Use of the Kwak Thyroid Image Reporting and Data System (K-TIRADS) in differential diagnosis of thyroid nodules: systematic review and meta-analysis

Ultrasound imaging is a basic technique used in the visualization and characterization of focal thyroid lesions and to estimate the risk of malignancy. The suspicious nature of lesions is confirmed on the basis of cytological examination of specimens collected via ultrasound-guided fine-needle aspiration biopsy (FNAB) and further histological examination if necessary [1]. Currently, ultrasound and FNAB serve as essential tools for diagnosing thyroid nodules [1, 2]. Numerous studies have shown that ultrasonography has an important place in the diagnosis of malignant and benign thyroid lesions and is marked by high sensitivity and low specificity [3]. Many lesions without suspicious features may be observed conservatively in this manner without the necessity of biopsy [2, 4].

Currently, the fundamental technique used for clinical assessment of thyroid lesions is grey-scale imaging (B-mode). In B-mode, suspicious features include: solid nature; low or very low echogenicity; irregular, microlobular or blurred borders; vertical shape or an anteroposterior diameter greater than lateral; and microcalcifications [2, 5, 6].

There are numerous guidelines of many medical societies worldwide that describe the clinical and ultrasound features that necessitate FNAB [7‐9]. However, there is a need for application of routine, adequate and common standardization system of thyroid nodules ultrasound classification. The system has already been proposed and is called Thyroid Imaging Reporting and Data System (TIRADS). It relies on B-mode imaging and represents an important step in standardization of ultrasound examination of the thyroid. TIRADS has its foundation in the Breast Imaging Reporting and Data System (BIRADS) classification [10‐12], which is based on varying, increasing the risk of malignancy of focal lesions in different categories. Data relating to TIRADS classification were first published in 2009 by two independent teams led by Horvath [13] and Park [14]. The two different approaches proposed by the teams, in our opinion, proved to be complicated and difficult to use on a daily basis. The later study published by Kwak in 2011 [6] had a different approach to this classification. Whether the individual lesion belongs to a TIRADS category or not was determined based on the number of suspicious features, including solid structure, low or very low echogenicity, irregular or microlobular borders, microcalcifications and vertical shape (TIRADS 3 = no suspicious features; TIRADS 4a = 1 suspicious feature; TIRADS 4b = 2 suspicious features; TIRADS 4c = 3 or 4 suspicious features; TIRADS 5= 5 suspicious features).

Furthermore, other groups proposed different TIRADS interpretations [15‐17].

The aim of this study was to conduct a systematic literature review and to assess the diagnostic utility of the Kwak’s TIRADS classification in the risk stratification of thyroid nodules in adults.

Ultrasonography is considered to be the test of choice in preoperative diagnosis of thyroid nodules [6]. There is no doubt that the coexistence of a greater number of suspicious features in a focal lesion significantly increases the risk of malignancy compared to a single suspicious feature. At the same time, the lack of a unified system for categorizing nodules often caused misunderstandings between radiologists and clinicians. The proposed TIRADS classification was intended to be a response to these needs. A large number of studies have been published evaluating the discriminatory capabilities of TIRADS based on different classifications [13, 32‐34] and recent WFUMB (World Federation for Ultrasound in Medicine and Biology) guidelines have suggested using TIRADS in order to improve characterization of thyroid nodules, and especially the communication between specialists and patients [35].

Previous meta-analyses evaluating the diagnostic utility of the TIRADS classification took into account its different variants, which could significantly affect the results obtained. In Wei's 2016 study, reported pooled sensitivity and specificity were 0.79 and 0.71, respectively, and for AUC was 0.92 for the summarized ROC. The results of the current meta-analysis including six studies and a total of 10,926 nodules show that the classification proposed by Kwak has a much higher pooled sensitivity of 0.983 and lower specificity of 0.552 in differentiation of thyroid nodules, compared to the previous meta-analyses [36, 37]. Other parameters, such as pooled positive likelihood ratio (2.67, 95 % Cl 1.69–4.2) and negative likelihood ratio (0.05, 95 % Cl 0.03–0.07), were slightly lower than those reported in Wei's study, while the diagnostic odds ratio of 51.02 (95 % CI 15.24–170.79) was considerably higher. Apart from these differences, the parameters also indicate good diagnostic value of the TIRADS classification proposed by Kwak.

