台湾长庚研究数据库出院状态验证。
Discharge status validation of the Chang Gung Research database in Taiwan.
机构信息
Center for Big Data Analytics and Statistics, Department of Medical Research and Development, Chang Gung Memorial Hospital at Linkou, Taoyuan, Taiwan.
Division of Geriatrics and General Internal Medicine, Department of Internal Medicine, Chang Gung Memorial Hospital, Taoyuan, Taiwan; School of Medicine, College of Medicine, Chang Gung University, Taoyuan, Taiwan.
出版信息
Biomed J. 2022 Dec;45(6):907-913. doi: 10.1016/j.bj.2021.12.006. Epub 2021 Dec 28.
BACKGROUND
The Chang Gung Research Database (CGRD) is the largest multi-institutional electronic medical records database in Taiwan and has been widely used to establish evidence studies. However, the accuracy of CGRD has rarely been validated. This study aims to validate the discharge status, especially with a focus on mortality, of admission data under CGRD.
METHODS
We constructed an observational study using CGRD linked with TDR to validate the discharge status. The CGRD and TDR data were obtained from the Chang Gung Memorial Hospital system and the Health and Welfare Data Science Center, respectively. The accuracy, positive predictive value (PPV), and underestimated mortality rate (UEM) were employed as indicators for validation. Year, sex, age, and the primary cause for admission (PCA) were analyzed.
RESULTS
A total of 1,972,044 admission records under CGRD were analyzed. The overall accuracy for mortality coding on discharge status was higher than 97% within one week after discharge. The accuracy increased by year and was more than 98% after 2010. A similar result was observed in UEM; the UEM within one week was lower than 10% after 2010. These indicators varied by age group and PCA-elderly patients had relatively lower accuracy and higher UEM (approximately 11%). The presence of UEM within one week was better but varied by disease.
CONCLUSIONS
Considering the data accuracy and UEM discharge status, prioritizing the use of inpatient data after 2010 under CGRD for mortality outcome follow-up studies is recommended.
背景
长庚研究数据库（CGRD）是台湾最大的多机构电子病历数据库，已广泛用于建立证据研究。然而，CGRD 的准确性很少得到验证。本研究旨在验证 CGRD 下的入院数据的出院状态，特别是死亡率。
方法
我们使用 CGRD 与 TDR 构建了一项观察性研究，以验证出院状态。CGRD 和 TDR 数据分别来自长庚纪念医院系统和卫生福利数据科学中心。准确性、阳性预测值（PPV）和低估死亡率（UEM）被用作验证指标。分析了年份、性别、年龄和入院的主要原因（PCA）。
结果
共分析了 1972044 份 CGRD 下的入院记录。出院状态下死亡率编码的整体准确性在出院后一周内高于 97%。准确性随年份增加，2010 年后超过 98%。UEM 也观察到类似的结果；2010 年后一周内的 UEM 低于 10%。这些指标按年龄组和 PCA 变化；老年患者的准确性相对较低，UEM 较高（约 11%）。一周内存在 UEM 的情况更好，但因疾病而异。
结论
考虑到数据准确性和 UEM 出院状态，建议优先在 2010 年后使用 CGRD 中的住院数据进行死亡率随访研究。
Zotero 插件