John A. Burns School of Medicine, University of Hawai'i, Honolulu, HI (BKS).
Department of Surgery, John A. Burns School of Medicine, University of Hawai'i, Honolulu, HI (PIT).
Hawaii J Health Soc Welf. 2023 Oct;82(10 Suppl 1):84-88.
Studies that examine racial disparities in health outcomes often include analyses that account or adjust for baseline differences in co-morbid conditions. Often, these conditions are defined as dichotomous (Yes/No) variables, and few analyses include clinical and/or laboratory data that could allow for more nuanced estimates of disease severity. However, disease severity - not just prevalence - can differ substantially by race and is an underappreciated mechanism for health disparities. Thus, relying on dichotomous disease indicators may not fully describe health disparities. This study explores the effect of substituting continuous clinical and/or laboratory data for dichotomous disease indicators on racial disparities, using data from the Queen's Medical Center's (QMC) cardiac surgery database (a subset of the national Society of Thoracic Surgeon's cardiothoracic surgery database) as an example case. Two logistic regression models predicting in-hospital mortality were constructed: (I) a baseline model including race and dichotomous (Yes/No) indicators of disease (diabetes, heart failure, liver disease, kidney disease), and (II) a more detailed model with continuous laboratory values in place of the dichotomous indicators (eg, including Hemoglobin A1c level rather than just diabetes yes/no). When only dichotomous disease indicators were used in the model, Native Hawaiian and other Pacific Islander (NHPI) race was significantly associated with in-hospital mortality (OR: 1.57[1.29,2.47], P=.04). Yet when the more specific laboratory values were included, NHPI race was no longer associated with in-hospital mortality (OR: 1.67[0.92,2.28], P=.28). Thus, researchers should be thoughtful in their choice of independent variables and understand the potential impact of how clinical measures are operationalized in their research.
研究种族健康结果差异时，通常会进行分析，以说明或调整共病状况的基线差异。这些状况通常被定义为二分类（是/否）变量，很少有分析包括可以更细致地估计疾病严重程度的临床和/或实验室数据。然而，疾病严重程度——而不仅仅是患病率——可能因种族而有很大差异，这是健康差异的一个未被充分认识的机制。因此，仅依赖二分类疾病指标可能无法充分描述健康差异。本研究以皇后医疗中心（QMC）心脏手术数据库（全国胸外科医生学会心胸外科手术数据库的一个子集）的数据为例，探讨了用连续临床和/或实验室数据替代二分类疾病指标对种族差异的影响。构建了两个预测住院死亡率的逻辑回归模型：（I）基线模型，包含种族和疾病的二分类（是/否）指标（糖尿病、心力衰竭、肝病、肾病）；（II）更详细的模型，用连续实验室值替代二分类指标（例如，用血红蛋白 A1c 水平代替糖尿病是/否）。当模型中仅使用二分类疾病指标时，夏威夷原住民和其他太平洋岛民（NHPI）种族与住院死亡率显著相关（OR：1.57[1.29,2.47]，P=.04）。然而，当纳入更具体的实验室值时，NHPI 种族与住院死亡率不再相关（OR：1.67[0.92,2.28]，P=.28）。因此，研究人员在选择自变量时应深思熟虑，并了解临床测量在其研究中的操作方式的潜在影响。
