Preoperative Fusion Class Identifies Two Clinically Distinct Phenotypes in Thyroid Tumors: A Single-Institution Cohort Study

Structured Abstract Objective Fusion-positive thyroid tumors detected by preoperative molecular testing are biologically heterogeneous. We evaluated whether fusion class was associated with clinicopathologic features, nodal disease, and surgical management among operated patients with an isolated fusion (no confirmed co-driver mutation). Methods Single-institution cohort of 38 adults undergoing initial thyroid surgery (2017–2024) with an isolated preoperative driver fusion. Fusions were classified as RTK/BRAF-like (RET, NTRK; n=12) or RAS-like/follicular-pattern (PAX8::PPARG, THADA::IGF2BP3, CREB3L2::PPARG; n=26). Principal outcomes were cervical nodal metastasis and final histology. Analyses were unadjusted and restricted to invasive malignancies (n=32); a prespecified sensitivity analysis restricted the nodal endpoint to patients with nodes examined. Results RTK/BRAF-like tumors were uniformly papillary carcinoma (12/12) and smaller than RAS-like/follicular-pattern tumors (median 1.8 vs 4.1 cm) but more often node-positive (6/12 vs 0/20; risk difference 0.50, 95% CI 0.21–0.75), with more extrathyroidal extension, lymphatic invasion, and lateral neck dissection. Nodes were examined in 8/12 versus 8/20 (p=0.27); among these, node-positive disease was present in 6/8 versus 0/8 (risk difference 0.75, 95% CI 0.28–0.93). RAS-like/follicular-pattern tumors were heterogeneous, with predominant capsular invasion (14/20 vs 1/12), and included all six noninvasive or benign lesions. Thyroidectomy extent did not differ. Conclusion In this exploratory cohort, fusion class identified two distinct phenotypes: RTK/BRAF-like fusions with a higher observed frequency of nodal metastasis, and RAS-like/follicular-pattern fusions with larger tumors and a low observed nodal burden, although nodes were not examined in most. Fusion class may complement established preoperative risk assessment, but these hypothesis-generating findings require multi-institutional validation.