Evaluation of multi-assay algorithms for cross-sectional HIV incidence estimation in settings with universal antiretroviral treatment

Table 3 Performance of multi-assay algorithms for identifying individuals with recent HIV infection

	Classification	LAg + VL MAA^a	LAg + BR MAA	LAg + PepPair MAA
Seroconverters infected < 1 year (N = 219)	MAA recent	58	80	55
Seroconverters infected < 1 year (N = 219)	MAA non-recent	160	139	164
Non-seroconverters infected > 1 year (N = 4376)	MAA recent	11	65	71
Non-seroconverters infected > 1 year (N = 4376)	MAA non-recent	4365	4311	4305
Non-seroconverters infected > 2 years (subset, N = 3483)	MAA recent	6	42	51
Non-seroconverters infected > 2 years (subset, N = 3483)	MAA non-recent	3477	3441	3432
Sensitivity^b		26.5%	36.5%	25.1%
Specificity^b		99.7%	98.5%	98.4%
Positive predictive value^b		84.1%	55.2%	43.7%
Negative predictive value^b		96.4%	96.9%	96.3%
False recent rate^c		0.2%	1.2%	1.5%

The table shows the performance characteristics for three multi-assay algorithms (MAAs) for identifying individuals with recent HIV infection (infected < 1 year). The cohort included 219 persons infected < 1 year (seroconverters) and 4376 persons infected for > 1 year (non-seroconverters). Sixty-three (28.8%) of the seroconverters and 3205 (73.2%) of the non-seroconverters were virally suppressed (viral load ≤ 400 copies/mL)
^aOne participant did not have a viral load result and could not be classified as recent vs. nonrecent by the LAg+VL MAA
^bSensitivity, specificity, positive predictive value, and negative predictive value were calculated classifying recent infection as <1 year since the last negative HIV test
^cThe false recent rate indicates the proportion of participants infected > 2 years who were classified as recently infected

ISSN: 1471-2334