Table 3.

ROC analyses of CFMQ score screening for the top 2.5^th percentile of facial recognition ability.

ROC Analyses - Screening for top 2.5%-ile
Optimization type	Cutoff score	Sensitivity	95% CI	Specificity	95% CI	PPV	95% CI
Youden’s J	72	76.92%	61.66% – 87.35%	67.68%	65.25% – 70.02%	5.91%	4.17% – 8.31%
Standard	85	7.69%	2.65% – 20.32%	99.73%	99.31% – 99.89%	42.86%	15.82% – 74.95%
Maximum PPV	85	7.69%	2.65% – 20.32%	99.73%	99.31% – 99.89%	42.86%	15.82% – 74.95%

Note. PPV = Positive Predictive Value

Youden’s J returns the operating point where sensitivity rate + specificity rate -1 is maximized. Thus, with this optimization, the cost of a loss in sensitivity rate is equal to the cost of a loss in specificity rate.

The standard optimization is similar to Youden’s J, except that costs are adjusted to reflect that those without prosopagnosia greatly outnumber those who with prosopagnosia. The cost of loss of specificity, therefore, greatly outweighs the cost of loss of sensitivity. In practice, this second criterion attempts a balance between sensitivity and positive predictive value, and so returns the main CFMQ cutoff of interest.