Development of a malignancy risk prediction model combining ultrasound and cytology for Bethesda category III thyroid nodules

Abstract

Background

Bethesda category III thyroid nodules pose substantial management challenges, and a validated malignancy risk stratification system is urgently needed to guide clinical decision-making. This study aims to identify specific sonographic and cytological features associated with malignant Bethesda III thyroid nodules and to develop a practical model for predicting malignancy risk through an integrated analysis of multimodal data from histopathologically confirmed cases.

Methods

This retrospective study used clinical data from patients with both Bethesda III cytological diagnoses and corresponding surgical pathology outcomes between July 2016 and December 2024. The sonographic, clinical, and cytological characteristics of benign and malignant nodules were systematically assessed and compared. Univariable and multivariable logistic regression analyses were conducted to evaluate the association between these features and malignancy. A prediction nomogram, incorporating sonographic and cytological features independently associated with malignancy, was developed and validated to assess its performance.

Results

In this study, a total of 187 Bethesda III thyroid nodules were analyzed, consisting of 77 benign nodules and 110 malignant nodules. Factors such as maximum diameter ≤ 1 cm, absence of smooth margins, microcalcifications, and nuclear atypia in cytology were identified as independent factors associated with malignancy. A prediction nomogram model was developed using these variables, demonstrating strong performance in distinguishing between benign and malignant nodules categorized as Bethesda III, with an area under the curve (AUC) of 0.874.

Conclusions

In Bethesda III thyroid nodules, malignant cases were associated with suspicious sonographic characteristics, smaller size (≤ 1 cm), and cytological nuclear atypia. The nomogram developed from these predictors demonstrates potential utility in estimating malignancy risk for Bethesda III thyroid nodules.

Background

The thyroid gland, consisting of two connected lobes, is one of the largest endocrine glands in the human body, weighing 20–30 g in adults. Thyroid lesions are highly prevalent in the general population and are frequently discovered during routine physical examinations. Most of them are asymptomatic, and thyroid hormone secretion is normal [1]. Ultrasound is the primary imaging modality for evaluating thyroid nodules. The widespread use of high-resolution ultrasound technology has resulted in detection rates of up to 68% in population-based screenings [2]. Beyond detection, ultrasound enables the comprehensive assessment of nodule malignancy potential. The Thyroid Imaging Reporting and Data System (TIRADS) has gained widespread clinical acceptance, providing standardized criteria for malignancy risk stratification and clinical decision-making. This systematic approach not only estimates the likelihood of malignancy but also establishes clear indications for biopsy based on nodule size and risk classification [3].

In cases where imaging findings are indeterminate, ultrasound-guided fine-needle aspiration biopsy (FNAB) is recommended as the next diagnostic step. FNAB of thyroid nodules is widely used in clinical practice due to its minimal invasiveness, ease of operation, and diagnostic accuracy, making it the gold standard for preoperative evaluation [4]. This procedure plays a crucial role by identifying candidates for surgical intervention and providing definitive diagnostic information for indeterminate nodules. The cellular samples obtained through FNAB are categorized using the Bethesda System for Reporting Thyroid Cytopathology (TBSRTC), which has revolutionized cytopathological reporting. This standardized classification system greatly improves inter-clinician communication and ensures consistent diagnostic interpretation across institutions [5].

The TBSRTC’s strength lies in its robust predictive validity. Benign specimens (Bethesda II) show a low malignancy risk of 2–7% with a high negative predictive value of 96.3%, while malignant lesions (Bethesda VI) present near-certain malignancy risk of 97–100% with an exceptional positive predictive value of 98.6%⁶. However, atypia of undetermined significance (AUS, Bethesda III) lesions introduce significant clinical ambiguity, necessitating more detailed discussions among surgeons, radiologists, and patients to balance surgical intervention against active surveillance. Accounting for approximately 10% of thyroid FNABs, Bethesda III nodules pose substantial management challenges [6]. The 2023 TBSRTC update reports an adjusted risk of malignancy of 13–30% (mean 22%) for this category [7], with surgical histology revealing benign outcomes in about 80% of resected Bethesda III nodules [8]. Current management strategies recommend sequential approaches such as repeat FNAB, molecular testing, or diagnostic lobectomy [9]. Nevertheless, 17–40% of these indeterminate nodules remain diagnostically unresolved despite comprehensive evaluation. Definitive diagnosis through thyroidectomy carries inherent surgical risks, including hemorrhage, wound infection, and recurrent laryngeal nerve injury that may result in permanent voice changes. Postoperative hypothyroidism affects 20–30% of lobectomy patients, necessitating lifelong thyroid hormone replacement [10]. These points emphasize the clinical need for improved risk stratification in Bethesda III nodules to reduce unnecessary surgeries and enable timely intervention for malignant cases.

