Analysis of age-related variations in laboratory tests by use of latent information recovered from routine laboratory data

1996年，我开发了一种数据挖掘方法，从常规实验室数据中推导出实验室测试的参考区间(RI)，称为“潜在参考值提取方法”。它的特点是同时设置多个相互关联的测试项目的RI，并严格排除其他测试项目中存在异常值的个人。因此，需要迭代处理来优化RIS。在这项研究中，我开发了一种现实的算法来获得RIS的适当初始估计，排除明显异常的个体，并针对患者群体中常见的扭曲的年龄分布进行调整。基于修订方法的软件不仅具有优化RIS推导的灵活能力，而且还具有使用逐步年龄调整排除异常值来推导特定年龄参考区间的能力。该软件被应用于各种临床实验室数据。在冈山，与年龄相关的RI是针对常见测试项目得出的。在所有分析物中都观察到了RIS与年龄相关的一致变化。它还被应用于揭示主要实验室分析物的RIS(0至23岁)的逐年变化。所有这些人都表现出独特的个人资料，而使用传统方法会导致明显扭曲的资料。它也被发现适用于使用来自血液透析患者的大量数据来获得所谓的疾病特异性RI。

I have developed a data-mining method to derive a reference interval (RI) for laboratory tests from a routine laboratory data, called "latent reference value extraction method", in 1996. It features simultaneous setting of RIs for multiple test items mutually related and a rigid exclusion of those individuals with abnormal values in other test items. Thus it requires an iterative processing to optimize RIs. In this research I developed a realistic algorithm to obtain appropriate initial estimates of RIs, to exclude apparently abnormal individuals, and to adjust for a distorted age distribution commonly seen in patients' population. A software based on the revised methodology not only has a flexible ability to optimize derivation of RIs but also has an ability to derive age-specific reference interval using a stepwise age-adjusted exclusion of abnormal values. The software was applied to various clinical laboratory data. In Okayama, age-related RIs were derived for common test items. Consistent age-related changes in RIs were observed in all the analytes. It was also applied to reveal year-by-year changes in RIs (0 to 23 year old) for major laboratory analytes. All of them showed peculiar profiles while a use of conventional methodology led to apparently distorted profiles. It was also found to be applicable to obtain so-called disease-specific RIs using a large data from hemodialysis patients.