CROI 2020 Abstract eBook

Abstract eBook

Poster Abstracts

905 KEY POPULATION SIZE ESTIMATION IN NIGERIA: NOVEL APPROACHES TO SAMPLING AND ANALYSIS Anne McIntyre 1 , Andrew Mitchell 2 , Samuel Nwafor 2 , Victor Sebastian 3 , Amee M. Schwitters 3 , Julia Lo 2 , Ibrahim Dalhatu 3 , Mahesh Swaminathan 3 , Kristen Stafford 2 , Man Charurat 2 1 CDC, Atlanta, GA, USA, 2 University of Maryland, Baltimore, MD, USA, 3CDC Nigeria, Abuja, Nigeria Background: Nigeria has the fourth largest HIV burden globally. Key populations (KP), including female sex workers (FSW), men who have sex with men (MSM), and people who inject drugs (PWID), are more vulnerable to HIV than the general population owing to stigma and discrimination, and often have poor social visibility. Previous population size estimates (PSE) in Nigeria were based on programmatic mapping of hotspots with enumeration of KP at venues. The results failed to account for KP who were not present at venues, resulting in underestimates of population sizes that also lacked precision. Reliable PSE are needed to guide focused and appropriately scaled HIV epidemic response efforts for KP. We used novel approaches for sampling and analysis to calculate PSE in Nigeria. Methods: We used three-source capture-recapture (3S-CRC) to estimate the size of KPs in seven states in Nigeria (October–December 2018). Hotspots were mapped just before 3S-CRC sampling. We independently sampled FSW, MSM, and PWID 3 times approximately 1 week apart. During encounters at KP hotspots, distributors offered inexpensive and memorable objects to FSW, MSM, and PWID that were unique to each capture round and KP. In subsequent rounds, participants were offered an object and asked to describe those received during previous rounds; we tallied correct identifications of the object. Distributors recorded responses on tablets using REDCap™ software and uploaded data to a secure central server. Data were aggregated by KP and state for analysis. Median PSEs were derived using Bayesian nonparametric latent-class models with 80% highest density intervals (HDI) for precision. Results: During three rounds of independent captures in each state, there were approximately 310,000 encounters in 13,899 hotspots. Table 1 summarizes median PSE by KP and state. Conclusion: We are the first to implement 3S-CRC to calculate median PSE with 80% HDI in Nigeria. Overall, our PSEs were larger than previously documented for each KP in each state. Empirical methods and analysis using Bayesian models that account for factors (i.e., social visibility and stigma) that influence heterogeneous capture probabilities may produce more accurate PSE. The large estimates suggest a need for programmatic scale-up to reach these populations with high HIV risk. 3S-CRC methods, in similar epidemic settings, could help estimate critical population denominator data needed to inform HIV prevention and treatment programs.

(documented negative HIV serology in prior year). The PIRC EDI was weakly correlated with the CD4 model EDI (R2 = 0.017) (Figure). Among the 159 (23%) PIRC participants with follow-up CD4 data for ≥1 year prior to starting ART, we also used the pre-ART CD4 to calculate the CD4 model EDI (i.e., sampled during chronic infection). The pre-ART CD4 EDI also was weakly correlated with the PIRC EDI (R2=0.00058). When using the PIRC EDI as the gold standard, the sensitivity of the CD4 model was 51% (95% CI 47%-55%) and specificity was 60% (95% CI 52%-67%). Conclusion: The CD4 depletion model did not correctly identify persons with incident infection and did not differentiate persons with incident and chronic infections. The CD4 depletion model is not an appropriate model for monitoring incidence trends among smaller sample sizes, such are likely to be represented within the highest burden EtHE jurisdictions. Alternative strategies are needed, including scale-up of objective measures of HIV incidence, to measure the primary outcome of the EtHE initiative.

Poster Abstracts

904 ESTIMATING INCIDENCE AT A REGIONAL LEVEL WITH THE CD4 DEPLETION MODEL Sanjay R. Mehta 1 , Michael E. Tang 1 , Christy M. Anderson 1 , Susan J. Little 1 1 University of California San Diego, La Jolla, CA, USA Background: The Ending the HIV Epidemic (EtHE) initiative targets a 75% decline in HIV incidence in 5 years and a 90% decline in 10 years. Currently HIV incidence in the U.S are derived from the Center for Disease Control’s CD4 depletion model. The EtHE initiative requires an understanding of HIV incidence at a regional and local level to evaluate the impact of prevention interventions. Here we examine the accuracy of the CD4 depletion model for measuring incidence in sub-epidemics. Methods: Using the San Diego Primary Infection Resource Consortium estimated date of infection (PIRC EDI) model as a gold standard (a model that estimates recency using the limiting-antigen [LAg] avidity assay in combination with viral load information which has an estimated false recency rate of 1%), we found that the sensitivity of the CD4 model was 51% (95% CI 47%-55%) and specificity was 60% (95% CI 52%-67%) (see abstract 1291). We used this to calculate the predictive values of CD4 recency testing in various epidemic scenarios. Results: Using the above estimates, we calculated the positive predictive value (PPV), negative predictive value (NPV), and the posterior odds (PO), for various proportions of incident infections, ranging from 5% to 50%, in a setting of 1000 newly diagnosed infections. For a test on a single individual, PPV ranged from 6.3% to 56.0%, NPV ranged from 95.9% to 55.1%, and PO from 0.67 to 1.28. Using a fixed proportion of 25% incident infections among all new diagnoses, we varied the size of the sampled population from 250 to 10,000 to evaluate the accuracy of the CD4 model in predicting the number of incident cases in different size epidemics. The estimated values were approximately 1.7 fold greater, ranging from 106.8 (95% CI 105.5 to 135.3) incident infections (true value 62.5) for a population of 250, to 4275 (95% CI 3850 to 4775) incident infections for a population of 10000 (true value 2500). Conclusion: Although the CD4 model is not designed to predict if an infection is incident at an individual test level, the uncertainty in this test also impacts population scale estimates. As interventions to prevent HIV transmission are scaled up as part of the EtHE effort, we need more accurate estimates of incidence that can be applied at smaller population scales, so that we will be able to measure the impact of our outcomes.

906 5GEOSPATIAL HIV DYNAMICS IN FRANCE: A GRAVITY EFFECT MODEL Marie-Laure Chaix Baudier 1 , Benoit Visseaux 1 , Lambert Assoumou 1 , Marc Wirden 1 , Lot Florence 2 , Laurence Morand-Joubert 3 , Stephanie Raymond 4 , Laurence Bocket 5 , Constance Delaugerre 1 , Francis Barin 6 , Diane Descamps 1 , Davey M. Smith 7 , Antoine Chaillon 7 , for the ANRS AC43 study group 1 INSERM, Paris, France, 2 Santé Publique France, Saint-Maurice, France, 3 Saint- Antoine Hospital, Paris, France, 4 Toulouse University Hospital, Toulouse, France, 5 CHU de Lille, Lille, France, 6 Université François Rabelais, Tours, France, 7 University of California San Diego, San Diego, CA, USA Background: HIV epidemiology is constantly evolving and regular surveillance studies are needed to monitor the HIV genetic diversity shifts or global transmission patterns. Individuals during the primary HIV infection contribute disproportionately to the spread of the HIV epidemic due both to at risk behaviors and to high viral loads. Thus, monitoring HIV epidemiology in this population is crucial in tracking the leading edge of HIV epidemics. We explored

CROI 2020 339