Department of Epidemiology and Public Health, University College London, London, UK.
Diabet Med. 2010 Jan;27(1):46-53. doi: 10.1111/j.1464-5491.2009.02891.x.
To compare the performance of nine published strategies for the selection of individuals prior to screening for undiagnosed diabetes.
We conducted a validation study, based on a cross-sectional analysis of 6990 participants of the Whitehall II study, an occupational cohort of civil servants in London. We calculated sensitivity, specificity and the area under the receiver operating characteristic (ROC) curve, indicative of the ability of a risk score to correctly identify those with undiagnosed diabetes.
The prevalence of unknown diabetes was 2.0%. At a set level of sensitivity (0.70), the specificity of the different scores ranged between 0.41 and 0.57. A reference model, based solely on age and body mass index had an area under the ROC curve of 0.67 [95% confidence interval (CI): 0.62, 0.72]. Four scores had a lower area under the ROC curve (lowest ROC AUC: 0.62; 95% CI: 0.58, 0.67) compared with the reference model, while the other five scores had similar areas (highest ROC AUC: 0.68; 95% CI: 0.63, 0.72). All ROC curve areas were lower than those reported in the original publications and validation studies.
Existing risk scores for the detection of undiagnosed diabetes perform less well in a large validation cohort compared with previous validation studies. Our study indicates that non-invasive risk scores require further refinement and testing before they can be used as the first step in a diabetes screening programme.
比较 9 种已发表的策略在筛选未诊断糖尿病个体前的表现。
我们进行了一项验证性研究，基于对伦敦公务员职业队列 Whitehall II 研究的 6990 名参与者的横断面分析。我们计算了敏感性、特异性和接受者操作特征（ROC）曲线下面积（表明风险评分正确识别未诊断糖尿病患者的能力）。
未知糖尿病的患病率为 2.0%。在设定的敏感性（0.70）水平下，不同评分的特异性在 0.41 至 0.57 之间。一个仅基于年龄和体重指数的参考模型的 ROC 曲线下面积为 0.67 [95%置信区间（CI）：0.62，0.72]。与参考模型相比，有 4 个评分的 ROC 曲线下面积较低（ROC AUC 最低：0.62；95%CI：0.58，0.67），而其他 5 个评分的 ROC 曲线下面积相似（ROC AUC 最高：0.68；95%CI：0.63，0.72）。所有 ROC 曲线下面积均低于原始出版物和验证研究报告的面积。
与之前的验证研究相比，现有的用于检测未诊断糖尿病的风险评分在大型验证队列中的表现较差。我们的研究表明，非侵入性风险评分在用于糖尿病筛查计划的第一步之前需要进一步改进和测试。
