Table 2.
Models | Candidate predictorsa | AUC (95% CI) | Sensitivity (SD) |
PPV (SD) |
||
---|---|---|---|---|---|---|
90% specificity | 95% specificity | 90% specificity | 95% specificity | |||
Fusion enhanced model | Local features + fused risk scores | 0.86 (0.84, 0.89) | 0.70 (0.03) | 0.65 (0.02) | 0.082 (0.008) | 0.134 (0.015) |
Conventional model | Local features | 0.82 (0.81, 0.84) | 0.42 (0.07) | 0.24 (0.07) | 0.050 (0.009) | 0.057 (0.018) |
Fusion only model | Fused risk scores | 0.52 (0.49, 0.55) | 0.10 (0.03) | 0.04 (0.03) | 0.012 (0.003) | 0.011 (0.006) |
AUC: area under the receiver operating characteristics curve; CI: confidence interval; PPV: positive predictive value; SD: standard deviation.
Local features included demographics and diagnosis codes collected in target cohort, that is, HIDD cohort.