Table 3.

Performance metrics of the model and eyepacs graders on the secondary clinical validation set.

Metric	Model	EyePACS Graders
Positive predictive value (%), 95% CI	35% [27–44%]	18% [13–23%]
Negative predictive value (%), 95% CI	96% [95–98%]	95% [94–97%]
Sensitivity (%), 95% CI	57% [47–69%]	55% [43–66%]
Specificity (%), 95% CI	91% [89–93%]	79% [76–82%]
Accuracy (%), 95% CI	88% [86–91%]	77% [74–80%]
Cohen’s Kappa, 95% CI	0.38 [0.29–0.47]	0.17 [0.11–0.24]

For the model we chose an operating point that matched the sensitivity of the eyepacs graders to calculate the metrics. Brackets denote 95% confidence intervals. n = 990 images for all calculations