Current clinical practice employs various diagnostic modalities to differentiate benign and malignant Bethesda III thyroid nodules, including molecular testing, ultrasound feature analysis, elastography, and nuclear medicine imaging [11]. However, these approaches exhibit variable efficacy and lack standardized diagnostic criteria. Studies utilizing conventional ultrasound and elastography have identified malignancy-associated features, such as hypoechogenicity, microcalcifications, taller-than-wide shape, irregular margins, and increased stiffness [8, 12]. Despite this, the diagnostic specificity of these features is limited, as nodules selected for FNAB often already display suspicious ultrasound characteristics. Molecular testing of FNAB specimens has become increasingly important for surgical decision-making, with emerging evidence supporting its predictive value for Bethesda III nodules through methods such as gene expression classifiers, gene mutation panels, and microRNA expression [13–15]. Unfortunately, molecular diagnostic tests are costly, not covered by insurance, and limited by the availability of suitable specimens, which currently restricts their widespread use [16]. The 2023 TBSRTC update further stratifies Bethesda III lesions into two subtypes: AUS-nuclear atypia and AUS-other, with the former exhibiting approximately nine times the malignancy risk (59% vs. 6.5%) of the latter [17, 18]. While subtyping within Bethesda III lesions provides additional stratification of malignancy risk for thyroid nodules, its diagnostic accuracy remains unsatisfactory.

Consequently, thyroid nodules categorized as Bethesda III present diagnostic challenges for both radiologists and surgeons. Given the limited reliability of ultrasound features and cytological characteristics in distinguishing between benign and malignant Bethesda III lesions, a validated malignancy risk stratification system is urgently needed to guide clinical decision-making. This study aims to identify sonographic and cytological characteristics specific to malignant Bethesda III nodules and develop a practical malignancy risk prediction model through an integrated analysis of multimodal data, including ultrasound parameters, clinical variables, and cytological findings from histologically confirmed cases. The goal is to improve the clinical management and therapeutic strategies of Bethesda III thyroid nodules, and reduce unnecessary surgeries.

Materials and methods

Approval for this study was obtained from the institutional review board, which granted a waiver of informed consent due to the retrospective nature of the data collection and analysis.

Patient selection

This retrospective study utilized clinical data from patients who underwent both FNAB and thyroidectomy procedures at Peking University Third Hospital. Pathological records from July 2016 to December 2024 were systematically reviewed to identify cases with both Bethesda III cytological diagnoses and their corresponding surgical pathology outcomes. Eligible cases were those that showed Bethesda III cytology results with confirmed postoperative pathology. The clinical management after obtaining Bethesda III cytological results was determined by surgeons based on TIRADS categories, nodule size and location, as well as patient preferences. Thyroid nodule sonograms were retrieved from the Picture Archiving and Communication System (PACS) for analysis. Exclusion criteria included: (1) Poor-quality ultrasound images that hindered accurate evaluation of nodular characteristics; (2) Inability to conclusively match FNAB-targeted nodules with corresponding surgical pathology reports; (3) Histological evidence of neoplasms with undetermined malignant potential. The study design’s flowchart is depicted in Fig. 1.

Fig. 1.

Flowchart of the study design

Ultrasound image review

Following the anonymization of clinical data, all ultrasound images were assigned random numbers as independent files for blinded evaluation. Each image in the PACS database was systematically analyzed by two radiologists specializing in thyroid imaging, with 11 and 18 years of diagnostic experience, respectively, to achieve a consensus interpretation. Discrepancies in initial assessments were resolved through joint reevaluation to establish a conclusive assessment. Both evaluators remained blinded to the original ultrasound reports and histopathological outcomes throughout the process. For subjects with multiple ultrasound examinations, the study prioritized the examination closest in time to the FNAB procedure. Nodule characterization was based on the 2017 ACR TI-RADS (ACR Thyroid Imaging Reporting and Data System) lexicon, with systematic evaluation of ultrasound features including: the thyroid background of nodules (the thyroid parenchyma surrounding the nodules), three-dimensional measurements (anteroposterior, transverse, and longitudinal dimensions), margin, composition, echogenicity, shape (taller-than-wide or wider than tall), echogenic foci (microcalcifications, macrocalcifications, peripheral calcifications), and color Doppler signals [19].

