Methods for estimating HIV prevalence: A comparison of extrapolation from surveys on infection rate and risk behaviour with back-calculation for the Netherlands

Objectives: To compare HIV prevalence estimates (total number infected) by using extrapolation from surveys on infection rate and risk behaviour (EIR) in specific segments of the population and back-calculation (BC) on reported AIDS cases. To discuss potential sources of bias and error, and to identify areas for improvement of the methodology. Design: Systematic comparison and epidemiological assessment of data input, underlying assumptions, and output. Methods: Low, possibly unbiased and high estimates of HIV prevalence as of January 1996 for homo/bisexual men, injecting drug users, heterosexual men and women with multiple partners, and blood transfusion recipients and haemophiliacs were derived from surveys and continuous data collections on HIV infection rate and risk behaviour in the Netherlands between 1992 and 1996. These were compared with estimates (point and 95% CI) by empirical Bayesian BC on AIDS cases 1982-1995. Results and conclusions: The estimate of HIV prevalence by EIR was 13,806 with low and high estimates of 9619 and 17,700, respectively. The HIV prevalence estimate by BC was 8812 (95% CI: 7759-9867). The available data from EIR are too limited for accurate estimates of HIV prevalence. EIR estimates could be improved considerably with more precise data on prevalence of risk behaviours and HIV prevalence rate for homosexual men. More confidence can be put in the BC estimates, but these could be underestimates because of the age effect on incubation time, pre-AIDS treatment and relapse of risk behaviour. BC estimates could be improved by a better representation of the incubation time distribution (including the effect of age thereupon), better data on the effectiveness and uptake of pre-AIDS antiretroviral treatment and prophylaxis of opportunistic infections, and on the level of underreporting.