Uncertainty in estimates of HIV/AIDS: the estimation and application of plausibility bounds
- 1Department of Infectious Disease Epidemiology, Imperial College London, Norfolk Place, London W2 1PG, UK
- 2Centers for Disease Control and Prevention, 1600 Clifton Road, Atlanta, GA 30333, USA
- 3UNICEF, 3 United Nations Plaza, New York, USA
- 4UNAIDS, 20 Avenue Appia, CH-1211 Geneva 27, Switzerland
- 5The Futures Group, Glastonbury, Connecticut, USA
- 6The East-West Center, Honolulu, Hawaii, USA
- Correspondence to: Dr P D Ghys UNAIDS, 20 Avenue Appia, CH-1211 Geneva 27, Geneva, Switzerland;
Objectives: To establish the accuracy of the country specific estimates of HIV prevalence, incidence, and AIDS mortality published every 2 years by UNAIDS and WHO.
Methods: We review sources of error in the data used to generate national HIV/AIDS and where possible estimate their statistical properties. We use numerical and approximate analytic methods to estimate the combined impact of these errors on HIV/AIDS estimates. Heuristic rules are then derived to produce plausible bounds about these estimates for countries with different types of epidemic and different qualities of surveillance system.
Results: Although 95% confidence intervals (CIs) can be estimated for some sources of error, the sizes of other sources of error must be based on expert judgment. We therefore produce plausible bounds about HIV/AIDS estimates rather than statistical CIs. The magnitude of these bounds depends on the stage of the epidemic and the quality and coverage of the sentinel HIV surveillance system. The bounds for adult estimates are narrower than those for children, and those for prevalence are narrower than those for new infections.
Conclusions: This paper presents a first attempt at a rigorous description of the errors associated with estimation of global statistics of an infectious disease. The proposed methods work well in countries with generalised epidemics (>1% adult HIV prevalence) where the quality of surveillance is good. Although methods have also been derived for countries with low level or concentrated epidemics, more data on the biases in the estimation process are required.