Additive predictive value of preoperative thyroid ultrasound parameters for postoperative hypothyroidism after hemithyroidectomy: a single-center retrospective study

Article information

Ultrasonography. 2026;45(2):129-140

Publication date (electronic) : 2026 January 5

doi : https://doi.org/10.14366/usg.25165

Yeseul Kang¹, Seok-Jae Heo², Kyung Hee Ko¹, Juyeon Lee¹, Chorok Lee³, Eun-Kyung Kim¹, Si Eun Lee¹

¹Department of Radiology, Yongin Severance Hospital, Yonsei University College of Medicine, Yongin, Korea

²Biostatistics Collaboration Unit, Department of Biomedical Systems Informatics, Yonsei University College of Medicine, Seoul, Korea

³Department of Surgery, Yongin Severance Hospital, Yonsei University College of Medicine, Yongin, Korea

Correspondence to: Si Eun Lee, MD, PhD, Department of Radiology, Yongin Severance Hospital, Yonsei University College of Medicine, 363 Dongbaekjukjeon-daero, Giheung-gu, Yongin 16995, Korea Tel. +82-31-5189-8321 Fax. +82-2-2227-8337 E-mail: SIEUN@yuhs.ac

Received 2025 August 21; Revised 2025 December 20; Accepted 2025 December 26.

Abstract

Graphical abstract

Introduction

Hemithyroidectomy is a commonly used surgical procedure for patients with unilateral benign thyroid disease and low-risk differentiated thyroid carcinoma (DTC) [1–3]. The risk of complications such as recurrent laryngeal nerve injury, hypocalcemia, and hemorrhage is significantly lower with hemithyroidectomy than with total thyroidectomy [4]. Consequently, hemithyroidectomy has been the primary treatment option for patients with DTC and tumors ≤2 cm without evidence of extrathyroidal extension or lymph node metastasis. In recent years, its indications have expanded to include tumors between 2 and 4 cm, based on consideration of recurrence risk and surgical morbidity [1–3]. As a result, the use of hemithyroidectomy continues to increase relative to total thyroidectomy [5].

Although hemithyroidectomy is generally expected to maintain euthyroid status without the need for thyroid hormone supplementation, postoperative hypothyroidism remains a common complication. Previous studies have reported its incidence to range from 9.5% to 64.2%, with overt hypothyroidism observed in approximately 4% [6–8]. Identified risk factors include older age, elevated preoperative thyroid-stimulating hormone (TSH) levels, female sex, right-sided lobectomy, positivity for anti-microsomal (thyroid peroxidase [TPO]) antibodies, and histologic evidence of Hashimoto thyroiditis [6,7].

Despite the routine use of preoperative imaging studies such as thyroid ultrasound (US) and neck computed tomography (CT), radiologic features that may help predict postoperative hypothyroidism remain under-researched. Several studies have reported that a smaller body surface area (BSA)–adjusted remnant thyroid volume is independently associated with postoperative hypothyroidism [9,10]. In addition, diffuse thyroid diseases such as Hashimoto thyroiditis demonstrate characteristic US features, including decreased parenchymal echogenicity, coarse echotexture, hypoechoic micronodules, and altered vascularity. Given that histopathologic lymphocytic infiltration has been associated with an increased risk of postoperative hypothyroidism, these US features may also serve as predictive markers. Furthermore, previous studies have demonstrated that ultrasonographic abnormalities in the thyroid parenchyma can precede measurable abnormalities in thyroid function tests or autoantibody levels in asymptomatic individuals [11,12]. Therefore, it is necessary to investigate the association between abnormal US features observed during routine preoperative imaging and the development of postoperative hypothyroidism. In this study, radiologic parameters, including remnant thyroid volume and sonographic features suggestive of diffuse thyroid disease, were analyzed to identify their association with the development of postoperative hypothyroidism.

Therefore, this study aimed to examine the association between preoperative thyroid US features and the risk of developing hypothyroidism after hemithyroidectomy.

Materials and Methods

Radiological Parameters

All imaging data were assessed with a picture archiving and communication system (ZeTTA PACS Viewer, Taeyoung Soft, Anyang, Korea). Preoperative neck CT scans were used to measure the anteroposterior diameter of the thyroid lobe on the axial plane. Remnant thyroid volume was calculated using the ellipsoid formula: length×width×depth×π/6. Width and depth were measured on the axial plane, and length was measured on the sagittal plane. To account for the established correlation between thyroid volume and BSA, remnant lobe volume was standardized by dividing it by the patient’s BSA [15]. BSA was estimated using the Mosteller formula: BSA (m²)=√[height (cm)×weight (kg)/3600].

