SNOMED CT is a comprehensive and evolving clinical reference terminology that has been widely adopted as a common vocabulary to promote interoperability between Electronic Health Records. Owing to its importance in healthcare, quality assurance becomes an integral part of the lifecycle of SNOMED CT. While, manual auditing of every concept in SNOMED CT is difficult and labor intensive, identifying inconsistencies in the modeling of concepts without any context can be challenging. Algorithmic techniques are needed to identify modeling inconsistencies, if any, in SNOMED CT. This study proposes a context-based, machine learning quality assurance technique to identify concepts in SNOMED CT that may be in need of auditing. The Clinical Finding and the Procedure hierarchies are used as a testbed to check the efficacy of the method. Results of auditing show that the method identified inconsistencies in 72% of the concept pairs that were deemed inconsistent by the algorithm. The method is shown to be effective in both maximizing the yield of correction, as well as providing a context to identify the inconsistencies. Such methods, along with SNOMED International's own efforts, can greatly help reduce inconsistencies in SNOMED CT.

中文翻译：

---

使用机器学习技术检测 SNOMED CT 中的建模不一致

SNOMED CT 是一个全面且不断发展的临床参考术语，已被广泛采用作为促进电子健康记录之间互操作性的通用词汇。由于其在医疗保健中的重要性，质量保证成为 SNOMED CT 生命周期不可或缺的一部分。虽然手动审核 SNOMED CT 中的每个概念既困难又费力，但在没有任何上下文的情况下识别概念建模中的不一致可能具有挑战性。需要算法技术来识别 SNOMED CT 中的建模不一致（如果有）。本研究提出了一种基于上下文的机器学习质量保证技术，以识别 SNOMED CT 中可能需要审核的概念。临床发现和程序层次结构用作检查方法有效性的测试平台。审计结果表明，该方法在 72% 的概念对中发现了不一致，这些概念对被算法视为不一致。该方法被证明在最大化校正产量以及提供上下文以识别不一致方面都是有效的。此类方法加上 SNOMED International 自己的努力，可以极大地帮助减少 SNOMED CT 中的不一致性。