Predicting biochemical hypothyroidism after thyroid lobectomy using ultrasound elastography.

BACKGROUND: Predicting hypothyroidism after thyroid lobectomy is important for treatment selection, patient counseling, and postoperative management. Although thyroiditis is a recognized risk factor, the predictive value of ultrasound strain elastography remains unclear. METHODS: We conducted a retrospective cohort study based on prospectively collected elastography data from consecutive adults undergoing thyroid lobectomy between January 2023 and December 2024. Biochemical hypothyroidism was defined as an elevated thyroid-stimulating hormone level at the final assessment within 1 year after surgery or before initiation of levothyroxine therapy. Multivariable logistic regression, restricted cubic spline analysis, and 3 machine learning models were used to identify predictors of postoperative hypothyroidism. RESULTS: Among the 269 patients included, 86 (32%) developed biochemical hypothyroidism. In multivariable logistic regression, preoperative thyroid-stimulating hormone level and thyroid parenchymal strain ratio were independent predictors of postoperative hypothyroidism (both P < .001). Restricted cubic spline analysis demonstrated a positive nonlinear association between strain ratio and predicted risk without an apparent threshold. Across all machine learning models, preoperative thyroid-stimulating hormone level and strain ratio consistently ranked as the 2 most influential predictors based on SHapley Additive exPlanations. The elastic net model achieved the highest discrimination (area under the receiver operating characteristic curve, 0.800), whereas the random forest model demonstrated the best calibration. CONCLUSION: Thyroid parenchymal strain ratio measured by ultrasound elastography and preoperative thyroid-stimulating hormone level are independent predictors of biochemical hypothyroidism after thyroid lobectomy. Incorporating quantitative assessment of thyroid stiffness into preoperative risk stratification may provide incremental predictive value beyond conventional clinicopathologic factors and facilitate individualized surgical counseling, treatment selection, and postoperative surveillance.