Questions regarding predictor directionality and outcome definition in ultrasound-based prediction of postoperative hypothyroidism

I read with great interest the recent study by Kang et al. [1], which examined the association between preoperative thyroid ultrasonographic parenchymal features and postoperative hypothyroidism after hemithyroidectomy. The topic is clinically important, and the authors should be commended for developing a composite ultrasound (US) parenchymal score based on routine preoperative imaging. However, several methodological and interpretative aspects require further clarification.

First, there appears to be a directional inconsistency between Table 1 and the regression analyses in Table 2. In Table 1, both the unweighted and weighted US thyroid parenchymal scores are higher in the euthyroid group than in the hypothyroid group. A higher score in the euthyroid group implies that, in univariable analysis, greater parenchymal abnormality is associated with a lower risk of hypothyroidism; however, the authors report odds ratios (ORs) greater than 1, indicating the opposite relationship. The same contradiction applies to preoperative thyroid-stimulating hormone (TSH) level, which was lower in the hypothyroid group than in the euthyroid group (1.57 vs. 2.33 mIU/mL, P<0.001), despite an OR of 2.18 for predicting hypothyroidism. It also applies to body surface area–adjusted remnant thyroid volume, which was higher in the hypothyroid group than in the euthyroid group (5.26 vs. 4.72 cm³/m², P<0.001), despite an OR of 0.85, implying that larger volume is protective. These findings appear inconsistent both internally and with prior literature [2–5]. Clarification of whether this reflects a labeling issue, model specification, or another factor would be helpful, as such inconsistencies may substantially affect interpretation of the study’s principal conclusions, particularly whether the identified imaging and clinical parameters truly increase or decrease the risk of postoperative hypothyroidism. They may also complicate the clinical interpretation and reproducibility of the reported predictors.

Second, defining postoperative hypothyroidism as continued levothyroxine use at 2 years may conflate two distinct entities. Post-lobectomy hypothyroidism is conventionally defined biochemically, based on a serum TSH level above the reference range, rather than by medication use [6–8]. Because levothyroxine was initiated in all patients after surgery, including those with differentiated thyroid cancer who required TSH suppression, “continued use” may capture both true thyroid insufficiency and iatrogenic suppression for oncologic indications. Given the high prevalence of malignancy in Supplementary Table 1, this definition could inflate the apparent incidence and distort the predictor associations. Medication status may also overstate true insufficiency: when mild hypothyroidism is followed without replacement, approximately two-thirds of patients recover spontaneously [8], and roughly half can be weaned from levothyroxine when discontinuation is attempted [6]. A biochemical definition based on TSH thresholds, together with analyses restricted to benign pathology, would therefore more accurately reflect true thyroid insufficiency and could materially change both the reported incidence and the predictor associations on which the study’s conclusions rest.

Third, the derivation and evaluation of the weighted US thyroid parenchymal score require further consideration. The weighting coefficients for echogenicity, echotexture, and vascularity—0.48, 0.23, and 0.24, respectively—appear to have been derived from a multivariable logistic regression model; however, the performance of the resulting score was assessed in the same dataset used for model development. This approach may introduce optimistic bias [9,10], as regression coefficients are inherently optimized for the derivation sample. Consequently, the reported associations and improvements in discrimination may overestimate the score’s true predictive performance. Incorporating internal validation methods, such as cross-validation, or conducting external validation in an independent cohort would strengthen confidence in the model.

Finally, there appears to be an inconsistency regarding the TSH variable included in the multivariable models. Although the methods state that only preoperative TSH was used, Table 2 presents results for postoperative TSH instead, with preoperative TSH omitted. Clarifying which variable was included, and providing corrections if needed, would improve transparency and reproducibility.

In conclusion, this study offers valuable insight into the potential role of US-based parenchymal assessment in predicting postoperative hypothyroidism. Addressing the points above would further strengthen the manuscript and support its clinical applicability.