Institute for Medical Informatics and Biometry, Carl Gustav Carus Faculty of Medicine, Technische Universität Dresden, Dresden, Germany.
Stud Health Technol Inform. 2022 Aug 31;298:73-77. doi: 10.3233/SHTI220910.
Data quality is essential for utilizing real world data (RWD) in scientific context. Based on drug prescriptions in a hospital information system (HIS), algorithms performed a mapping of unstructured drug data to ATC codes. Visualization of the resulting distribution of structured to unstructured data based on ATC codes was created and used to explore a defined limitation of the current drug prescription highlighting the example of proton pump inhibitors. As a second step, a generalization of this approach was inductively created. As result we were able to identify 4 crucial steps for a feedback loop framework: The first step being the actual use of the HIS by clinician for drug prescription, second the processing of the entered unstructured and structured data and performing automatic analyses and visualization of the resulting distributions. The third step included an interdisciplinary expert evaluation of the data distribution followed by the fourth step, consisting of feedback to the stakeholders and generating actions as teaching or re-modelling of the system incorporating the actual learning process. The presented approach represents a continuously learning system based on RWD, although it is limited by analyzing the distribution of mapped unstructured text to ATC codes and therefore does not allow to analyze free text not mapped to ATC codes (false negatives). Future work will focus on the evaluation of this approach to analyze the impact on prescription data quality and the potential improvement on patient safety in general.
数据质量对于在科学环境中利用真实世界数据（RWD）至关重要。基于医院信息系统（HIS）中的药物处方，算法执行了将非结构化药物数据映射到 ATC 代码的操作。根据 ATC 代码创建了结构化到非结构化数据的分布的可视化效果，并用于探索当前药物处方的一个明确限制，以质子泵抑制剂为例。作为第二步，我们归纳性地创建了这种方法的推广。结果，我们能够确定反馈循环框架的 4 个关键步骤：第一步是临床医生实际使用 HIS 进行药物处方，第二步是处理输入的非结构化和结构化数据，并对结果分布执行自动分析和可视化。第三步包括对数据分布进行跨学科专家评估，然后是第四步，向利益相关者反馈并生成操作，例如教学或重新建模系统，将实际学习过程纳入其中。所提出的方法代表了一个基于 RWD 的持续学习系统，尽管它受到分析映射到 ATC 代码的非结构化文本分布的限制，因此无法分析未映射到 ATC 代码的自由文本（假阴性）。未来的工作将侧重于评估这种方法，以分析其对处方数据质量的影响以及对总体患者安全的潜在改善。
