Table 2:

Performance metrics using the majority vote of the three most senior paediatric ophthalmologists (CR1, CR2, and CR3) as reference standard

	Healthy		Pre-plus disease		Plus disease
	Sensitivity	Specificity	Sensitivity	Specificity	Sensitivity	Specificity
Bespoke model*	0.973	0.900 (0.640–0.978)	0.860	0.860 (0.612–0.943)	0.522	0.981 (0.948–1.000)
CFDL model*	0.973	0.843 (0.700–0.978)	0.860	0.866 (0.796–0.930)	0.522	1.000 (0.994–1.000)
CR4	0.973	0.955	0.860	0.841	0.522	0.987
JR1	0.964	0.955	0.860	0.873	0.652	0.987
AHP1	0.928	0.865	0.674	0.860	0.696	0.987
JR2	0.964	0.921	0.744	0.930	0.826	0.968
JR3	0.964	0.775	0.442	0.866	0.587	0.961
JR4	0.748	0.989	0.372	0.834	0.935	0.799
JR5	0.901	0.843	0.558	0.796	0.522	0.961

Data are sensitivity, specificity, or specificity (95% CI). CR4 is the consultant rater who was part of the group of seven additional raters for the internal validation of the models but not part of the three consultant raters who provided the reference standard. AHP=allied health professional. CFDL=code-free deep learning. CR=consultant rater. JR=junior rater.

^*Sensitivity of the bespoke and CFDL models were matched to CR4.