At the authors’ hospital, preoperative thyroid US examinations were performed using three different US machines: Aplio i800 (Canon Medical Systems, Ōtawara, Japan), LOGIQ E10 (GE HealthCare, Chicago, IL, USA), and EPIQ 7G (Philips, Bothell, WA, USA), with 4–18 Hz, 4–20 Hz, or 6–15 Hz linear probes. For the assessment of US features suggestive of diffuse thyroid disease, two board-certified radiologists (K.H.K. and J.L., with 23 and 3 years of experience in breast and thyroid imaging, respectively) independently reviewed the preoperative thyroid US images while blinded to clinical data. They qualitatively evaluated the echogenicity, echotexture, and vascularity of the thyroid parenchyma (Fig. 1). Echogenicity was scored as (0) normal, defined as hyperechoic relative to the strap muscles, or (1) hypoechogenicity, defined as isoechoic or hypoechoic relative to the strap muscles. Echotexture was assessed as (0) normal, defined as a smooth and uniform parenchymal echo pattern; (1) mildly coarse, defined as a fine, diffuse granular texture resembling a sandy pattern, which may be accompanied by thin, scattered fibrous septa; or (2) markedly coarse, defined by the presence of distinct hypoechoic micronodules creating a cobblestone or micronodular appearance and/or thick, coarse fibrous septa causing architectural distortion. Vascularity was classified as (0) normal or (1) increased, defined as a diffuse increase in parenchymal flow, often with a turbulent, flame-like appearance. To increase reproducibility in this retrospective analysis, the assessments of echogenicity and vascularity were simplified into two categories. In cases of discrepant evaluations, the final decision was made collaboratively at a consensus meeting. Based on these parameters, the US thyroid parenchymal score was defined as the sum of the qualitative scores for echogenicity, echotexture, and vascularity [12].

Fig. 1.

Assessment of ultrasonographic features suggestive of diffuse thyroid disease.

Results

Baseline Characteristics

Among 460 patients who underwent hemithyroidectomy at Yongin Severance Hospital between March 2020 and June 2023, 62 were excluded (Fig. 2). Consequently, 398 patients were included in the final analysis. Table 1 summarizes the baseline characteristics of the study population. Of the 398 patients, 191 (47.9%) developed postoperative hypothyroidism.

Fig. 2.

Flow diagram of participant selection.

TFT, thyroid function test.

Table 1.

Patient characteristics

Clinical factors significantly associated with postoperative hypothyroidism included thyroid volume, BSA-adjusted remnant thyroid volume, preoperative serum TSH level, postoperative serum TSH level, and the presence of thyroiditis on histopathology. Although the individual US features of the thyroid parenchyma (echotexture, echogenicity, and vascularity) were not significant predictors, the US thyroid parenchymal score showed a statistically significant association with postoperative hypothyroidism. The weighted US thyroid parenchymal score, calculated as follows, also displayed statistical significance: weighted US thyroid parenchymal score=0.48×echogenicity score+0.23×echotexture score+0.24×vascularity score. The distributions of the US thyroid parenchymal score and the weighted US thyroid parenchymal score are illustrated in Fig. 3. Representative clinical cases of patients with and without postoperative hypothyroidism are presented in Fig. 4. Final pathological diagnoses are summarized in Supplementary Table 1.

Fig. 3.

Distribution of unweighted and weighted ultrasound (US) thyroid parenchymal scores by postoperative hypothyroidism status.

A. Violin/box plot shows the distribution of US thyroid parenchymal scores (unweighted) according to the development of postoperative hypothyroidism. B. Violin/box plot shows the distribution of weighted US thyroid parenchymal scores according to the development of postoperative hypothyroidism.

Fig. 4.

Representative clinical cases of patients with and without postoperative hypothyroidism.

Preoperative thyroid ultrasound (US) images show a patient who maintained euthyroidism postoperatively and another who developed postoperative hypothyroidism. A, B. A 50-year-old man maintained euthyroidism postoperatively. His preoperative data were as follows: thyroid-stimulating hormone (TSH), 1.43 mIU/L; body surface area (BSA)-adjusted thyroid volume, 6.09 cm³/m²; US thyroid parenchymal score, 1 (echotexture, 1; echogenicity, 0; vascularity, 0); and weighted US thyroid parenchymal score, 0.23. C, D. A 47-year-old woman developed postoperative hypothyroidism. Her preoperative data were as follows: TSH, 1.64 mIU/L; BSA-adjusted thyroid volume, 10.79 cm³/m²; US thyroid parenchymal score, 4 (echotexture, 2; echogenicity, 1; vascularity, 1); and weighted US thyroid parenchymal score, 1.17.

