Division of Drug Development and Regulatory Science, Graduate School of Pharmaceutical Sciences, Keio University, Shibakoen, Minato-ku, Tokyo, Japan.
Department of Dermatology, Keio University School of Medicine, Shinanomachi, Shinjuku-ku, Tokyo, Japan.
PLoS One. 2019 Aug 13;14(8):e0221130. doi: 10.1371/journal.pone.0221130. eCollection 2019.
Stevens-Johnson syndrome (SJS) and toxic epidermal necrolysis (TEN), severe drug reactions, are often misdiagnosed due to their rarity and lack of information on differential diagnosis. The objective of the study was to develop an electronic medical record (EMR)-based algorithm to identify patients with SJS/TEN for future application in database studies. From the EMRs of a university hospital, two dermatologists identified all 13 patients with SJS/TEN seen at the Department of Dermatology as the case group. Another 1472 patients who visited the Department of Dermatology were identified using the ICD-10 codes for diseases requiring differentiation from SJS/TEN. One hundred of these patients were then randomly sampled as controls. Based on clinical guidelines for SJS/TEN and the experience of the dermatologists, we tested 128 algorithms based on the use of ICD-10 codes, clinical courses for SJS/TEN, medical encounters for mucocutaneous lesions from SJS/TEN, and items to exclude paraneoplastic pemphigus. The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and diagnostic odds ratio (DOR) of each algorithm were calculated, and the optimal algorithm was defined as that with high PPV and maximal sensitivity and specificity. One algorithm, consisting of a combination of clinical course for SJS/TEN, medical encounters for mucocutaneous lesions from SJS/TEN, and items to exclude paraneoplastic pemphigus, but not ICD-10 codes, showed a sensitivity of 76.9%, specificity of 99.0%, PPV of 40.5%, NPV of 99.8%, and DOR of 330.00. We developed a potentially optimized algorithm for identifying SJS/TEN based on clinical practice records. The almost perfect specificity of this algorithm will prevent bias in estimating relative risks of SJS/TEN in database studies. Considering the small sample size, this algorithm should be further tested in different settings.
史蒂文斯-约翰逊综合征（SJS）和中毒性表皮坏死松解症（TEN）是严重的药物反应，由于其罕见性和缺乏鉴别诊断信息，常被误诊。本研究的目的是开发一种基于电子病历（EMR）的算法，以识别 SJS/TEN 患者，以便将来在数据库研究中应用。从一家大学医院的 EMR 中，两名皮肤科医生确定了皮肤科就诊的 13 名 SJS/TEN 患者作为病例组。另使用 SJS/TEN 需要鉴别诊断的疾病的 ICD-10 编码，从皮肤科就诊患者中确定了 1472 名患者。然后从这 1472 名患者中随机抽取 100 名作为对照组。基于 SJS/TEN 的临床指南和皮肤科医生的经验，我们测试了 128 种基于 ICD-10 编码、SJS/TEN 临床病程、SJS/TEN 黏膜皮肤病变的就诊情况以及排除副肿瘤天疱疮的项目的算法。计算了每种算法的敏感性、特异性、阳性预测值（PPV）、阴性预测值（NPV）和诊断比值比（DOR），并将高 PPV 和最大敏感性和特异性的最佳算法定义为最优算法。一种算法，由 SJS/TEN 的临床病程、SJS/TEN 的黏膜皮肤病变就诊情况和排除副肿瘤天疱疮的项目组成，但不包括 ICD-10 编码，其敏感性为 76.9%，特异性为 99.0%，PPV 为 40.5%，NPV 为 99.8%，DOR 为 330.00。我们基于临床实践记录开发了一种潜在优化的 SJS/TEN 识别算法。该算法几乎完美的特异性将防止在数据库研究中估计 SJS/TEN 的相对风险时出现偏差。考虑到样本量小，该算法应在不同环境中进一步测试。
