e18072 Background: PTC is a common endocrine cancer with a good prognosis, but aggressive subtypes, such as Hürthle cell (HCC) and columnar cell variants (CCV), pose challenges due to their higher recurrence and metastasis rates. To provide personalized care, we applied machine learning to evaluate the treatment effectiveness and develop precise prognostic models for PTC variants. Methods: The Surveillance, Epidemiology, and End Results (SEER) database provided the data used for this study’s analysis (2000–2019). Patients who met any of the following criteria were excluded: diagnosis not confirmed by histology, previous history of cancer or with other concurrent malignancies, and unknown data. To identify prognostic variables, we conducted Cox regression analysis and constructed prognostic models using machine learning (ML) algorithms to predict the 5-year survival. Patient records were randomly divided into training (70 %) and validation (30 %) sets. A validation method incorporating the area under the curve (AUC) of the receiver operating characteristic curve was used to validate the accuracy and reliability of the ML models. Results: The study population comprised 3690 patients. Among them 3180 patients with CCV and 510 patients with HCC, respectively. Most patients (62.8%) were 45 years or older, with a median age of 52 years. A total of 56.9% of patients had a tumor size greater than 2 cm, with a median tumor size of 3.1 cm. The largest racial group was white, comprising 83.8% of the cases, and 11.8% of the cases were Asian. Most cases were regional (53.4%, n=1969), followed by localized (38.3%, n= 1413). Multivariate Cox regression analysis revealed that N1 negatively affected the survival of HCC patients. CCV has a favorable prognosis after surgery, radiotherapy, or total thyroidectomy. Poor prognosis in CCV is associated with black race, large tumor size, and T4 stage. Improved survival in the localized/regional stage and decreased survival with male sex, older age, distant metastasis, and advanced AJCC stage in both PTC subtypes. ML models revealed that the random forest classifier (RFC) and K-Nearest Neighbors (KNN) accurately predicted outcomes, followed by Logistic Regression (LR) models. The highest contributing factors were AJCC staging, tumor size, and T aspect of TNM staging. Conclusions: Our study offers a method for evaluating and treating patients with PTC variants. The machine learning model that we created serves as a useful and personalized resource to aid in clinical decision-making processes.[Table: see text]