Figure 3

A receiver-operating characteristic curve for three models of type 2 diabetes risk assignment based on genotyping specific SNPs. The detection rate (vertical axis) is plotted versus the false-positive rate (horizontal axis) for 4, 8, and 30 genomic markers associated with type 2 diabetes. The dashed line indicates a “useless” test in which the false-positive rate and detection rate are equal. The four-marker model (open squares) has an area under the curve (AUC) of 0.547 and includes the strongest markers with the highest credibility ( Table 2 ). The eight-marker model (open crossed diamonds) improved AUC to 0.551, by adding four more widely used markers. The 30-marker model includes all markers on identified panels, along with additional markers identified through the examination of genome-wide association studies. The AUC increases to 0.570.