All evaluated preoperative US features and the US thyroid parenchymal score were significantly associated with histopathologic evidence of thyroiditis (all P<0.001) (Supplementary Table 2). Interobserver agreement for US features of the thyroid parenchyma, assessed using unweighted and weighted kappa values, was as follows: echogenicity, 0.31; echotexture, 0.43; and vascularity, 0.49, indicating fair to moderate agreement.

Significant Clinical and Thyroid US Parameters for Predicting Postoperative Hypothyroidism

In univariable logistic regression analysis, preoperative TSH (odds ratio [OR], 2.18; 95% confidence interval [CI], 1.71 to 2.78), postoperative TSH (OR, 1.37; 95% CI, 1.18 to 1.61), BSA-adjusted remnant thyroid volume (OR, 0.85; 95% CI, 0.76 to 0.95), presence of Hashimoto thyroiditis on histopathology (OR, 2.47; 95% CI, 1.56 to 3.92), US thyroid parenchymal score (OR, 1.33; 95% CI, 1.05 to 1.68), and weighted US thyroid parenchymal score (OR, 2.72; 95% CI, 1.22 to 6.28) were identified as significant predictors of postoperative hypothyroidism (Table 2).

Table 2.

Univariable and multivariable logistic regression analysis for postoperative hypothyroidism

To account for multicollinearity and clinical relevance, multivariable logistic regression analysis was conducted using BSA-adjusted remnant thyroid volume, preoperative TSH level, and either the US thyroid parenchymal score or the weighted US thyroid parenchymal score. Postoperative TSH and histopathologic evidence of thyroiditis were excluded because these parameters are not available when developing a preoperative predictive model. To assess the additive value of US parameters for predicting postoperative hypothyroidism, three models were constructed: model A included preoperative TSH and BSA-adjusted remnant thyroid volume; model B included preoperative TSH, BSA-adjusted remnant thyroid volume, and the US thyroid parenchymal score; and model C included preoperative TSH, BSA-adjusted remnant thyroid volume, and the weighted US thyroid parenchymal score. None of the variables included in each model had a variance inflation factor greater than 10.

Across all models, preoperative TSH (OR, 2.12 [95% CI, 1.66 to 2.72]; OR, 2.02 [95% CI, 1.62 to 2.66]; and OR, 2.08 [95% CI, 1.63 to 2.66], respectively) and BSA-adjusted remnant thyroid volume (OR, 0.88 [95% CI, 0.78 to 0.99]; OR, 0.85 [95% CI, 0.75 to 0.97]; and OR, 0.85 [95% CI, 0.75 to 0.96], respectively) were significant predictors of postoperative hypothyroidism. The US thyroid parenchymal score (OR, 1.36; 95% CI, 1.04 to 1.78) and the weighted US thyroid parenchymal score (OR, 3.05; 95% CI, 1.22 to 7.62) were significant predictors in models B and C, respectively (Table 2).

Assessment of Prediction Model Performance with US Findings Included

The AUC (95% CI), sensitivity, and specificity of each model were as follows: model A, 0.71 (0.66 to 0.77), 0.58, and 0.77, respectively; model B, 0.72 (0.68 to 0.77), 0.60, and 0.73, respectively; and model C, 0.72 (0.67 to 0.77), 0.60, and 0.76, respectively (Table 3). The diagnostic performance of each model is illustrated by its ROC curve (Fig. 5).

Table 3.

AUCs and other metrics of each model for the prediction of postoperative hypothyroidism

Fig. 5.

Receiver operating characteristic (ROC) curve of each model.

Model A, preoperative thyroid-stimulating hormone+body surface area-adjusted remnant thyroid volume; model B, model A+ultrasound (US) thyroid parenchymal score; model C, model A+weighted US thyroid parenchymal score. AUC, area under the ROC curve.

In pairwise comparisons using the DeLong method, differences in AUC were not statistically significant between models A and B (P=0.35) or between models A and C (P=0.20). However, adding the US thyroid parenchymal score to model A (to create model B) yielded a borderline significant cNRI of 0.20 (95% CI, 0 to 0.39; P=0.05) and a significant IDI of 0.01 (95% CI, 0.00 to 0.02; P=0.04). Adding the weighted score to model A (to create model C) resulted in a significant IDI of 0.01 (95% CI, 0.00 to 0.02; P=0.03), although the improvement in cNRI was not statistically significant (0.13; 95% CI, −0.06 to 0.32; P=0.18) (Table 4).

Table 4.

