未定性甲状腺结节（ITN）十分常见，常常导致（有时是不必要的）诊断性手术。我们旨在评估基于常规特征的两种机器学习方法（ML）的绩效，以预测ITN的恶性风险（RM）。该多中心诊断性回顾性队列研究从2010年至2020年进行。纳入至少接受一个Bethesda III-V甲状腺结节（TN）手术并具有完全可用医疗记录的成年患者。在审查了7,917份记录后，符合资格标准的有1,288名患者和1,335个TN。患者分为训练组（940 TN）和验证组（395 TN）。评估多元逻辑回归模型（LR）及其评分表以及随机森林模型（RF）在预测TN的性质和RM方面的诊断绩效。收集并使用患者的所有可用临床，生化，超声和细胞学数据构建两种算法。253（19％），693（52％）和389（29％）TN分别被归类为Bethesda III，IV和V，整体RM为35％。两个组基线特征均平衡。两个模型都在验证组上进行验证，根据LR模型和RF模型的特异性，敏感性，阳性预测值，阴性预测值和受试者工作特征曲线下面积的表现，分别为90％，57.3％，73.4％，81.4％，84％（CI95％：78.5-89.5％）和87.6％，54.7％，68.1％，80％，82.6％（CI95％：77.4-87.9％）。我们的ML模型在预测Bethesda III-V TN的性质方面表现良好。此外，我们的免费在线评分表有助于进一步细化RM，识别出一些低风险TN，这些TN可能从监测中受益，从而可以减少不必要的手术数量。©作者（S）2023。由牛津大学出版社代表（ESE）欧洲内分泌学会发表。

Indeterminate thyroid nodules (ITN) are common and often lead to (sometimes unnecessary) diagnostic surgery. We aimed to evaluate the performance of two machine learning methods (ML), based on routinely available features to predict the risk of malignancy (RM) of ITN.Multicentric diagnostic retrospective cohort study conducted between 2010 and 2020.Adult patients who underwent surgery for at least one Bethesda III-V thyroid nodule (TN) with fully available medical records were included. Of the 7,917 records reviewed, eligibility criteria were met in 1,288 patients with 1,335 TN. Patients were divided in a training (940 TN) and validation cohort (395 TN). The diagnostic performance of a multivariate logistic regression model (LR) and its nomogram, and a random forest model (RF) in predicting the nature and RM of a TN were evaluated. All available clinical, biological, ultrasound, and cytological data of the patients were collected and used to construct the two algorithms.There were 253 (19%), 693 (52%) and 389 (29%) TN classified as Bethesda III, IV and V respectively, with an overall RM of 35%. Both cohorts were well balanced for baseline characteristics. Both models were validated on the validation cohort, with performances in terms of specificity, sensitivity, positive predictive value, negative predictive value and area under the receiver operating characteristic curve of 90%, 57.3%, 73.4%, 81.4%, 84% (CI95%: 78.5-89.5%) for the LR model, and 87.6%, 54.7%, 68.1%, 80%, 82.6% (CI95%: 77.4-87.9%) for the RF model, respectively.Our ML models performed well in predicting the nature of Bethesda III-V TN. In addition, our freely available online nomogram helped to refine the RM, identifying low-risk TN that may benefit from surveillance in up to a third of ITN, and thus may reduce the number of unnecessary surgeries.© The Author(s) 2023. Published by Oxford University Press on behalf of (ESE) European Society of Endocrinology.