Evaluation of bias in HIV seroprevalence estimates from national household surveys

Abstract

Objectives:To evaluate HIV seroprevalence estimates from demographic and health surveys (DHS) and AIDS indicator surveys (AIS) for potential bias because of non-response and exclusion of non-household population groups.Methods:Data are from 14 DHS/AIS surveys with HIV testing, conducted during 2003–6. Blood samples were collected and analysed for HIV using standard laboratory and quality control procedures. HIV prevalence among non-tested adults was predicted based on multivariate statistical models of HIV for those who were interviewed and tested, using a common set of predictor variables. Estimates of the size of non-household populations in national censuses were used to assess potential bias because of their exclusion in the household surveys under different assumptions about proportion of adults and HIV prevalence in non-household populations.Results:Non-tested men had significantly higher predicted HIV prevalence than those tested in eight of the 14 countries, while non-tested women had significantly higher predicted prevalence than those tested in seven of the 14 countries. Effects of non-response were somewhat stronger in lower-prevalence countries. The overall effect of non-response on observed national HIV estimates was small and insignificant in all countries. Estimated effects of exclusion of non-household population groups were generally small, even in concentrated epidemics in India and Cambodia under the scenario that 75% of the non-household population was adults having 20 times greater HIV prevalence than adults in household surveys.Conclusions:Non-response and the exclusion of non-household population groups tend to have small, insignificant effects on national HIV seroprevalence estimates obtained from household surveys.

A Simultaneous Equation Approach to Estimating HIV Prevalence With Nonignorable Missing Responses

“…estimates is that the latter is the recommended approach for dealing with missing data in HIV research by UNAIDS/WHO, and is also very popular in the applied literature for dealing with data affected by missingness. As was found in previous research, imputation estimates are almost identical to those in Table 1 which were based only on observations without missing data (Mishra et al, 2008;Marston et al, 2008;Hogan et al, 2012;. Moreover, the imputation-based confidence intervals are, similarly, between 3 and 4 percentage points wide.…”

Women's and Men's Fertility Preferences and Contraceptive Behaviors by HIV Status in 10 Sub-Saharan African Countries

“…Finally, adults who participate in a household survey and consent to being tested for HIV may be selectively different than those who do not, and these unobserved factors may affect the associations observed in this study. Comparative analysis of DHS data from 14 countries found that men and women who did not get tested were predicted to have significantly higher HIV prevalence than those who did get tested in eight and seven countries, respectively, of the 14 analyzed, although not to a degree that would bias estimates of overall HIV prevalence from the DHS (Mishra, Barrere, Hong, & Khan, 2008).…”

The effect of changes in condom usage and antiretroviral treatment coverage on human immunodeficiency virus incidence in South Africa: a model-based analysis

“…For example, it may be unrealistic to assume that household prevalence surveys provide unbiased estimates of HIV prevalence in the general population. Although non-response bias appears to be minimal if it is assumed that recorded demographic and behavioural characteristics fully account for all variation in HIV prevalence [56], recent studies have suggested that individuals who do not participate in HIV prevalence surveys are more likely to be HIV-positive, independent of their demographic and behavioural characteristics [57][58][59]. The assumption that the antenatal bias parameter is constant over time may also be unrealistic.…”