in assignment of TI-RADS level in adults but does use size to suggest thresholds for when a nodule should undergo FNAB [6]. In children, on the other hand, it is recommended that sonographically suspicious nodules undergo FNAB regardless of size because malignancy is better predicted by ultrasound features and clinical factors rather than size alone [7, 9]. The results of the current study support this strategy, because, although there was a significant correlation between size and malignancy on univariable analysis, higher ACR TI-RADS categories remained associated with increased likelihood of malignancy on multivariable analysis, whereas nodule size did not.

Although the ACR TI-RADS was useful as a predictor of malignancy in children, its imperfections should be considered. Although intraobserver reproducibility for ACR TI-RADS category was high, interobserver reliability was only moderate, which was not as robust as hoped. Additional analysis of the interobserver data indicated that many of the disagreements were between adjacent TI-RADS categories (e.g., category 3 vs 4, rather than category 3 vs 5). Therefore, although the results were divergent, the extent of the discord was often not substantial.

Of greater concern was the large number of higher suspicion ratings, which is congruent with prior studies in children using different risk stratification guidelines [12, 18]. In our study, we found that the majority of both benign and malignant nodules were rated as ACR TI-RADS category 4 or 5. Furthermore, although ROC analysis confirmed that TI-RADS 5 was the category at which discrimination between malignant and benign nodules was most effective, a higher than desired number of false-positives were observed. Specifically, 19 of 36 (53%) benign nodules were erroneously classified as category 5, meaning these nodules may have undergone unnecessary FNAB. This was reflected in suppressed positive predictive value and specificity at TI-RADS category 5, findings that are supported by a recent study of ACR TI-RADS in adults that reported similarly low specificities, ranging from 44% to 51% [17]. Despite these shortcomings, the correct classification of 35 of 38 (92%) of benign nodules at ACR TI-RADS categories 1, 2, 3, or 4 suggests that, in these cases, FNAB could have been avoided and ultrasound surveillance would have been an alternative management option.

Although this study did not analyze which individual ultrasound features were most responsible for the high false-positive rate of ACR TI-RADS category, some general relevant trends were observed. Punctate echo-genic foci, a feature favoring malignancy, were found not only in most of our malignant nodules, but also in most of our benign nodules a finding supported in other studies [5, 23]. This result could be explained by prior sonographic and histopathologic correlation in children showing that punctate echo-genic foci may represent not only exclusively psammoma bodies that are associated with papillary thyroid carcinoma, but also other entities, such as stromal calcifications and sticky colloid [19]. Nevertheless, radiologists using the ACR TI-RADS should be aware that, with the current point allocation system, the presence of punctate echogenic foci is accorded 3 points, resulting in a disproportionate influence on elevating the TI-RADS score and therefore establishing the level of suspicion. The category of punctate echogenic foci was found in a recent study in adults to be one of the ACR TI-RADS sonographic features for which interobserver agreement was lowest, suggesting that radiologists could benefit from education regarding this important distinguishing feature [16].

Taller-than-wide shape, a suspicious feature, has been reported as rare in pediatric nodules [7, 12]. However, we found this feature in 9.3% of our benign nodules, which is an unexpectedly high frequency and which we hypothesize could also contribute to the high false-positive rate of ACR TI-RADS category. All of the benign nodules with this configuration occurred in the same 14-year-old boy with multinodular goiter (Fig. 3). Inspection of his images revealed that the taller-than-wide shape could be attributed to distortion and mass effect from the sheer size and number of neighboring nodules. This case serves as a reminder that the significance of the taller-than-wide shape should be interpreted in the context of surrounding thyroid tissue because it may represent a manifestation of secondary, rather than intrinsic, contour deformity. Refinements to the current ACR TI-RADS format may be necessary to account for this specific situation.

Although the large number of false-positives poses a challenge to use of the ACR TI-RADS, of equal concern is the false-negative we encountered in which a malignant nodule was assigned a benign rating of TI-RADS category 1. The individual ultrasound features that caused this 0.8-cm malignant nodule in a 17-year-old girl to be assigned ACR TI-RADS category 1 were the selection of cystic for composition and anechoic for echogenicity, both of which correspond to 0 points (Fig. 5B). Although it is possible that the nodule was genuinely cystic, an alternative consideration is that this was a solid nodule that should have received a very hypoechoic rating. This possible misinterpretation illustrates the ease with which very hypoechoic, a feature favoring malignancy and defined as echogenicity less than that of the neck musculature, can be misconstrued as anechoic, a feature favoring benignity, and vice versa, despite their diverging clinical implications [13]. This may be an issue particularly for subcentimeter nodules. In quantitative terms, the difference in the ACR TI-RADS between the two features is 3 points, which could be enough to erroneously shift management recommendations to deferral of FNAB in a nodule that ought to undergo tissue sampling, as well as the reverse.

This case emphasizes the point that sub-centimeter nodules may be difficult to characterize accurately with sonography, mirroring prior studies in children showing that six of 52 malignant nodules were incorrectly classified as very low and low suspicion with ATA criteria [12]. Furthermore, because this patient also had a separate 2.5-cm malignant right nodule (Fig. 5A), management of pediatric nodules should be based on not only its sonographic appearance and ACR TI-RADS rating but also on the broader clinical context, including the presence of additional suspicious nodules.

Limitations of our study include its retrospective design, which resulted in the use of different ultrasound vendors and potential variation in image quality. Another restriction was the small study population and even smaller number of malignant nodules (n = 20), possibly limiting our ability to extrapolate the true population effect. The high rate of false-positives observed at ACR TIRADS category 5 also requires further research to determine whether these positive but preliminary results can be replicated while simultaneously increasing the positive predictive value.

This study shows that the ACR TI-RADS discriminates well between malignant and benign nodules in a pediatric population, particularly at TI-RADS category 5, and is useful in malignancy risk stratification. However, practitioners should be aware of the high rate of false-positives at higher TIRADS categories, as well as the possibility of false-negatives, especially in subcentimeter nodules.