Early postoperative voice-change phenotypes after thyroid surgery: a prospective cohort study.

Background: Post-thyroidectomy voice change is common, but early postoperative objective voice-change patterns and their short-term clinical relevance remain insufficiently characterized. Methods: In this single-center prospective cohort study, patients undergoing thyroid surgery were screened consecutively from October 2025 to November 2025. Standardized voice recordings were obtained at baseline and postoperative day 2 (POD2), and clinical voice outcomes were assessed at postoperative day 7 (POD7). Five POD2 spectral change features relative to baseline were used to derive phenotypes through a prespecified two-stage unsupervised clustering workflow based on partitioning around medoids, Manhattan distance, and silhouette-based cluster-number selection. Clinical validation used change in Voice Handicap Index-30 (VHI-30), the prespecified VHI-30 responder definition (Delta VHI-30 ≥ 13), and change in Grade within the Grade, Roughness, Breathiness, Asthenia, Strain scale. Robustness was evaluated using consensus clustering, proportion of ambiguous clustering, sample-level stability, and resampling-based adjusted Rand index reproducibility. Sensitivity analyses compared the primary workflow with one-step k = 3 and k = 4 clustering alternatives, and exploratory analyses evaluated operative and demographic factors in relation to POD7 outcomes. Results: Of 401 screened patients, 245 were included in the final analytic cohort. The final phenotypes were A/B/C = 59/56/130. All five spectral clustering features differed significantly across phenotypes (all P < 0.001). At POD7, phenotypes differed significantly in VHI-30 change (epsilon-squared = 0.641), VHI-30 responder rate (Cramér's V = 0.639), and Grade change (epsilon-squared = 0.203) (all P < 0.001). Phenotype B showed the greatest short-term burden, including a responder rate of 53.6% (30/56; exact 95% confidence interval, 39.7%-67.0%), compared with 1.7% (1/59; exact 95% confidence interval, 0.0%-9.1%) in phenotype A and 1.5% (2/130; exact 95% confidence interval, 0.2%-5.4%) in phenotype C. Sensitivity analyses did not support replacing the prespecified two-stage workflow with one-step k = 3 or k = 4 alternatives. Exploratory analyses suggested that the clinical profile of phenotype B was not simply explained by thyroidectomy extent, surgical approach, lateral neck dissection, or operative time. Conclusions: Early postoperative objective voice changes after thyroid surgery can be organized into clinically interpretable short-term phenotypes associated with patient-reported and supportive perceptual outcomes. These phenotypes should be interpreted as early postoperative voice-change patterns rather than as nerve-injury phenotypes or persistent long-term voice categories.