Center for Crohn's and Colitis, Department of Gastroenterology, Kyung Hee University College of Medicine, Seoul, South Korea.
Department of Biostatistics, Yonsei University Wonju College of Medicine, Wonju, Republic of Korea.
J Gastroenterol Hepatol. 2020 May;35(5):760-768. doi: 10.1111/jgh.14855. Epub 2019 Oct 27.
We conducted a nationwide validation study of diagnostic algorithms to identify cases of inflammatory bowel disease (IBD) within the Korea National Health Insurance System (NHIS) database.
Using the NHIS dataset, we developed 44 algorithms combining the International Classification of Diseases (ICD)-10 codes, codes for Rare and Intractable Diseases (RID) registration and claims data for health care encounters, and pharmaceutical prescriptions for IBD-specific drugs. For each algorithm, we compared the case identification results from electronic medical records data with the gold standard (chart-based diagnosis). A multiple sampling test verified the validation results from the entire study population.
A random nationwide sample of 1697 patients (848 potential cases and 849 negative control cases) from 17 hospitals were included for validation. A combination of the ICD-10 code, ≥ 1 claims for health care encounters, and ≥ 1 prescription claims (reference algorithm) achieved excellent performance (sensitivity, 93.1% [95% confidence interval 91-94.7]; specificity, 98.1% [96.9-98.8]; positive predictive value, 97.5% [96.1-98.5]; negative predictive value, 94.5% [92.8-95.8]) with the lowest error rate (4.2% [3.3-5.3]). The multiple sampling test confirmed that the reference algorithm achieves the best performance regarding IBD diagnosis. Algorithms including the RID registration codes exhibited poorer performance compared with that of the reference algorithm, particularly for the diagnosis of patients affiliated with secondary hospitals. The performance of the reference algorithm showed no statistical difference depending on the hospital volume or IBD type, with P-value < 0.05.
We strongly recommend the reference algorithm as a uniform standard operational definition for future studies using the NHIS database.
我们在韩国国民健康保险系统（NHIS）数据库中对用于识别炎症性肠病（IBD）病例的诊断算法进行了全国性验证研究。
我们使用 NHIS 数据集，结合国际疾病分类（ICD-10）代码、罕见和疑难疾病（RID）登记代码、医疗保健就诊记录代码和 IBD 特异性药物的处方数据，开发了 44 种算法。对于每种算法，我们将电子病历数据中的病例识别结果与金标准（基于图表的诊断）进行比较。多次抽样检验验证了整个研究人群的验证结果。
我们从 17 家医院的 1697 名患者（848 例疑似病例和 849 例阴性对照病例）中进行了随机全国性抽样验证。ICD-10 代码、≥1 次医疗保健就诊记录和≥1 次处方记录的组合（参考算法）具有出色的性能（敏感性 93.1%[95%置信区间 91-94.7]；特异性 98.1%[96.9-98.8]；阳性预测值 97.5%[96.1-98.5]；阴性预测值 94.5%[92.8-95.8]），且错误率最低（4.2%[3.3-5.3]）。多次抽样检验证实，参考算法在 IBD 诊断方面表现最佳。与参考算法相比，包含 RID 登记代码的算法表现较差，尤其是在诊断二级医院患者时。参考算法的性能与医院规模或 IBD 类型无关，P 值<0.05。
我们强烈建议将参考算法作为使用 NHIS 数据库进行未来研究的统一标准操作定义。