Imaging technique

Twenty radiologists, each with a minimum of three years of specialized experience in thyroid imaging, performed the ultrasound examinations. The procedures were performed using ultrasound systems (Samsung Edison RS 80, Mindray Resona 7, or GE Logiq 9) equipped with high-frequency linear-array transducers (5–12 MHz or 8–15 MHz). The imaging protocol included standardized acquisition of transverse and longitudinal gray-scale planes to document thyroid nodule characteristics, with additional color Doppler imaging to evaluate the vascular patterns within the lesions.

Pathologic result review

All cytological slides obtained from thyroid FNAB were re-evaluated by a pathologist with over 15 years of specialized experience in thyroid pathology. This secondary review aimed to subclassify Bethesda III nodules into two distinct subtypes: AUS-nuclear atypia and AUS-other. Subsequent analyses of surgical reports were performed to confirm diagnostic accuracy. Both cytological and histopathological assessments followed the diagnostic criteria outlined in the WHO 2017 classification for endocrine tumors and the revised 2023 Bethesda System for Reporting Thyroid Cytopathology (TBSRTC).

Statistical analysis

Statistical analyses were performed using SPSS software (version 27.0, IBM, Armonk, NY, USA) and the R analysis platform (version 4.4.2). Data distribution patterns were assessed through the Kolmogorov-Smirnov normality test. Normally distributed data were expressed as mean values with standard deviations (SD), while non-normally distributed data were represented by medians and interquartile ranges. Comparative analyses between benign and malignant groups were conducted using the independent samples t-test for normally distributed variables and the Mann-Whitney U test for non-normally distributed variables. Categorical variables were compared via Pearson’s chi-square test or Fisher’s exact test, with results presented as frequencies.

Logistic regression was used to evaluate the predictive value of sonographic and cytological characteristics for malignancy risk prediction. Variables with significant associations in univariate analysis were retained for multivariable analysis, and effect sizes were quantified using odds ratios (OR) and 95% confidence intervals (CI). A clinical prediction nomogram was then developed, implementing a proportional weighting system in which the strongest predictor was assigned a score of 100 points. Model performance was evaluated through the area under the receiver operating characteristic curve (AUC), supplemented with internal validation via 1,000 bootstrap resamples. Calibration curve analysis, comparing predicted and observed malignancy probabilities, included two reference components: the “ideal” line, which represents perfect prediction alignment, and the “apparent” curve, which reflects the calibration of the entire cohort.

Statistical significance was set at P < 0.05 (two-tailed) for all analyses.

Results

Distribution of pathological types

A total of 201 patients (202 nodules) with Bethesda category III cytological results and subsequent surgical pathology confirmation were initially identified.

After excluding 2 nodules with inadequate imaging quality, 2 FNAB-targeted nodules that could not be definitively matched with surgical pathology reports, and 11 nodules with undetermined malignant potential, 186 patients (187 nodules) were included in the final analysis. Among these, 77 nodules (41.2%) were benign, and 110 (58.8%) were malignant. As shown in Table 1, nodular goiter was the most common benign lesion (51.9%), while papillary thyroid carcinoma was the predominant malignant tumor (90.9%).

Table 1.

Distribution of pathological types in Bethesda III thyroid nodules

Benign(n = 77)	Malignant(n = 110)
Nodular goiter/40	Papillary Thyroid Carcinoma/100
Follicular Thyroid Adenoma/24	Follicular Thyroid Carcinoma/6
Subacute Thyroiditis/5	Medullary Thyroid Carcinoma/1
Hashimoto’s thyroiditis/8	Hürthle Cell Carcinoma/3

