Sunday, October 21, 2007

"HIV prevalence estimates: Fact or fiction?"

Read the full article by Kristen Jill Kresge in IAVI Report.

In recent years HIV prevalence estimates have been revised for many nations. The revised estimates, based on improved data, in almost all cases are lower than previous ones, sometimes dramatically so. "A few years ago UNAIDS estimated that 42 million people were HIV infected. Now the number stands just below 40 million, according to the 2006 Report on the Global AIDS Epidemic.....HIV prevalence estimates are generated by epidemiologists using HIV infection data from small subsets of the population that can be extrapolated using mathematical models. These models combine national population estimates and epidemiological data collected in a country and then churn out estimates of national HIV prevalence, based on a series of assumptions.......Previous prevalence estimates have been based primarily on sentinel surveillance data collected from pregnant women who visited antenatal clinics, one of the few settings where there is mandatory HIV testing.....'Data from antenatal clinics help monitor trends over time,' says Karen Stanecki, a senior advisor at UNAIDS in Switzerland. 'The intent [with data from pregnant women] is to monitor changes, not to predict the actual number of people who are infected,' says Prabhat Jha, professor of epidemiology at the Center for Global Health Research at the University of Toronto.....Following pressure from donor organizations to come up with more accurate prevalence estimates, more countries began conducting population-based surveys, often leading to a drop, sometimes precipitous......Now 30 countries have conducted population-based surveys to help better gauge the extent of their HIV/AIDS epidemics. In Benin, Mali, and Niger the results from these surveys were very similar to the figures estimated using sentinel surveillance data from antenatal clinics, but in the majority of cases the new figures were lower......Population-based surveys have several advantages—they reach more individuals in rural areas and include men. But they have disadvantages as well. 'The other side of the coin is that people may refuse HIV testing,' says Stanecki. 'This introduces a bias.' These household surveys are also limited to countries where there is a well-developed HIV/AIDS epidemic. "We don't recommend that they be conducted in countries with low-level prevalence," Stanecki adds. Population-based surveys are only applicable in countries where 1% or more of the population is HIV infected, which excludes many Asian countries where the HIV epidemic has not progressed as rapidly as in sub-Saharan Africa.

Comment: The collection of information requires resources, and resources are scarce. The resources needed to get good epidemiological data are quite scarce anywhere, and especially so in developing nations. Therefore those resources should be allocated so as to obtain the most important and useful information for decision making.

Sometimes, therefore, it is appropriate to allocate these resources to the collection of information that is very easily available, but only of limited utility.

Clearly it makes a lot of sense to test pregnant women coming into antenatal clinics for HIV infection, since medication of the HIV infected women is a cost effective means of prevention of transmission to the children to which they will give birth. Keeping track of this data is easy, and important for resource planning for future service delivery as well as for detecting trends that are relevent to the planning of prevention campaigns aimed at such women and their sexual partners.

Similarly, it is reasonable to test sex workers and other very high risk people for HIV infection, both as case finding for treatable disease and for monitoring the effectiveness of key prevention campains.

When making estimates of national prevalence and incidence of HIV (or any other disease), it makes sense to utilize a wide variety of information sources, including many sources that were designed (primarily) for other purposes.

It is important to realize that information also only has to be "good enough" for its purposes, not perfect. Does the difference between a worldwide prevalence of 42 million and 40 million make any difference in the decisions that are made relevant to HIV/AIDS? If not, either estimate is "good enough" for the global decision making. On the other hand, it may be important to know whether the incidence of HIV infection is increasing or decreasing globally, regionally, and nationally. If one is inferring incidence from changes in prevalence (combined with death rates), the need for accuracy in the prevalence estimates is greatly increased.

It is important, however, to understand how the estimates are made, and how and why the data was collected on which the estimates are based. Estimates of prevalence based on national sample surveys done specifically for the estimation of that national prevalence are different than those based primarily on extrapolation from clinical records of prevalence in specific target populations. The latter may be "good enough" for many practical purposes, especially if a nation deems it a poor use of epidemiological survey resources to make more precise estimates.

Governments do play politics with morbidity estimates, and it is sometimes important not to be fooled by a government official seeking to spin or even falsify that information. However, inaccuracy in estimates that were made with a reasonable allocation of available resources, and deemed "good enough" for their purposes by reasonable decision makers need not be meritricious. JAD

No comments: