CROI 2018 Abstract eBook

Abstract eBook

Poster Abstracts

cohort by sex, gender and age. In the derivation set myalgia, fever and weight loss were significant in the multivariate model (Table 1). In the validation set the SDSS yielded AUC of 0.85(95%CI 0.77-0.93). A cut-off score of ≥9 was 71% sensitive, 97% specific, with positive predictive value of 76%, negative predictive value of 96% and diagnostic odds ratio of 65.1(95%CI 26.4-160.8). Conclusion: The SDSS accurately predicts for AHI in a cohort self-presenting for screening, and may inform allocation of diagnostic resources in settings that do not routinely test for AHI. Validation in other populations with different risk behaviors is needed.

adjusted by survey-specific FRR and an MDRI calibrated for the population’s subtype distribution, the HIV incidence estimates were similar to observed incidence and detected declines in incidence.

1002 PERFORMANCE VALIDATION OF THE SEDIA LAG ASSAY IN SOUTH AFRICAN BLOOD DONORS Eduard Grebe 1 , Marion Vermeulen 2 , Tinus Brits 2 , Ronel Swanevelder 2 , Genevieve Jacobs 2 , Michael P. Busch 3 , Alex Welte 1

Poster Abstracts

1 Stellenbosch University, Stellenbosch, South Africa, 2 South African National Blood Service, Johannesburg, South Africa, 3 Blood Systems Research Institute, San Francisco, CA, USA Background: Tests for ‘recent’ HIV infection have found significant application in major population-level incidence (and trend) estimation projects. Design and validation of such tests depends crucially on robust estimation of a Mean Duration of Recent Infection (MDRI), which is usually based on data with repeated observations of seroconverting subjects from studies with different primary goals, rendering the production of such estimates particularly difficult and expensive. Methods: The MDRI of a test is estimated by fitting a statistical model for the probability of testing recent as a function of time since infection. In the usual context of seroconverter cohorts, this is applied to multiple observations anchored to a ‘point estimate’ of infection time, based on interval censoring applied to diagnostic test data. Midpoints of seroconversion intervals are not appropriate to single observations at the time of first positive specimens from previously negative blood donors. We adapted the methodology to cater to single observations, even with very imprecise infection time estimates by averaging over unobserved (possible) infection times. Intervals were adjusted for the sensitivity of screening assays used. We estimated MDRI for the established Sedia LAg recency test, using data from the South African National Blood Service, at a range of discriminatory thresholds. We used data from 2,973 seroconverting repeat blood donors, overwhelmingly with subtype C infections and with a median inter-donation interval of 357 days. Reproducibility of the MDRI estimates was investigated using bootstrap resampling. Results: The figure shows estimated MDRI values, as a function of normalised optical density (ODn) recency discrimination threshold. The MDRI at the standard threshold of 1.5 is 189 days (95% CI: 161-215) and at a threshold of 2.0 is 257 days (95% CI: 223-282). Conclusion: The Sedia LAg assay’s performance in South African blood donors is similar to its performance in a previously-described multi-origin subtype C panel. These results provide context-specific recency test characteristics to support estimating incidence in first-time blood donors. The results demonstrate, for the first time, that data from repeat donors can be used to calibrate tests for recent infection. Specimens from seroconverting repeat blood donors are a potentially rich source of data on early pathogenesis and early infection biomarkers, as long as appropriate averaging is performed over uncertain infection times.

1001 CAN THE LAG-AVIDITY ASSAY MEASURE AN INCIDENCE DIFFERENCE IN EAST AFRICA? Oliver Laeyendecker 1 , Ronald H. Gray 2 , Kate Grabowski 3 , Steven J. Reynolds 1 , Anthony Ndyanabo 2 , Joseph Ssekasanvu 3 , Reinaldo Fernandez 3 , Maria Wawer 3 , David Serwadda 4 , Thomas C. Quinn 1 1 NIAID, Baltimore, MD, USA, 2 Rakai Health Sciences Program, Kalisizo, Uganda, 3 The Johns Hopkins University, Baltimore, MD, USA, 4 Makerere University, Kampala, Uganda Background: A combination of the Limiting-Antigen Avidity Assay (LAg- Avidity) with viral load (VL) >1000 copies/mL is being used internationally for cross-sectional estimation of population-level HIV incidence. However, the capacity of this method to measure a point estimate and a change in incidence over time has not been validated in an East African setting where HIV-1 subtypes A and D circulate. Methods: We analyzed longitudinal data for two time periods in the Rakai Community Cohort Study (RCCS) in Uganda. Between survey Rounds 12 and 13 (2006-2007 and 2008-2009) the observed incidence was 1.05/100 person years (95% CI 0.90, 1.23). Between Rounds 14 and 15 (2010-2011 and 2012-2013) the observed incidence was significantly lower, 0.66/100 person years (95% CI 0.52, 0.83), P<0.05. The performance of the current LAg-Avidity protocol, with mean duration of recent infection (MDRI) of 130 days and false recent rate (FRR) of 0% was compared to a subtype specific MMDRI and FRR calibrated by survey round. Results: Based on gp41 sequence data, the subtype proportion in the RCCS was 45% A and 55% D, and the subtype adjusted MDRI was 184 days. In Round 13 there were 9,973 subjects, of whom 1244 were HIV-positive, with 422 were on ART. Of the 742/822 remaining subjects with samples available for testing, 49 were classified as recent and the FRR was 1.1% (6/544). In Round 15 there were 6749 subjects, of whomwere 985 HIV-positive, with 423 were on ART. Of the 500/562 remaining subjects with samples available for testing, 52 were classified as recent and the FRR was 4.8% (16/332). Per protocol cross-sectional R13 incidence was 1.63 (95% CI 0.97, 2.30), and the R15 estimate was 2.55 (95% CI 1.51, 3.59), P<0.05. Both per protocol estimates exceeded the observed incidence. When using the MDRI adjusted for subtype and a round specific FRR the incidence estimates was 0.88% (95%CI 0.44, 1.33) for R13 and 0.67% (95% CI 0.00, 1.68) for R15, P=0.54, which were closer to the observed incidence. Conclusion: In this subtype A/D epidemic, the per protocol methods of LAg- Avidity +VL over estimated observed incidence and failed to detect the decline in incidence between the two time periods. In contrast, when the method was

CROI 2018 383