Clinical, cytopathological and sonographic characteristics

A comprehensive comparison of patient demographics, cytological findings, and sonographic features between benign and malignant nodules is summarized in Table 2. No significant differences were observed between the groups in terms of age or gender distribution. Benign nodules exhibited significantly larger maximum diameters compared to malignant ones (1.7 ± 1.1 cm vs. 1.1 ± 0.8 cm), with 67.3% of malignant nodules measuring ≤ 1 cm, compared to 22.1% of benign lesions. The malignant group also showed a significantly higher prevalence of suspicious sonographic features, including solid composition, very hypoechoic appearance, taller-than-wide shape, irregular margins, and microcalcifications (all P < 0.01). Macrocalcifications were relatively common in both groups, with no significant difference between the benign and malignant groups (20.8% vs. 30.9%). Cytological analysis revealed significantly higher rates of nuclear atypia in malignant nodules (87.3% vs. 36.4%, P < 0.01). Malignant nodules were more strongly associated with Hashimoto’s thyroiditis, whereas benign nodules more commonly exhibited hypervascularity.

Table 2.

Comparison of clinical, cytopathological and sonographic features between benign and malignant nodules

	Benign(n = 77)	Malignant(n = 110)	p
Age(years), mean ± SD	50.7 ± 12.6	47.3 ± 13.5	0.06
Gender			0.32
Male	68.8%(53/77)	75.5%(83/110)
Female	31.2%(24/77)	24.5%(27/110)
AUS nuclear atypia	36.4%(28/77)	87.3%(96/110)	< 0.01*
Maximum diameter(cm)	1.7 ± 1.1	1.1 ± 0.8	< 0.01*
Maximum diameter ≤ 1 cm	22.1%(17/77)	67.3%(74/110)	< 0.01*
Solid composition	77.9%(60/77)	93.6%(103/110)	< 0.01*
Very hypoechoic echogenicity	9.1%(7/77)	27.3(30/110)	< 0.01*
Taller-than-wide shape	15.6%(12/77)	55.5%(61/110)	< 0.01*
Margin
Irregular	48.1(37/77)	78.2%(86/110)	< 0.01*
Ill-defined	16.9%(13/77)	20.0%(22/110)	0.7
Smooth	33.8(26/77)	1.8%(2/110)	< 0.01*
Calcification
Microcalcification	37.7%(29/77)	71.8%(79/110)	< 0.01*
Macrocalcification	20.8%(16/77)	30.9%(34/110)	0.13
Peripheral calcification	2.6%(2/77)	0.9%(1/110)	0.57
Hypervascularity	83.1%(64/77)	66.4%(73/110)	0.01*
Hashimoto’s thyroiditis background	20.8%(16/77)	49.1%(54/110)	< 0.01*

* Statistically significant difference

Development of the malignancy risk prediction model for Bethesda III nodules

We developed a logistic regression model that integrates sonographic and cytological parameters, which show significant differences between the benign and malignant groups (Table 3). The initial univariate analysis identified eight predictors of malignancy: AUS nuclear atypia, maximum diameter ≤ 1 cm, solid composition, very hypoechoic appearance, taller-than-wide shape, irregular margin, microcalcification, and hashimoto’s thyroiditis background. In contrast, smooth margin and hypervascularity were associated with benignity. The multivariate analysis revealed four independent determinants for differential diagnosis: AUS nuclear atypia, maximum diameter ≤ 1 cm, smooth margin, and microcalcification. Based on these parameters, we developed a nomogram (Fig. 2) with a weighted scoring system: the absence of smooth margins received 100 points, while other factors were scaled proportionally according to their effect relative to this reference parameter.

Table 3.

Univariate and multivariate analysis of benign and malignant groups

	Univariate Analysis			Multivariate Analysis
	OR	95% CI	P	OR	95% CI	P
AUS nuclear atypia	12.0	5.8–24.9	< 0.01*	5.3	2.0–13.6.0.6	< 0.01*
Maximum diameter1cm	7.3	3.7–14.2	< 0.01*	2.9	1.1–7.9	0.03*
Solid composition	4.2	1.6–10.7	< 0.01*	0.6	0.2–2.3	0.46
Very hypoechoic appearance	3.8	1.6–9.1	< 0.01*	1.3	0.4–4.2	0.23
Taller-than-wide shape	6.7	3.3–13.9	< 0.01*	2.1	0.8–5.3	0.11
Irregular margin	3.8	2.1–7.3	< 0.01*	1.6	0.6–4.4	0.4
Smooth margin	0.03	0.01–0.16	< 0.01*	0.2	0.0–0.9.0.9	0.04*
Microcalcification	4.2	2.3–7.8	< 0.01*	3.4	1.5–7.7	< 0.01*
Hypervascularity	0.4	0.2–0.8	0.01*	1.0	0.3–3.0.3.0	0.97
Hashimoto’s thyroiditis background	3.7	1.9–7.2	< 0.01*	1.8	0.7–4.5	0.24

* Statistically significant difference

Fig. 2.

