Table 2.

Cross-Validated Efficiency, Defined as the Rate of Positive Predictions (Proportion of the Population Flagged as High Risk) to Achieve a Fixed Sensitivity for Correct Classification of Seroconversions (Top); and Cross-Validated Sensitivity That Would Have Been Achieved When Fixing the Rate of Positive Predictions (Bottom)

	Rate of Positive Predictions, %
Needed to meet a minimum sensitivity, %	Risk Groupa	Model-based	Machine Learning
50	42	27	18
60	NA	39	26
70	NA	51	37
80	NA	63	48
	Sensitivity Achieved, %
Limiting the rate of positive predictions, %	Risk Groupb	Model-based	Machine Learning
20	8	40	52
30	8	55	65
40	8	68	74
45	58	68	78

Abbreviation: NA, not applicable.

^aA strategy to target all persons with at least 1 known risk factor (score ≥ 1) would have offered intensified prevention to 42% of the population; a lower threshold (score ≥ 0) is NA.

^bA strategy to target all persons with at least 2 known risk factors (score ≥ 2) would have achieved 8% sensitivity, while a strategy to target all persons with at least 1 known risk factor (score ≥ 1) would have achieved 58% sensitivity.