Husain N, Blais P, Kramer J, Kowalkowski M, Richardson P, El-Serag H B, Kanwal F
Department of Medicine, Baylor College of Medicine, Houston, TX, USA; Section of Gastroenterology and Hepatology, Baylor College of Medicine and Michael E. DeBakey Veterans Affairs Medical Center, Houston, TX, USA.
Aliment Pharmacol Ther. 2014 Oct;40(8):949-54. doi: 10.1111/apt.12923. Epub 2014 Aug 26.
In practice, nonalcoholic fatty liver disease (NAFLD) is diagnosed based on elevated liver enzymes and confirmatory liver biopsy or abdominal imaging. Neither method is feasible in identifying individuals with NAFLD in a large-scale healthcare system.
To develop and validate an algorithm to identify patients with NAFLD using automated data.
Using the Veterans Administration Corporate Data Warehouse, we identified patients who had persistent ALT elevation (≥2 values ≥40 IU/mL ≥6 months apart) and did not have evidence of hepatitis B, hepatitis C or excessive alcohol use. We conducted a structured chart review of 450 patients classified as NAFLD and 150 patients who were classified as non-NAFLD by the database algorithm, and subsequently refined the database algorithm.
The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) for the initial database definition of NAFLD were 78.4% (95% CI: 70.0-86.8%), 74.5% (95% CI: 68.1-80.9%), 64.1% (95% CI: 56.4-71.7%) and 85.6% (95% CI: 79.4-91.8%), respectively. Reclassifying patients as having NAFLD if they had two elevated ALTs that were at least 6 months apart but within 2 years of each other, increased the specificity and PPV of the algorithm to 92.4% (95% CI: 88.8-96.0%) and 80.8% (95% CI: 72.5-89.0%), respectively. However, the sensitivity and NPV decreased to 55.0% (95% CI: 46.1-63.9%) and 78.0% (95% CI: 72.1-83.8%), respectively.
Predictive algorithms using automated data can be used to identify patients with NAFLD, determine prevalence of NAFLD at the system-wide level, and may help select a target population for future clinical studies in veterans with NAFLD.
在实际操作中，非酒精性脂肪性肝病（NAFLD）是根据肝酶升高以及确诊性肝活检或腹部影像学检查来诊断的。在大规模医疗保健系统中，这两种方法对于识别NAFLD患者都不可行。
开发并验证一种使用自动化数据识别NAFLD患者的算法。
利用退伍军人事务部企业数据仓库，我们识别出谷丙转氨酶（ALT）持续升高（≥2次，每次≥40 IU/mL，间隔≥6个月）且无乙肝、丙肝或过度饮酒证据的患者。我们对数据库算法分类为NAFLD的450例患者和分类为非NAFLD的150例患者进行了结构化图表审查，随后完善了数据库算法。
NAFLD初始数据库定义的敏感性、特异性、阳性预测值（PPV）、阴性预测值（NPV）分别为78.4%（95%置信区间：70.0 - 86.8%）、74.5%（95%置信区间：68.1 - 80.9%）、64.1%（95%置信区间：56.4 - 71.7%）和85.6%（95%置信区间：79.4 - 91.8%）。如果患者有两次间隔至少6个月但在彼此2年内升高的ALT，则将其重新分类为患有NAFLD，这将算法的特异性和PPV分别提高到92.4%（95%置信区间：88.8 - 96.0%）和80.8%（95%置信区间：72.5 - 89.0%）。然而，敏感性和NPV分别降至55.0%（95%置信区间：46.1 - 63.9%）和78.0%（95%置信区间：72.1 - 83.8%）。
使用自动化数据的预测算法可用于识别NAFLD患者，确定全系统范围内NAFLD的患病率，并可能有助于为未来NAFLD退伍军人的临床研究选择目标人群。