The nomogram integrating sonographic and cytological predictors. The scoring scale ranges from 0 to 100 points, as shown on the top axis. Predictive variables (axes 2–5) are aligned with their corresponding point values on the primary scale. The total risk score for the nodule (axis 6) is calculated by summing the points of the predictors, with the final malignancy probability displayed on axis 7

Evaluation and validation of the prediction model

The calibration curves depicted in Fig. 3 demonstrate a good fit between predicted and observed malignancy probabilities (Brier score = 0.134). Analysis of the receiver operating characteristic curve revealed clinically significant discriminatory performance, with an AUC of 0.874 (95%CI: 0.822–0.926) as shown in Fig. 4.

Fig. 3.

Calibration curves for the malignancy risk prediction model. The accuracy of the model is indicated by a higher level of alignment between the solid line and the “Ideal” line

Fig. 4.

Receiver operating characteristic (ROC) curves for the nomogram. AUC, area under the receiver operating characteristic curve; CI, confidence interval

Discussion

Risk stratification of thyroid nodules categorized as Bethesda III on cytology poses inherent challenges in their clinical management. This study explores the reliability of integrating sonographic and cytological features to develop a diagnostic stratification system for distinguishing between benign and malignant Bethesda III nodules. Histopathological results revealed malignancy in 58.8% of the resected nodules. This high malignancy rate underscores the diagnostic challenge of Bethesda III nodules, while the benign pathology found in 41.2% of cases indicates a significant rate of potentially unnecessary surgeries. This malignancy rate substantially exceeds the 22% reference value reported in current guideline for Bethesda III nodules [7], although it aligns with the wide range (12–80%) documented in previous studies [5, 20, 21], suggesting these nodules may have a higher risk of malignancy than traditionally believed [22]. The observed discrepancy may be attributed to selection bias in our cohort, as nodules referred for FNAB exhibited more suspicious sonographic characteristics. This clinical dilemma surrounding Bethesda III thyroid nodules not only creates diagnostic uncertainty but also leads to patient anxiety and increased healthcare costs. Establishing an evidence-based malignancy risk prediction system is therefore crucial for optimizing clinical management and surgical decision.

This study demonstrates that the AUS-nuclear atypia subtype within Bethesda III is associated with a higher malignancy rate. The malignancy risk for AUS-nuclear atypia is approximately 3.5 times higher than that of AUS-other (77.4% vs. 22.2%). However, it does not reach the discriminatory capacity for malignancy risk proposed by the 2023 TBSRTC (59% vs. 6.5%) for the two subtypes [7]. Notably, nuclear atypia was observed in 36.4% of benign nodules and 87.3% of malignant cases. While nuclear atypia showed independent predictive value in multivariate analysis and showed moderate utility in differentiating benign and malignant Bethesda III nodules, its sensitivity and specificity remains suboptimal. These limitations suggest that nuclear atypia alone is insufficient as a predictor and should be interpreted in conjunction with comprehensive evaluations.

Most versions of the Thyroid Imaging Reporting and Data System (TIRADS) recommend biopsy based on nodule size [3, 19, 23]. Previous studies have suggested that nodule size is associated with the diagnostic accuracy of FNAB, indicating that thyroid nodules smaller than 1 cm may yield limited FNAB diagnostic accuracy due to insufficient cellular sampling, higher rates of nondiagnostic results, and false-negative cytology [24–26]. Our study revealed that Bethesda III nodules larger than 1 cm were more likely to be benign (62.5%), significantly higher than those smaller than 1 cm (18.7%). We hypothesize that it is less likely to fail in identifying malignant features due to insufficient cellularity for larger nodules. Consequently, nodules larger than 1 cm diagnosed as Bethesda III on cytology are more likely to be benign. In multivariate analysis, a maximum diameter greater than 1 cm was also identified as an independent predictor of benignity.

