Table 3. Performance of risk scores in longitudinal validation for the prediction of incident diabetes.
| Risk score | Cutoff | Patients at high risk(%) | AROC (95% CI) | Sensitivity (%) | Specificity (%) | Youden Index | PPV (%) | NPV (%) | P value* |
|---|---|---|---|---|---|---|---|---|---|
| Korean Risk Score (KRS) [13] | ≥6 | 47 | 0.696 (0.656–0.736) | 74 | 54 | 29 | 7.1 | 98 | N/A |
| - | ≥5 | 64 | - | 89 | 37 | 26 | 6.2 | 99 | - |
| Australian score (AUSDRISK study) [16] | ≥11 | 43 | 0.721 (0.679–0.762) | 73 | 58 | 31 | 7.6 | 98 | 0.121 |
| - | ≥12 | 35 | - | 62 | 66 | 28 | 7.8 | 97 | - |
| Finnish score (DETECT-2 study) [22] | ≥4 | 55 | 0.718 (0.678–0.758) | 85 | 46 | 31 | 6.9 | 99 | 0.205 |
| - | ≥7 | 21 | - | 46 | 80 | 26 | 9.8 | 97 | - |
| Thai score [21] | ≥6 | 56 | 0.713 (0.675–0.752) | 86 | 45 | 31 | 6.9 | 99 | 0.309 |
| - | ≥7 | 47 | - | 76 | 54 | 30 | 7.2 | 98 | - |
| Danish score [17] | ≥25 | 38 | 0.700 (0.658–0.742) | 65 | 64 | 29 | 7.8 | 98 | 0.845 |
| - | ≥31 | 22 | - | 43 | 79 | 22 | 8.8 | 97 | - |
| The Leicester Risk Assessment score [20] | ≥13 | 45 | 0.697 (0.655–0.739) | 74 | 57 | 30 | 7.4 | 98 | 0.967 |
| - | ≥16 | 35 | - | 62 | 66 | 28 | 8.0 | 97 | - |
| Japanese score (TOPICS-10 study) [19] | ≥8† | 33 | 0.696 (0.655–0.738) | 58 | 68 | 26 | 7.9 | 97 | 0.976 |
| Chinese score [26] | ≥16 | 37 | 0.692 (0.651–0.732) | 66 | 64 | 31 | 8.0 | 98 | 0.838 |
| - | ≥14 | 58 | - | 84 | 43 | 27 | 6.4 | 98 | - |
| Indian score [24] | ≥18 | 38 | 0.689 (0.648–0.729) | 64 | 63 | 27 | 7.5 | 97 | 0.703 |
| - | ≥17 | 42 | - | 68 | 59 | 26 | 7.2 | 97 | - |
| Japanese score (Doi et al.) [25] | ≥12 | 39 | 0.688 (0.643–0.733) | 69 | 62 | 31 | 7.9 | 98 | 0.658 |
| - | ≥14 | 28 | - | 57 | 73 | 30 | 9.0 | 97 | - |
| ADA questionnaire [18] | ≥3 | 38 | 0.688 (0.644–0.731) | 66 | 63 | 29 | 7.8 | 98 | 0.697 |
| - | ≥5 | 5 | - | 17 | 95 | 12 | 14.7 | 96 | - |
| Brazilian score [23] | ≥12 | 40 | 0.683 (0.640–0.727) | 68 | 61 | 29 | 7.5 | 98 | 0.554 |
| - | ≥18 | 15 | - | 30 | 86 | 16 | 9.1 | 96 | - |
| Oman score [28] | ≥9 | 57 | 0.680 (0.639–0.722) | 82 | 45 | 27 | 6.5 | 98 | 0.479 |
| - | ≥10 | 39 | - | 64 | 62 | 26 | 7.3 | 97 | - |
| British score [12] | ≥4 | 44 | 0.670 (0.627–0.714) | 69 | 58 | 27 | 7.1 | 98 | 0.187 |
| - | ≥6 | 16 | - | 32 | 85 | 17 | 9.0 | 96 | - |
| French score (DESIR study) [29] | ≥3† | 26 | 0.654 (0.608–0.699) | 51 | 75 | 27 | 8.9 | 97 | 0.006 |
| Rotterdam model [27] | ≥32.9 | 47 | 0.646 (0.599–0.693) | 71 | 54 | 25 | 6.8 | 98 | 0.035 |
| - | ≥37 | 32 | - | 50 | 69 | 19 | 7.1 | 97 | - |
| Kuwait score [30] | ≥22 | 40 | 0.630 (0.587–0.674) | 59 | 61 | 20 | 6.6 | 97 | 0.002 |
| - | ≥32 | 8 | - | 18 | 92 | 11 | 10.0 | 96 | - |
Abbreviations: AROC, area under the curve of receiver operating characteristic curve; CI, confidence interval; PPV, positive predictive value; NPV, negative predictive value; N/A, not applicable
Risk scores other than the KRS were arranged in order of the higher AROC value. The performance of different risk scores for the prediction of incident diabetes was evaluated with each model’s original cutoff value and new cutoff value showing the highest Youden index. The results in the upper row of each score are based on the new cutoff value with the highest Youden index. The results in the bottom row of each score are based on the original cutoff value.
*P values for the comparison of ROC curves between the KRS and other scores were calculated using DeLong’s methods [31].
†The new cutoff and original cutoff were the same for Japanese scores and French scores.