Predictive performance of each model

Discussion

To the best of current knowledge, this is the first study to examine the association between thyroid parenchymal abnormalities on preoperative US and the risk of hypothyroidism after hemithyroidectomy. This study demonstrated the potential value of thyroid parenchymal features on preoperative US for predicting postoperative hypothyroidism. Although adding the US thyroid parenchymal score and the weighted US thyroid parenchymal score did not significantly increase the model’s overall discrimination, it significantly improved the IDI (P=0.04 and P=0.03, respectively). This finding is clinically relevant, as it suggests that improved risk stratification can be achieved using only routine preoperative examinations, minimizing the need for additional medical resources.

The central hypothesis was that preoperative US features of the thyroid parenchyma can serve as predictive markers for postoperative hypothyroidism. All evaluated preoperative US features, as well as the US thyroid parenchymal score and the weighted US thyroid parenchymal score, were significantly associated with histopathologic evidence of thyroiditis, consistent with previous studies on US findings of diffuse thyroid disease and supporting this hypothesis [11,12,18]. Although individual US features (echotexture, echogenicity, and vascularity) were not significant predictors when assessed separately, their combined assessment provided meaningful predictive value. The combined score was a statistically significant predictor in the multivariable analysis. Both the unweighted score (OR, 1.36; 95% CI, 1.04 to 1.78) and the weighted score (OR, 3.05; 95% CI, 1.22 to 7.62) remained significant after adjustment for strong predictors such as preoperative TSH and BSA-adjusted remnant thyroid volume.

In this study, the AUCs (95% CI) were 0.71 (0.66 to 0.77) for model A, 0.70 (0.65 to 0.75) for preoperative TSH alone, and 0.60 (0.54 to 0.65) for BSA-adjusted remnant thyroid volume alone. Lang et al. [9] reported AUCs of 0.70 (0.60 to 0.80) for preoperative TSH, 0.74 (0.66 to 0.82) for BSA-adjusted remnant thyroid volume, and 0.80 (0.72 to 0.87) for a model including both parameters, comparable to model A in this study. Similarly, Prevot et al. [10] reported AUCs of 0.76 (0.67 to 0.86) for preoperative TSH and 0.74 (0.64 to 0.84) for BSA-adjusted remnant thyroid volume. Cao et al. [19] also reported an AUC of 0.68 (0.62 to 0.74) for preoperative TSH. Although the predictive ability of preoperative TSH in this study was comparable to that reported previously, the AUC for BSA-adjusted remnant thyroid volume was notably lower. This discrepancy may be attributable to several factors related to the study methodology and patient cohort. First, thyroid volume was measured on preoperative neck CT, which may yield higher volume measurements than US-based assessments. Second, whereas Lang et al. [9] excluded patients with thyroid nodules ≥1 cm, this cohort included such patients, which may have influenced the association between remnant volume and postoperative hypothyroidism. Third, differences in patient demographics and clinical characteristics, including a higher prevalence of thyroiditis, may have affected the discriminative ability of BSA-adjusted remnant thyroid volume.

Multivariable models combining preoperative TSH, BSA-adjusted remnant thyroid volume, and either the US thyroid parenchymal score (model B) or the weighted US thyroid parenchymal score (model C) were constructed. The AUCs of models B and C were 0.72 (95% CI, 0.68 to 0.77) and 0.72 (95% CI, 0.67 to 0.77), respectively, which are lower than those described in previous studies. For example, Lang et al. [9] reported an AUC of 0.83 (95% CI, 0.76 to 0.90) for a multivariable model that included preoperative TSH, BSA-adjusted remnant thyroid volume, and the number of nodules in the ipsilateral lobe. Similarly, Prevot et al. [10] reported an AUC of 0.82 (95% CI, 0.75 to 0.90) for a model combining preoperative TSH, BSA-adjusted remnant thyroid volume, and age. This difference may stem from several factors, including the lower predictive value of BSA-adjusted remnant thyroid volume in this study, variations in model composition, and differences in variable handling. Prevot et al. [10] converted continuous variables, such as preoperative TSH, into binary categories by calculating an optimal cutoff value, whereas these variables were treated as continuous in this study.

Unfortunately, despite the high OR and statistical significance of the weighted US thyroid parenchymal score, its inclusion did not meaningfully improve the model’s overall discrimination as measured by AUC (P=0.20). This apparent paradox likely reflects the low variability and limited granularity of the score, which comprised only 10 unique values across the cohort. This constrained distribution may have limited its capacity to provide additional discriminative power beyond the already strong predictive capacity of preoperative TSH. Notably, a subgroup of patients (18.6%) remained poorly predicted across models incorporating preoperative TSH alone, preoperative TSH with BSA-adjusted remnant thyroid volume, and additional inclusion of the weighted US thyroid parenchymal score. This finding suggests that the risk of postoperative hypothyroidism in these individuals may be influenced by unmeasured factors beyond currently available laboratory, volumetric, and parenchymal imaging parameters.