The initial assessment of obtained studies suggested heterogeneity, which after more detailed analysis turned out to be statistically insignificant with p > 0.05. Exploration of possible causes revealed that important factors that may increase the heterogeneity of results include study type and country of origin. Analysis of subgroups showed a statistically significant difference between retrospective and prospective studies with p < 0.0001 and significantly higher values of diagnostic odds ratio in prospective studies (361.91 vs. 23.023 for retrospective studies). Further analysis demonstrated that in prospective studies, the pooled sensitivity was lower than in the retrospective study group (0.972 vs. 0.985). The pooled specificity was significantly higher (0.911 vs. 0.263). The results of prospective studies that verify nodules by means of cytological and histopathological evaluation better reflect everyday practice than retrospective studies, which classify material via histopathological verification.

Another essential feature impacting the heterogeneity of results was the country of origin. It is worth noting that three of the six studies originated from Korea, one was from India, and two were from China. Subgroup analysis demonstrated a statistically significant difference between the studies originating from Korea versus other countries (p = 0.0244). In China, cytological verification of focal lesions is rarely performed and thyroid cancer is confirmed mainly on the basis of histological examination [37]. The result is that in the case of patients with nodules classified as TIRADS 1–3 and first cytological, non-diagnostic, atypical or follicular lesion of undetermined significance (Bethesda categories I and III), the invasive diagnostics were not enhanced with another aspiration biopsy or surgery. As a consequence, this causes a lack of diagnosis of cancer in these TIRADS categories and negatively affects the assessment of specificity for this classification [37].

In analysed publications the cut-off point was different. In two cases the cut-off was 3/4a, while in the remaining four it was 4a/4b (Table 2). There is a high difference in specificity, which is higher for 3/4a 0.874 (95 % CI 0.863–0.884) compared to 4a/4b 0.392 (95 % CI 0.379–0.405), when sensitivity for 3/4a 0.979 (95 % CI 0.959–0.991) is almost parallel to 4a/4b 0.971 (95 % CI 0.961–0.980). Similar results were obtained for accepted reference methods. In two cases authors used cytology, histology and ultrasound follow-up for non-suspicious lesions in ultrasound examination or after initial benign cytology (Table 2). In both cases accepted cut-off differences and reference standards did not significantly influence homogeneity of pooled data p > 0.05 (Table 2).

In our results comparing French-TIRADS we report slightly lower specificity, 0.552 versus 0.61, but sensitivity was higher, 0.983 versus 0.957 [15]. The specificity difference could be the consequence of implementation of elastography in French-TIRADS. Obtained pooled data turned out to be lower compared to Horvath’s recent work from 2016 reporting sensitivity of 0.996 and specificity of 0.744. A recently published work on TI-RADS by ACR (American College of Radiology) proposed a different approach [38]. The study is based on a previously published ACR Thyroid Ultrasound Reporting Lexicon [39]. The authors proposed a five-grade scale of TR1 (benign) to TR5 (high suspicion of malignancy) based on scoring of five nodule characteristics (composition, echogenicity, shape, margin and echogenic foci). It was noted that due to the lack of elastography in each ultrasound scanner, it was not included in the ACR TI-RADS. This system needs to be verified as a tool in thyroid nodule stratification.

To minimise the erroneous selection of publications in the course of this systematic review, four databases were used: PubMed, Cochrane database, ScienceDirect and EMBASE. In addition, in order to assess the quality of publications the updated version of the QUADAS-2 tool was used [19].

This systematic review also has some limitations. First, in some studies it was not clearly established if the interpretation of ultrasound images of focal lesions on the basis of tested K-TIRADS classification was carried out without knowledge of the results of reference tests and vice versa. At this level, we cannot clearly determine whether this error resulted from incorrect planning of the original study methodology or from inadequate reporting. Currently, it is recommended that people studying diagnostic accuracy use the Standard for Reporting Diagnostic Accuracy check-list to minimize errors in publication of results [40]. Second, the final diagnosis was not always established on the basis of histopathological examination. Patients in whom lesions were classified as categories 1–3 were included into routine ultrasound follow-up similarly, and not in every case of a focal lesion with category 4–5 was histopathological verification performed; some diagnoses were established on the basis of cytological examination. Third, some articles found in the database searches were rejected as they were in a language other than those approved in the study protocol. Fourth, the disadvantage of previous studies is the small number of non-papillary carcinomas, which often present with a different appearance on the ultrasound examination. Particularly interesting is the group of follicular lesions of indeterminate cytology in which elastography may be useful in differentiating benign and malignant nodules [41]. The creation of a final, comprehensive, ultimate TIRADS classification in the future should include evaluation of a significant number of less prevalent non-papillary cancers as their proper diagnosis is a very important issue from a medical point of view.

K-TIRADS classification has high sensitivity and low specificity. Thus, it is very useful to discard the benign cases and to reduce the number of biopsies. Further prospective, multicentre studies are needed in order to systematize this classification.