AEGiS-15IAC: [LbOrC23] HOW WELL DO CROSS-SECTIONAL POPULATION SURVEYS MEASURE HIV PREVALENCE? EXPLORING THE EFFECTS OF NON-PARTICIPATION

15th International AIDS Conference

Bangkok, Thailand — July 11-July 16, 2004

[LbOrC23] HOW WELL DO CROSS-SECTIONAL POPULATION SURVEYS MEASURE HIV PREVALENCE? EXPLORING THE EFFECTS OF NON-PARTICIPATION

Int Conf AIDS. 2004 Jul 11-16;15:Abstract No. LbOrC23

B Zaba¹, M Marston¹, R Isingo², M Urassa², P D Ghys³
¹London School of Hygiene & Tropical medicine, London, United Kingdom; ²National Institute of Medical Research, Mwanza, Tanzania; ³UNAIDS, Geneva, Switzerland

BACKGROUND: Nationally representative population surveys in several African countries found lower HIV prevalence than expected from ante-natal clinic surveillance. Data from the Kisesa HIV cohort study in Northern Tanzania are used to explore possible bias due to non-participation of mobile adults in cross-sectional surveys.

METHODS: Participants in community-based serological surveys were offered free treatment for common illnesses, but no inducement for participating in demographic surveillance. Individual HIV status obtained in sero-surveys in 1994, 1997 and 2000 was linked to socio-demographic characteristics and migration histories obtained in 15 concurrent demographic rounds. Logistic regression was used to find the relationship between HIV status and age, sex, area of residence, marital status and mobility. HIV status of non-participants was deduced from past and future test results, or for those who had never had an HIV test, by predictions based on regression results.

RESULTS: Participation rates are lower for men, urban residents, un-married and mobile individuals. The table shows participants and absentees in serological and demographic surveys in 2000 and their observed and estimated HIV prevalence. In the sero-survey, prevalence in absentees with known HIV status is lower than in participants, but the opposite is true for demographic surveillance. Predicted prevalence for absentees who had never tested is higher than prevalence among participants, with a wider differential in demographic rounds. Relative risk of HIV for all sero-survey absentees is 1.0 (95% CL: 0.9-1.1), whereas for demographic round absentees it is 1.3 (1.1-1.5). The figure shows consistent under-estimation of prevalence over all 15 rounds: ~23% for males and ~12% for females.

	Serological survey Demographic surveillance
	Total	HIV infected	HIV prevalence	Total	HIV infected	HIV prevalence
Participants present at survey	6067	495	8.2%	4438	377	8.5%
Absentees with known HIV status	1576	79	5.0%	551	63	11.4%
Absentees never tested for HIV	4992		9.0% estimate	5788		10.1% estimate
Total eligible	12635		8.1% estimate	10777		9.5% estimate

CONCLUSION: Cross-sectional surveys with high absentee rates may under-estimate population prevalence due to higher HIV infection rates in mobile persons, but studies offering medical treatment to eligible subjects attract those who are unwell, compensating for this bias.

040711
LbOrC23