The prevalence of postoperative hypothyroidism in the present study (47.9%) was higher than the weighted pooled incidence (29%; 95% CI, 25% to 34%; P<0.001) reported in the meta-analysis by Cooper et al. [7]. However, that meta-analysis exhibited high heterogeneity (I² >75%), indicating substantial variation in incidence across studies and limiting the generalizability of the pooled estimate. In addition, this discrepancy may be partly attributable to differences in the definition of postoperative hypothyroidism. Cox et al. [20] reported that 350 of 478 patients (73.2%) had TSH levels >2 mIU/L within 1 year after thyroid lobectomy. Kim et al. [8] demonstrated that 59.0% and 66.0% of patients required levothyroxine supplementation at 2 and 5 years postoperatively, respectively. Therefore, the relatively high prevalence in the present study may reflect the definition of postoperative hypothyroidism as the continued need for thyroxine supplementation 2 years after surgery.

The proportion of Hashimoto thyroiditis on postoperative histopathology in the present study (27.6%) was higher than that reported in China (12%) but lower than that reported in a Korean study (34%) [8–10]. In this study, the mean BSA-adjusted remnant thyroid volume (5.00 mL) was higher than that reported in previous studies from China (3.7 mL) and Korea (3.1 mL) [9,18]. This difference may be related to variation in measurement methods, as US tends to yield smaller thyroid volume estimates than CT [21]. In addition, this study did not exclude patients with nodules in the remnant thyroid lobe, which may have introduced measurement inaccuracy in remnant thyroid volume. Nevertheless, thyroid volume displayed a significant association with the development of postoperative hypothyroidism, consistent with previous research [9,10,18].

Preoperative TSH, a well-established risk factor for postoperative hypothyroidism, was identified as significant predictor, consistent with previous research [6,7,9,10,14,22]. Histopathologic evidence of thyroiditis was also statistically significant. Several previous studies have indicated that histopathologic evidence of thyroiditis is an independent prognostic factor for postoperative hypothyroidism [6,7,23]. However, other studies have reported no significant association between lymphocytic thyroiditis and postoperative hypothyroidism [9,10,14,22]. The statistical significance of histopathologic evidence of thyroiditis in this study may relate to the relatively young age of the cohort (median, 41 years; interquartile range, 35 to 49 years). Piper et al. [24] reported that the association between lymphocytic infiltration and postoperative hypothyroidism is age dependent. Age and sex, which have shown inconsistent associations with postoperative hypothyroidism in previous research, were not associated with postoperative hypothyroidism in this study [6,7].

This study had several limitations. First, due to its single-center retrospective design, the findings have inherent limitations in generalizability. Because US is a real-time, operator-dependent examination, retrospective assessment of the thyroid parenchyma using images acquired by different operators across multiple US systems poses challenges for reproducibility. To mitigate this issue, the parenchymal classification scheme was simplified, and two board-certified radiologists specializing in thyroid imaging independently reviewed all images, with discrepancies resolved by consensus. Nevertheless, prospective studies using a standardized imaging protocol are needed to validate the proposed US parenchymal scoring system. Second, several potential confounding variables were not evaluated. Only clinical and radiologic parameters that were retrospectively obtainable from standard preoperative examinations were analyzed. As a result, known clinical predictors such as anti-microsomal (TPO) antibody status, as well as ultrasonographic features including glandular margin irregularity, scattered microcalcifications, and reactive lymphadenopathy, were not assessed. Third, postoperative hypothyroidism was defined as the continued requirement for thyroxine supplementation 2 years after surgery because thyroxine is routinely initiated immediately after surgery at this institution. This definition may have influenced the results. In addition, factors such as pregnancy or patient preference may also have affected the decision to continue levothyroxine therapy, which were beyond the control of this study.

In conclusion, this study highlights the potential utility of ultrasonographic parenchymal features for refining risk stratification for postoperative hypothyroidism. The US thyroid parenchymal score and the weighted US thyroid parenchymal score, which are readily obtainable from routine preoperative US, were statistically significant predictors independent of established risk factors such as preoperative TSH and BSA-adjusted remnant thyroid volume. Although inclusion of these scores did not significantly improve the AUC, their independent association with postoperative hypothyroidism underscores their clinical relevance. Prospective external validation with larger, multicenter cohorts is needed to confirm the generalizability and robustness of these findings.

Supplementary Material

Article information Continued

Notes