Evaluating concurrent partnership data from the 2005-2008 demographic and health surveys (DHS)
Presented by Martina Morris (United States).
M. Morris1, A. Leslie-Cook2, S.J. Nelson3
1University of Washington, Sociology and Statistics, Seattle, United States, 2University of Washigton, Statistics, Seattle, United States, 3University of Washigton, Epidemiology, Seattle, United States
Background: There is growing interest in determining whether concurrent sexual partnerships play an important role in the hyperepidemics of HIV in countries in Eastern and Southern Africa. Data from the DHS archives are being used to evaluate this question. This study examines the validity of DHS estimates of the point and cumulative prevalence of concurrency.
Methods: We examined data from every country with partner-specific questions for completeness of questionnaire design and levels of missing data. We evaluate the recommended algorithm for estimating the cumulative prevalence of concurrency in cases where the date of last sex questions are not asked for two of the three partners, using data from more complete surveys. We evaluate the UNAIDS recommendation for measuring the point prevalence of concurrency 6 months prior to interview by comparison with estimates from 1 to 12 months prior, checking both for trends and truncation bias.
Results: All DHS datasets are missing elements recommended by UNAIDS for measuring concurrency properly. The algorithms currently used to deal with missing date information underestimate the cumulative prevalence of concurrency by 20 - 80%, depending on country (Fig 1a). The UNAIDS recommended point prevalence measure of concurrency at 6 months prior to interview is systematically below the point prevalence measured in months 1-5 (Fig 1b).
Conclusions: Cumulative concurrency prevalence estimates from DHS surveys with missing date information are not valid, and recommended point prevalence estimates appear to be systematically biased. Inferences based on these data to assess the relationship between concurrency and HIV epidemic severity are likely to be misleading.
Download the e-Poster
Back to the session -
Back to the Programme-at-a-Glance