Performance comparisons for ordinal mRS prediction between the DLPD and the clinical readersa
| Gwet AC | MAE | ±1ACC (%) | ACC (%) | |
|---|---|---|---|---|
| Model/readers | ||||
| Neuroradiologist I | 0.70 (0.60–0.80) | 1.15 (0.94–1.38) | 71 (60–81) | 26 (18–36) |
| Neuroradiologist II | 0.69 (0.59–0.79) | 1.14 (0.93–1.38) | 70 (60–80) | 29 (19–39) |
| Neuroradiology fellow | 0.73 (0.65–0.81) | 1.04 (0.85–1.24) | 74 (6–84) | 31 (21–41) |
| Clinical readers | ||||
| Neurologist I | 0.75 (0.66–0.84) | 1.03 (0.83–1.25) | 75 (65–84) | 32 (22–44) |
| Neurologist II | 0.77 (0.67–0.86) | 0.91 (0.70–1.15) | 79 (69–88) | 41 (30–51) |
| Consensus read | 0.76 (0.67–0.84) | 0.95 (0.75–1.17) | 79 (70–88) | 36 (25–46) |
| DLPD | 0.79 (0.71–0.86) | 0.89 (0.70–1.11) | 81 (73–90) | 36 (26–46) |
| P value | P < .001 | P = .02 | P < .001 | P = .07 |
↵a The data in the parentheses represent the 95% confidence interval. The P value is for the noninferiority test between the consensus clinical reads and the DLPD with the predefined margin of 0.1 (MAE)/10% (±1ACC, ACC). The Gwet AC for agreement among 5 clinical readers is 0.83 (95% CI, 0.80–0.86), justifying the comparison with a consensus.