The reliability of sexual partnership histories: implications for the measurement of partnership concurrency during surveys

Concurrent partnerships (CPs) can accelerate the transmission of HIV (and other sexually transmitted infections) in a population [1–4]. CPs have been described as the “key driver” of generalized HIV epidemics [5]. Long-term CPs (i.e., partnerships that overlap for months, possibly years) in particular, may play a crucial role in connecting the sexual networks that transmit HIV [6]. Despite a recent surge of interest in targeting CPs for HIV prevention [7], the evidence that they are an important risk factor of HIV transmission remains limited [8–12]. The lack of association between partnership concurrency and HIV infection in empirical investigations could in part be due to the poor quality of survey data on CPs. To estimate the extent of CPs, survey respondents are typically asked questions about their 3–5 most recent sexual partners, including the dates at first and last sex, as well as whether the relationship with a given partner is still ongoing. These questions are used to check whether relationship intervals overlap [13, 14]. Such data are potentially affected by large biases [15–19], but few studies have evaluated the impact of measurement errors on survey estimates of partnership concurrency [14, 20], particularly in sub-Saharan settings. Only Morris and O'Gorman [15] argued that “measurement errors introduce a slight positive bias in estimates of the prevalence of CPs, and a slight negative bias in the length of the overlap [between CPs]”. This pioneering study suffers from two major limitations however: first, it only considered the impact of date heaping and recall errors on survey reports of partnership dates, but it did not consider the likely hypothesis that partnership histories (on which estimates are based) may not be reliable; second, it hypothesized that reporting errors occurred at random among population members, in ways that are unrelated to parameters of partnership concurrency. If patterns of misreporting are associated with partnership duration (for example), biases in concurrency measurements are likely to be significantly larger than previously thought. We use sociocentric network data collected on Likoma (an island located in the northern region of Lake Malawi) to (i) assess the inter-partner reliability of partnership histories collected during sexual behaviors surveys, and (ii) test whether reliability varies with partnership duration and timing. In doing so, and contrary to Morris and O'Gorman's previous assessment, we show that biases in survey measurements of CPs are large and of unknown direction.