Skip to main content
. Author manuscript; available in PMC: 2020 Apr 13.
Published in final edited form as: Lancet HIV. 2019 Jul 5;6(10):e688–e695. doi: 10.1016/S2352-3018(19)30137-7

Table 3.

Performance of full and simpler HIV risk prediction models among patients with high or very high risk scores in the validation dataset – Kaiser Permanente Northern California, 2015-2017 (N=606,701)

Sensitivity (% of incident HIV cases identified)

HIV risk prediction model N of
patients
flagged
Overall
(n=83)
Male
(n=69)
Female
(n=14)
Black
(n=20)
White
(n=20)
Sensitivity for
incident PrEP
users, %
(n=640)
Specificity, % PPV, % NPV, %
Full LASSO model 13,463 38·6 46·4 0 25·0 25·0 59·1 97·8 0·24 >99·9
MSM status and STI positivity, testing, and treatment 4617 28·9 34·8 0 15·0 35·0 42·7 99·2 0·52 >99·9
STI positivity, testing, and treatment 1458 20·5 24·6 0 10·0 15·0 16·2 99·8 1·17 >99·9
MSM status and STI positivity 3849 25·3 30·4 0 10·0 30·0 38·8 99·4 0·55 >99·9
MSM status 3790 25·3 30·4 0 10·0 30·0 38·1 99·4 0·55 >99·9
STI positivity 1016 6·0 7·2 0 10·0 0 6·4 99·8 0·49 >99·9

LASSO, least absolute shrinkage and selection operator; STI, sexually transmitted infection; MSM, men who have sex with men; PrEP, preexposure prophylaxis; PPV, positive predictive value; NPV, negative predictive value. High risk scores were defined as a predicted probability of an incident HIV diagnosis within three years of 0·20-0·99%, and very high risk scores as a predicted probability of ≥1·0%.