A CONTINUOUS MULTIVARIABLE MODEL COMBINING SERUM TSH AND ACR TI-RADS FOR PRE-BIOPSY RISK STRATIFICATION OF THYROID NODULES: DEVELOPMENT AND INTERNAL VALIDATION

Objective: To determine the incremental diagnostic value of serum thyroid-stimulating hormone (TSH) beyond ACR TI-RADS for pre-biopsy risk stratification of thyroid nodules, comparing binary integration strategies with a continuous multivariable risk model. Methods: This retrospective study included 172 adults who underwent thyroidectomy (2021–2024) at two Brazilian hospitals. The optimal TSH cutoff was determined by ROC analysis. A continuous multivariable model (TSH, ACR TI-RADS category, age, sex, and nodule size on ultrasound) was internally validated by bootstrapping and compared, using decision curve analysis, with the ACR TI-RADS FNA criteria alone, TSH alone, and two binary TSH–ACR TI-RADS strategies. Results: Malignancy prevalence was 47.1%. Median TSH was higher in patients with malignant nodules (1.74 vs. 1.31 mIU/L; p<0.01). TSH alone yielded an AUC of 0.652 (95% CI 0.571–0.734; p<0.01), with an optimal cutoff of 1.545 mIU/L. TSH was an independent predictor of malignancy (OR=1.67 per mIU/L; 95% CI 1.13–2.47; p=0.01). The binary strategies increased sensitivity from 71.6% to 84.0–90.1%, at the cost of lower specificity. The continuous model showed the highest optimism-corrected net benefit at threshold probabilities above approximately 10% (corrected AUC 0.829; calibration slope 0.83) Conclusions: In this surgical cohort, TSH provided incremental, low-cost diagnostic value beyond ACR TI-RADS, particularly when incorporated into a continuous multivariable model. External validation is warranted before clinical use.