Diagnostic accuracy of ultrasound characteristics in the identification of malignant thyroid nodules

Objective This study was aimed at determining the diagnostic accuracy of the ultrasound characteristics in the identification of malignant thyroid nodules in patients attending the surgical clinic at a tertiary care hospital in Sri Lanka. Results This prospective validation study was conducted among 263 patients attending the surgical clinic at a tertiary care hospital, Sri Lanka. The ultrasound characteristics having statistically significant associations with thyroid malignancies were identified by employing a multivariable analysis. These ultrasound characteristics were compared with Fine Needle Aspiration Cytology results and indicators of diagnostic accuracy were computed. The study sample consisted of 33 malignant and 230 benign nodules. Internal vascularity, hypoechogenicity and microcalcification showed statistically significant positive associations with thyroid malignancy (p < 0.05). The highest positive likelihood ratio was observed for the presence of microcalcifications (10.4, 95% CI 4.6–23.7) with a specificity of 96.5% (95% CI 93.3–98.5%). Hypoechogenicity had a sensitivity of 66.7% (95% CI 48.2–82.0%) and a specificity of 74.8% (95% CI 68.6–80.3%). The presence of at least one positive ultrasound characteristic yielded the highest sensitivity (87.9%, 95% CI 71.8–96.6%), whereas, the presence of all three characteristics yielded a near perfect specificity (99.6%, 95% CI 97.6–99.9%). Hence, the presence of microcalcifications was found to be the single most useful criterion in predicting thyroid malignancy.


Introduction
Globally, thyroid malignancy is the most common endocrine cancer and there has been a steady increase in the global incidence of thyroid cancer over the last few decades [1,2]. Since there is a 5-15% probability of malignancy in any given thyroid nodule [3,4], current thyroid guideline recommendations call for imaging modalities to detection of thyroid malignancies early leading to a better prognosis [5]. Ultrasound scanning is widely cited as the imaging modality of choice for early diagnosis owing to its wide availability, relative simplicity and nonexposure to ionizing radiation [6][7][8][9][10].
Despite the highlighted advantages in ultrasound imaging, the evidence is not conclusive in predicting thyroid malignancies [11,12]. Hence, the diagnostic accuracy of the ultrasound scanning needs to be evaluated by comparing with the results of a gold standard test. Fine needle aspiration cytology (FNAC) is considered as the most reliable cost effective method for definitive evaluation of thyroid nodules [13][14][15][16][17][18][19][20]. Thus, thyroid cancer management guidelines recommend conducting FNAC on any thyroid nodule, which is suspected to be malignant [5,[8][9][10]. Global evidence highlights a multitude of ultrasound characteristics that are indicative of thyroid malignancies [3, 7-9, 13, 14, 16, 19, 21-30]. However, collective global evidence base suggests inconsistent results across studies on predictive abilities.
Thyroid nodules are common clinical presentations to the surgical clinics in Sri Lanka [31]; hence, accurate identification of potential malignancy will have a great impact on patient management. In this background, it is important to utilise the ultrasound imaging techniques optimally, especially to decide on cases necessitating urgent surgical interventions. Despite the significance, only a limited number of studies have been conducted in Sri Lanka to evaluate the diagnostic accuracy of the thyroid ultrasound imaging [32]. In this research vacuum, the present study was conducted to determine the diagnostic accuracy of the ultrasound characteristics in identification of malignant thyroid nodules.

Methods
The findings are reported following the Standards for Reporting of Diagnostic Accuracy Studies (STARDS 2015) guidelines [33].

Study design and setting
This prospective validation study was carried out at the Radiology Department in one of the largest tertiary care centres in the Central Province, Sri Lanka for a period of 18 months from January 2014 to June 2015.

Participants
In this single-gate study, all patients referred to the Radiology Department during the study period for thyroid ultrasound scanning were considered as the study population, excluding patients with known bleeding diathesis. All patients who fulfilled the eligibility criteria were recruited for the study upon the informed written consent.

Hypoechogenicity
All 263 participants have undergone the ultrasound scanning of the thyroid nodules and the above-mentioned ultrasound characteristics were recorded in each participant.