This study found that malignant nodules categorized as Bethesda III were more likely to exhibit malignant sonographic features, such as very hypoechoic echogenicity, a taller-than-wide shape, and irregular margins, consistent with prior researches [27–29]. However, among Bethesda III nodules undergoing FNAB, regardless of the final pathological outcome (benign or malignant), there was a higher prevalence of ultrasound features indicative of malignancy and a tendency for higher TIRADS grades. These findings suggest a potential selection bias, as these nodules were biopsied due to suspicious imaging characteristics. Multivariate analysis identified a smooth margin as an independent and strong predictor of benignity, meaning its presence strongly suggests that the nodule is benign. According to multivariate analysis, this model incorporated only two ultrasound features—smooth margin and microcalcification. This suggests that the application of ultrasound features alone has relatively limited value in differentiating benign from malignant Bethesda III thyroid nodules.

The diagnosis of thyroid nodules categorized as Bethesda category III presents clinical challenges. Patients with indeterminate cytology may face three suboptimal management options: repeat fine-needle aspiration, molecular testing, or diagnostic thyroid surgery, which may produce similar indeterminate results, significant healthcare costs, or risks of complications. A comprehensive assessment approach incorporating ultrasonographic features, cytological examination, and other relevant parameters is essential for accurately differentiating between benign and malignant thyroid nodules, as even nodules classified as cytologically benign (Bethesda category II) carry a measurable risk of malignancy [30]. Our prediction model provides surgeons with additional insights into the malignant potential of these nodules, potentially guiding better clinical decisions. Malignancy risk prediction models for Bethesda III nodules remain limited in the literature, with most relying on radiomic features that require specialized imaging techniques not available in routine clinical practice. In contrast, our model incorporates readily accessible clinical parameters, including sonographic characteristics, nodule dimensions, and cytological subclassifications, which are routinely available in clinical settings. The limited number of predictors and the straightforward design of our model ensure its widespread clinical applicability without the need for specialized equipment or additional testing. This model provides additional evidence to support clinical management decisions for Bethesda III thyroid nodules. Based on the prediction results, surgeons may determine subsequent management strategies—such as repeat FNAB, molecular testing, surgery, or continued surveillance.

Four main limitations should be acknowledged. First, our retrospective study design using routine clinical data may introduce inherent biases. Specifically, because FNAB were primarily performed on nodules with suspicious ultrasound features, this selection bias likely elevated the observed malignancy rate. Prospective validation of the risk prediction model remains necessary to confirm its clinical utility. Secondly, as most retrospective samples lacked molecular testing results, this study did not incorporate molecular data. However, given that molecular testing is often expensive and challenging to implement routinely, the exclusion of this factor enhances the model’s generalizability. Consequently, the model can be applied before the decision of repeat FNAB or molecular testing. Thirdly, this study did not include Bethesda IV thyroid nodules—another category of indeterminate nodules within the Bethesda System. As Bethesda IV nodules represent follicular neoplasms with distinct sonographic features and diagnostic pathways compared to other thyroid nodule types, this lesion category will be analyzed separately in subsequent research. Lastly, the single-center nature of the study and its limited sample size restrict external validation. Future multicenter studies with larger cohorts are planned to confirm the validity of this model.

Conclusions

In Bethesda III thyroid nodules, malignant cases were associated with suspicious sonographic characteristics, smaller size (≤ 1 cm), and cytological nuclear atypia. The nomogram developed from these predictors demonstrates potential utility in estimating malignancy risk for Bethesda III thyroid nodules.

Authors’ contributions

FZ participated the design of this study, data analysis, model validation, and was a major contributor in writing the manuscript. FM was in charge of date collection, cytological result re-evaluation, and was also a major contributor in writing the manuscript. WC was the designer of this study, and reviewed and edited the manuscript. All authors read and approved the final manuscript.

Funding

Proof of Concept Program of Zhongguancun Science City and Peking University Third Hospital (HDCXZHKC2022210).

Data availability

The datasets analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Approval for this study was obtained from Peking University Third Hospital Medical Science Research Ethics Committee. The registration number is IRB00006761- M20250374.

Consent for publication

Informed consent was granted by Peking University Third Hospital Medical Science Research Ethics Committee due to the retrospective nature of the data collection and analysis.

Competing interests

The authors declare no competing interests.