Table 5.

PPV and NPV of the ED and admissions models, calibrated during training to 70% and 80% sensitivities, for identifying COVID-19 in test sets with various prevalences

		Prevalence of COVID-19 in test set
		1%	2%	5%	10%*	20%†	25%	33%	50%
ED model
Sensitivity 0·70
	PPV	0·203	0·383	0·613	0·763	0·834	0·902	0·888	0·979
	NPV	0·996	0·990	0·985	0·953	0·932	0·871	0·886	0·778
Sensitivity 0·80
	PPV	0·133	0·282	0·493	0·638	0·767	0·831	0·823	0·944
	NPV	0·997	0·993	0·991	0·962	0·946	0·909	0·908	0·820
Admissions model
Sensitivity 0·70
	PPV	0·175	0·304	0·513	0·595	0·830	0·859	0·876	0·950
	NPV	0·996	0·992	0·982	0·969	0·926	0·905	0·881	0·785
Sensitivity 0·80
	PPV	0·098	0·211	0·390	0·509	0·755	0·797	0·812	0·922
	NPV	0·998	0·994	0·986	0·977	0·942	0·920	0·907	0·841

ED=emergency department. NPV=negative predictive values. PPV=positive predictive values.

^*The 10% scenario approximates the observed prevalence of COVID-19 in patients presenting to the study hospitals during April 1–8, 2020.

^†The 20% scenario approximates the observed prevalence of COVID-19 in patients admitted to the study hospitals during April 1–8, 2020.