CROI 2017 Abstract e-Book

Abstract eBook

Poster and Themed Discussion Abstracts

reach more than 300,000 clients during 2015. This study analyzed HIV prevention program data to estimate incidence trends over time and assess the effect of prevention services on seroconversion. Methods: Data were extracted from the SyrEx database developed by Alliance to monitor service provision. The dataset included data on all HIV tests from Jan 2010 to Dec 2015 performed in all HIV prevention programs in Ukraine. Unique client coding system allowed tracking of individual testing history. Clients who had two or more tests were included in incidence calculation. Seroconversion was defined as a positive test after a negative one. Person time calculated as time between first negative test and last negative (or midpoint between last negative and first positive tests) was distributed across calendar years spent between the tests. Exact Poisson confidence intervals were calculated. Results: From 568,194 individual clients who received at least one prevention service in 2011-2015, 58.8% had at least one HIV test. Of those, 42.8% had a negative first test and were tested at least once more and contributed person time. Overall five-year incidence rate per 100 person-years was 0.65 for PWID (95% CI 0.61-0.69), 0.48 for MSM (0.40-0.57), and 0.24 for CSW (0.20-0.30). Over five years, there was a significant declining trend among PWID, but no meaningful change among MSM and CSW. Detailed results are presented in the Table. Conclusion: Electronic programmonitoring tools, such as SyrEx, are becoming a useful source of strategic information on HIV epidemic. Incidence among Ukrainian prevention clients is low compared to data from other studies on key populations in Ukraine. The declining trend among PWID may reflect the impact of continuously high prevention coverage, whereas the stable incidence among CSW and MSMmay warrant improvements both in coverage and quality of services. Association between service utilization and seroconversion will be the subject of further analysis.

Poster and Themed Discussion Abstracts

885 IMPROVED EVALUATION OF HIV PREVALENCE ADJUSTING FOR INFORMATIVE NONPARTICIPATION Linbo Wang 1 , Erik VanWidenfelt 2 , Chloe Auletta-Young 1 , Nealia Khan 1 , Etienne Kadima 2 , Unoda Chakalisa 2 , Tendani Gaolathe 2 , Eric Tchetgen Tchetgen 1 , Kathleen Wirth 1 1 Harvard Univ, Boston, MA, 2 Botswana Harvard AIDS Inst Partnership, Gaborone, Botswana Background: HIV prevalence is routinely estimated from household surveys. Estimates may be biased if HIV-infected persons are less likely to participate. A design-based method leveraging interviewer features as instrumental variables (IV) for non-ignorable missing data can be used to account for such bias. A valid IV satisfies two conditions: (A) it is a strong correlate of participation and (B) it is not directly related to a subject’s HIV status. Using an IV, we examined the presence and magnitude of bias due to missing data on HIV status within a large randomized trial in Botswana. Methods: The Botswana Combination Prevention Project is an ongoing cluster-randomized trial evaluating the effect of a combination prevention package on HIV incidence. From 2013 to 2015, a random 20% sample of households in 30 communities was selected for participation. A household representative enumerated all members with their age, gender, residency status and relationship to head of household. Present and consenting members aged 16-64 years who were (or married to) a Botswana citizen completed a survey and submitted to HIV testing in the absence of evidence of positive HIV status. For each interviewer, we obtained data on years of prior work experience as an IV which we used to estimate upper and lower bounds of HIV prevalence consistent with (A) and (B). Under a third condition (C) that selection bias does not vary with interviewer years of experience, we obtained HIV prevalence estimates overall and by community. Results: A total of 15,475 eligible household members were enumerated by 58 interviewers. Median (25th, 75th percentile) years of prior work experience among interviewers was 3.1 (0.7, 8.0) years. Field staff consented and enrolled 12,610 persons. The most common reason for non-participation was refusal (10%), followed by inability to locate the person in the household (9%). Among the 12,570 subjects with a known HIV status, 3,596 (28.6%; 95% CI: 26.4%-31.0%) were HIV-infected. Under (A) and (B), the estimated lower and upper bounds of HIV prevalence were 29.5% and 35.4%. Under (A), (B) and (C), estimated HIV prevalence was 32.3% (95% CI: 30.1%-34.0%). Figure 1 presents bias-corrected prevalence estimates by community. We found empirical evidence for selection bias overall and in nine communities. Conclusion: HIV prevalence estimates which ignore non-participation may be downwardly biased. Investigators should consider including IVs in the study design to safeguard against non-participation-induced bias. 886 RISK- AND SYMPTOM-BASED SCREENING IMPROVES IDENTIFICATION OF ACUTE HIV INFECTION Maartje Dijkstra 1 , Godelieve J. de Bree 2 , Ineke Stolte 1 , Udi Davidovich 1 , Eduard Sanders 3 , Maria Prins 1 , Maarten F. Schim van der Loeff 1 1 Pub Hlth Service Amsterdam, Netherlands, 2 Academic Med Cntr, Amsterdam, Netherlands, 3 Kenya Med Rsr Inst, Kilifi, Kenya Background: Identifying patients with acute HIV infection (AHI) is important; (1) patients with AHI benefit from immediate start of antiretroviral therapy (ART), (2) early treatment of AHI could have a significant impact on the ongoing HIV epidemic, (3) patients who start ART during AHI may offer insight into the potential for post treatment HIV control. The recent development of point-of-care HIV RNA tests has made prompt diagnosis of AHI possible at time of care seeking. However, these tests are expensive and guidelines on whom to test for AHI are lacking.

CROI 2017 383