Development of a Claims-Based Risk Score to Identify Obese Individuals
Development of a Claims-Based Risk Score to Identify Obese Individuals
复制标题
DOI：
10.1089/pop.2009.0051
复制
发表时间：
2010-08-01
影响因子：
2.5
通讯作者：
Weiner, Jonathan P.
中科院分区：
文献类型：
--
作者：
Clark, Jeanne M.;Chang, Hsien-Yen;Weiner, Jonathan P.
肥胖症诊断不足，阻碍了基于系统的健康促进和研究。我们的目标是开发和验证一个基于索赔的风险模型，以确定肥胖的人使用医疗诊断和处方记录。我们对3名蓝十字蓝盾计划的登记者的去识别索赔数据进行了横断面分析，这些人完成了健康风险评估，捕获了身高和体重。最终的71，057名参与者被随机分为2个子样本，用于开发和验证肥胖风险模型。使用约翰霍普金斯调整临床组病例组合/预测风险方法，我们对研究成员的诊断（ICD）代码进行分类。使用逻辑回归来确定哪些基于索赔的风险标志物与体重指数（BMI）>= 35 kg/m2相关。得分>= 90（th）百分位数检测肥胖的敏感性为26%至33%，而特异性为> 90%。受试者工作曲线下面积范围为0.67 - 0.73。相比之下，单独诊断肥胖或肥胖药物的敏感性非常低（分别为10%和1%）;肥胖风险模型识别了额外的22%的肥胖成员。将百分位数临界点从第70百分位数变化到第99百分位数，得到的阳性预测值范围为15.5 - 59.2。肥胖风险评分对于检测BMI >= 35 kg/m2具有高度特异性，并且大大增加了对肥胖成员的检测，超出了提供者编码的肥胖诊断或药物声明。该模型可用于肥胖护理管理和健康促进或肥胖相关研究。（人口健康管理2010;13：201-207）
Obesity is underdiagnosed, hampering system-based health promotion and research. Our objective was to develop and validate a claims-based risk model to identify obese persons using medical diagnosis and prescription records. We conducted a cross-sectional analysis of de-identified claims data from enrollees of 3 Blue Cross Blue Shield plans who completed a health risk assessment capturing height and weight. The final sample of 71,057 enrollees was randomly split into 2 subsamples for development and validation of the obesity risk model. Using the Johns Hopkins Adjusted Clinical Groups case-mix/predictive risk methodology, we categorized study members' diagnosis (ICD) codes. Logistic regression was used to determine which claims-based risk markers were associated with a body mass index (BMI) >= 35 kg/m(2). The sensitivities of the scores >= 90(th) percentile to detect obesity were 26% to 33%, while the specificities were >90%. The areas under the receiver operator curve ranged from 0.67 to 0.73. In contrast, a diagnosis of obesity or an obesity medication alone had very poor sensitivity (10% and 1%, respectively); the obesity risk model identified an additional 22% of obese members. Varying the percentile cut-point from the 70(th) to the 99(th) percentile resulted in positive predictive values ranging from 15.5 to 59.2. An obesity risk score was highly specific for detecting a BMI >= 35 kg/m(2) and substantially increased the detection of obese members beyond a provider-coded obesity diagnosis or medication claim. This model could be used for obesity care management and health promotion or for obesity-related research. (Population Health Management 2010;13:201-207)