Open Access Research

The effect of participant nonresponse on HIV prevalence estimates in a population-based survey in two informal settlements in Nairobi city

Abdhalah K Ziraba12*, Nyovani J Madise3, Mwau Matilu4, Eliya Zulu1, John Kebaso5, Samoel Khamadi4, Vincent Okoth4 and Alex C Ezeh1

Author Affiliations

1 African Population and Health Research Center, Shelter Afrique Centre, 2nd Floor, Longonot Road, Upper Hill P. O. Box 10787, 00100, Nairobi Kenya

2 Department of Epidemiology and Population Health, Centre for Population Studies. London School of Hygiene and Tropical Medicine. 49-51 Bedford Square, London, WC1B 3DP, UK

3 University of Southampton, School of Social Sciences, Southampton, SO17 1BJ, UK

4 Kenya Medical Research Institute (KEMRI). P. O. Box 54840, 00200, Nairobi Kenya

5 Loma Linda University, School of Public Health, 24951 North Circle Drive, Nichol Hall 1410 Loma Linda, California 92350, USA

For all author emails, please log on.

Population Health Metrics 2010, 8:22  doi:10.1186/1478-7954-8-22

Published: 22 July 2010

Abstract

Background

Participant nonresponse in an HIV serosurvey can affect estimates of HIV prevalence. Nonresponse can arise from a participant's refusal to provide a blood sample or the failure to trace a sampled individual. In a serosurvey conducted by the African Population and Health Research Center and Kenya Medical Research Centre in the slums of Nairobi, 43% of sampled individuals did not provide a blood sample. This paper describes selective participation in the serosurvey and estimates bias in HIV prevalence figures.

Methods

The paper uses data derived from an HIV serosurvey nested in an on-going demographic surveillance system. Nonresponse was assessed using logistic regression and multiple imputation methods to impute missing data for HIV status using a set of common variables available for all sampled participants.

Results

Age, residence, high mobility, wealth, and ethnicity were independent predictors of a sampled individual not being contacted. Individuals aged 30-34 years, females, individuals from the Kikuyu and Kamba ethnicity, married participants, and residents of Viwandani were all less likely to accept HIV testing when contacted. Although men were less likely to be contacted, those found were more willing to be tested compared to females. The overall observed HIV prevalence was overestimated by 2%. The observed prevalence for male participants was underestimated by about 1% and that for females was overestimated by 3%. These differences were small and did not affect the overall estimate substantially as the observed estimates fell within the confidence limits of the corrected prevalence estimate.

Conclusions

Nonresponse in the HIV serosurvey in the two informal settlements was high, however, the effect on overall prevalence estimate was minimal.