Performance of existing risk scores in screening for undiagnosed diabetes: an external validation study

Aim To compare the performance of nine published strategies for the selection of individuals prior to screening for undiagnosed diabetes. Methods 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. Results 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 ROCAUC: 0.68; 95% CI: 0.63, 0.72). All ROC curve areas were lower than those reported in the original publications and validation studies. Conclusions 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.