Amsterdam UMC, University of Amsterdam, Department of Medical Informatics, Amsterdam Public Health Research Institute, Meibergdreef 9, Amsterdam, Netherlands; Department of Research & Development, ChipSoft B.V., Amsterdam, Netherlands.
SNOMED CT Netherlands National Release Center, Nictiz, The Hague, Netherlands.
J Biomed Inform. 2022 May;129:104071. doi: 10.1016/j.jbi.2022.104071. Epub 2022 Apr 13.
Now that patients increasingly get access to their healthcare records, its contents require clarification. The use of patient-friendly terms and definitions can help patients and their significant others understand their medical data. However, it is costly to make patient-friendly descriptions for the myriad of terms used in the medical domain. Furthermore, a description in more general terms, leaving out some of the details, might already be sufficient for a layperson. We developed an algorithm that employs the SNOMED CT hierarchy to generalize diagnoses to a limited set of concepts with patient-friendly terms for this purpose. However, generalization essentially implies loss of detail and might result in errors, hence these generalizations remain to be validated by clinicians. We aim to assess the medical validity of diagnosis clarification by generalization to concepts with patient-friendly terms and definitions in SNOMED CT. Furthermore, we aim to identify the characteristics that render clarifications invalid.
Two raters identified errors in 12.7% (95% confidence interval - CI: 10.7-14.6%) of a random sample of 1,131 clarifications and they considered 14.3% (CI: 12.3-16.4%) of clarifications to be unacceptable to show to a patient. The intraclass correlation coefficient of the interrater reliability was 0.34 for correctness and 0.43 for acceptability. Errors were mostly related to the patient-friendly terms and definitions used in the clarifications themselves, but also to terminology mappings, terminology modelling, and the clarification algorithm. Clarifications considered to be most unacceptable were those that provide wrong information and might cause unnecessary worry.
We have identified problems in generalizing diagnoses to concepts with patient-friendly terms. Diagnosis generalization can be used to create a large amount of correct and acceptable clarifications, reusing patient-friendly terms and definitions across many medical concepts. However, the correctness and acceptability have a strong dependency on terminology mappings and modelling quality, as well as the quality of the terms and definitions themselves. Therefore, validation and quality improvement are required to prevent incorrect and unacceptable clarifications, before using the generalizations in practice.
现在，患者越来越多地获取自己的医疗记录，因此需要对其内容进行说明。使用通俗易懂的术语和定义可以帮助患者及其家属理解他们的医疗数据。但是，为医学领域中使用的众多术语制作通俗易懂的描述是非常昂贵的。此外，对于非专业人士来说，使用更一般的术语并省略一些细节可能已经足够了。我们开发了一种算法，该算法使用 SNOMED CT 层次结构将诊断概括为具有通俗易懂术语的有限数量的概念，以达到此目的。但是，概括本质上意味着损失细节，并且可能导致错误，因此这些概括仍然需要临床医生进行验证。我们旨在评估通过概括到具有通俗易懂术语和 SNOMED CT 定义的概念来阐明诊断的医学有效性。此外，我们旨在确定导致澄清无效的特征。
两名评估员在随机抽样的 1,131 条澄清中发现了 12.7%（95%置信区间 - CI：10.7-14.6%）的错误，他们认为 14.3%（CI：12.3-16.4%）的澄清不能向患者展示。正确性和可接受性的两位评估员间的组内相关系数分别为 0.34 和 0.43。错误主要与澄清本身中使用的通俗易懂术语和定义有关，但也与术语映射、术语建模和澄清算法有关。被认为最不可接受的澄清是那些提供错误信息并可能引起不必要的担忧的澄清。
我们已经确定了将诊断概括为通俗易懂术语的概念中存在的问题。诊断概括可用于创建大量正确且可接受的澄清，在许多医疗概念中重复使用通俗易懂的术语和定义。但是，正确性和可接受性强烈依赖于术语映射和建模质量以及术语和定义本身的质量。因此，在实践中使用概括之前，需要进行验证和质量改进，以防止出现不正确和不可接受的澄清。