Ultrasound guided FNAC of thyroid nodules
Following the ultrasound scanning, participants were subjected to ultrasound guided FNAC, upon their informed written consent. The FNAC of the thyroid nodules were carried out by the same Consultant Radiologist, who performed the ultrasound scanning, using the same ultrasound scanner used to detect the nodule. The procedure was done under strict aseptic conditions using a 23G needle and three samples were taken from each nodule. Prepared slides were immersed immediately in 95% ethyl alcohol solution, before transporting them to the laboratory.

Cytology of thyroid nodules
The thyroid cytopathology report was prepared according to the Bethesda classification [35]. An experienced Consultant Pathologist (with more than 25 years experience in histopathology), who was blinded to the ultrasound scan findings, evaluated all the samples.
Data analysis was carried out by using Statistical Package for the Social Sciences (SPSS) analytic software version 25.0. Based on FNAC findings, the sample was categorised as benign and malignant.
To identify the ultrasound characteristics with significant associations with thyroid malignancy, a bivariate analysis was conducted and the crude odds ratios (OR) were computed with 95% confidence intervals (95% CI). A binomial multiple logistic regression model was used to control for confounding. Ultrasound characteristics that showed statistical significance (p < 0.05) in the bivariate analysis were included in the multivariable analysis model using backward stepwise elimination method and the model produced adjusted odds ratios (AOR) and 95% CI.
The statistically significant ultrasound characteristics were compared with FNAC results separately and in combination. Performance characteristics of each ultrasound characteristic and combined ultrasound characteristics, namely, sensitivity, specificity, positive and negative predictive values and the positive likelihood ratios (LR+) were computed. Point estimates for the performance characteristics were supported by the 95% CI estimates.

Characteristics of the sample
The study sample consisted of 263 participants and all participants have undergone both ultrasound scanning and ultrasound guided FNAC. The majority of participants were females (n = 247, 93.9%). The age range of the sample was 16 years to 74 years (mean = 46.1 years, SD = 12.7 years).

Diagnosis of the ultrasound guided FNAC
Based on the cytological diagnosis of the thyroid nodules, the sample was categorised as benign and malignant. According to the FNAC results, Thy2 lesions were categorised as benign (n = 163). In addition, the benign category consisted of the Thy3 and Thy4 lesions (n = 79), which underwent surgery (n = 58) and histopathologically confirmed as benign (n = 46). Further, Thy3 lesions, which showed no interval changes after 6 months, were also included in this category (n = 21). The malignant group consisted of all the histological proven malignancies in Thy3 and Thy4 group (n = 12) and those that were Thy5 (n = 10) and Thy6 (n = 11) from the FNAC results. Thus, there were 33 proven malignant thyroid nodules and 230 benign nodules. Table 1 summarises the results of the bivariate analysis of the association between ultrasound characteristics and thyroid malignancy and out of the eight ultrasound characteristics assessed, except for the size and consistency, other characteristics showed statistically significant associations with thyroid malignancy at p < 0.05. Table 2 illustrates the results of the analysis of binary logistic regression of the ultrasound characteristics retained in the final model. Out of the five factors retained in the final model, three factors made unique statistically significant contributions at p < 0.05. When controlled for other factors included in the logistic regression analysis, internal vascularity, hypoechogenecity and microcalcification showed statistically significant positive associations with thyroid malignancy.

Diagnostic accuracy of ultrasound characteristics in the identification of thyroid malignancies
The performance characteristics of individual ultrasound characteristic evaluated against the final diagnosis are summarised in the Table 3.
Amongst the single ultrasound characteristics assessed, the highest LR+ was observed for the presence of microcalcifications and it had a very high specificity, but the sensitivity was relatively low. On the contrary, hypoechogenicity had a relatively higher sensitivity and specificity values.
Amongst the combined ultrasound characteristics, the presence of at least one positive characteristic yielded the highest sensitivity, whereas, the sensitivity values were substantially lower when more than one characteristic was combined. On the contrary, presence of all three characteristics yielded a near perfect specificity.

