Performance at a predictive model to identity undiagnosed diabetes in a health care setting
Performance at a predictive model to identity undiagnosed diabetes in a health care setting
复制标题
DOI：
10.2337/diacare.22.2.213
复制
发表时间：
1999-02-01
期刊：
影响因子：
16.2
通讯作者：
Feskens, EJM
中科院分区：
文献类型：
--
作者：
Baan, CA;Ruige, JB;Feskens, EJM
研究设计和方法-鹿特丹研究的参与者样本（n = 1，016），年龄55 - 75岁，不知道有糖尿病的人完成了一份关于糖尿病相关症状和风险因素的问卷，并进行了葡萄糖耐量试验。使用逐步逻辑回归分析开发预测模型，将新诊断的糖尿病的存在或不存在作为因变量，将与糖尿病有合理联系的各种项目作为自变量。在另一项荷兰基于人群的研究Hoorn研究（n = 2，364）中评估了这些模型，其中参与者年龄在50 - 74岁之间。预测模型的性能进行了比较，通过使用受试者-操作者特征（ROC）current.RESULTS-我们开发了三个预测模型（PM）。PM1包含全科医生常规收集的信息，而PM2还包含通过额外问题获得的变量。第三个预测模型PM3包括必须从体检中获得的变量。这些后一个变量没有附加的预测值，导致PM3类似于PM2。PM2的ROC曲线下面积高于PM1，但95%CI重叠（分别为0.74 [0.70 - 0.78]和0.68 [0.64 - 0.72]）。结论-仅使用全科医生薯条中通常存在的信息，开发了一种预测模型，该模型的表现类似于由额外问题获得的信息补充的模型。PM1的简单性使其易于在当前的医疗保健环境中实现。
OBJECTIVE - To develop a predictive model to identify individuals with an increased risk for undiagnosed diabetes, allowing for the availability of information within the health care system.RESEARCH DESIGN AND METHODS - A sample of participants from the Rotterdam Study (n = 1,016), aged 55-75 years, not known to have diabetes completed a questionnaire on diabetes-related symptoms and risk factors and underwent a glucose tolerance test. Predictive models were developed using stepwise logistic regression analyses with the absence or presence of newly diagnosed diabetes as the dependent variable and various items with a plausible connection to diabetes as the independent variables. The models were evaluated in another Dutch population-based study the Hoorn Study (n = 2,364), in which the participants were aged 50-74 years. Performances of the predictive models were compared by using receiver-operator characteristics (ROC) curves.RESULTS - We developed three predictive models (PMs). PM1 contained information routinely collected by the general practitioner, while PM2 also contained variables obtainable by additional questions. The third predictive model, PM3, included variables that had to be obtained from a physical examination. These latter variables did not have additive predictive value, resulting in a PM3 similar to PM2. The area under the ROC curve was higher for PM2 than for PM1,but the 95% CIs overlapped (0.74 [0.70-0.78] and 0.68 [0.64-0.72], respectively).CONCLUSIONS - Using only information normally present in the fries of a general practitioner, a predictive model was developed that performed similarly to one supplemented by information obtained from additional questions. The simplicity of PM1 makes it easy to implement in the current health care setting.