Discussion
Since the existing literature on diagnostic accuracy of ultrasound characteristics in predicting thyroid malignancies fails to provide conclusive evidence, we considered a number of important ultrasound characteristics cited in the literature for the present study.
Since experienced consultants in the relevant fields interpreted the findings of the index test and the reference standard separately based on clear, explicit guidelines, the risk of misclassification bias is minimised in the study. Furthermore, given that a single consultant carried out interpretation of ultrasound imaging and cytology separately, there was no inter-observer bias in gathering information. Since the Consultant Pathologist was blinded to the results of the ultrasound scan findings of the participants, there was no risk of bias in interpreting the reference standard test.
In the present study, ultrasound scanning was considered as the index test and FNAC diagnosis, which is cited as the gold standard diagnosis in thyroid malignancies [14,15,[17][18][19][20] was used as the reference standard.
Rather than evaluating all ultrasound characteristics for their diagnostic accuracy, statistically significant characteristics were identified by employing both bivariate analysis followed by the multivariable analysis. This analysis was important in addressing the potential confounding effect of ultrasound characteristics on each other and has aided in determining the most significant characteristics to be used in diagnostic accuracy evaluation. The analysis has yielded three significant ultrasound characteristics, namely, microcalcifications, hypoechogenicity and internal vascularity. These findings are consistent with most of the existing literature, as many studies have revealed satisfactory indicators of diagnostic accuracy of microcalcification [7-9, 14, 16, 19, 21, 29], hypoechogenicity [8,   [7,13,16,19,[28][29][30] in predicting thyroid malignancy. The presence of microcalcifications is associated with a LR+ of 10.4 (95% CI 4.6-23.7), indicating that the likelihood of having thyroid malignancy has increased by approximately ten-fold given the positive ultrasound characteristic. However, this characteristic was found to have a low sensitivity of 36.4%, which is consistent with low sensitivity (26.1-59.1%) values reported in the literature [3,14,16,21,29]. In contrast, the specificity of this ultrasound characteristic was very high (96.5%), which is consistent with previously reported values of 90.8% [21] and 87.8% [19]. The strong association of microcalcifications in thyroid malignancies is thought to be due to the fact that microcalcifications correspond pathologically to calcified 'psammoma bodies' that are typical of papillary cancer [21,36,37]. In contrast, macrocalcifications or coarse calcifications are related to fibrosis and degeneration [38,39] and benign nodules of long duration are mostly associated with coarse calcifications [40,41].
In the present study, hypoechogenicity was shown to have a sensitivity of 66.7% and a specificity of 74.8%. These findings are congruent with existing literature, in which the sensitivity was reported as 62.7% in a metaanalysis [19]. However, literature suggests that the sensitivity values range from 26.5% [14] to 87.2% [21]. In a similar vein, even though the meta-analysis findings reveal a specificity of 62.3% [19], the values range from 58.5% [21] to as high as 94.3% [14]. Furthermore, the LR+ reported in the present study is similar to that in the meta-analysis [19].
The study findings reveals that internal vascularity had a sensitivity of 48.5%, specificity of 79.1% and a LR+ of 2.3, corresponding with the meta-analysis findings of 45.9%, 78.0% and 2.1 of sensitivity, specificity and LR+ respectively [19]. On the contrary, some studies have concluded that the intra-nodular colour flow alone has a lower predictive value in assessing the malignant thyroid nodules [13,42].
In relation to the combined ultrasound characteristics, the presence of at least one positive characteristic yielded the highest sensitivity, whereas, the presence of all three characteristics yielded a near perfect specificity. Furthermore, when considering multiple ultrasound characteristics in combination, the LR+ of detecting malignant thyroid nodules increased. These findings are congruent with the findings of the existing literature [16,32].
Our study showed that the presence of microcalcifications and hypoechogenicity are the most significant criteria in predicting thyroid malignancies, which are also included in the TI-RADS classification. In addition to the characteristics included in the TI-RADS classification, internal vascularity has shown a statistically significant association in our study. Thus, further studies are needed to explore the role of vascularity in predicting thyroid malignancies.

Limitations
In this study, all the subjects with benign FNAC results did not attend the follow-up ultrasound scan in 6 months. Of Thy3 and Thy4 lesions, 58 subjects underwent surgical removal and the final diagnosis was obtained from the histopathological report. Twentyone subjects with Thy3 lesion did not undergo surgery, but attended follow-up ultrasound scan.