Belinda Beresford
The true HIV infection rate may be closer to one in eight South Africans, rather than the one in nine suggested by a government study released this week.
Using statistical models from the Actuarial Society of South Africa (ASSA), researchers have estimated that about 5,3- to 5,4-million people in South Africa are carrying HIV, or about 700 000 people more than the latest government figures.
Minister of Health Manto Tshabalala-Msimang hailed the latest study as showing that the HIV/Aids epidemic in South Africa was being brought under control.
The government estimate of 4,7-million people infected with HIV is drawn from a yearly survey of women attending public antenatal facilities. It compares to 4,2-million for 1999.
The ministry says the latest figures appear to confirm that the rate of infection among pregnant women has been plateauing out since 1998. In that year the prevalence of infection was 22,4%, and rose to 22,8% in 1999.
Previously the infection rate appeared to have been following an exponential curve, but after a sharp rise in 1998 the growth has been slower, fuelling the government’s assertion that the rate has been flattening out.
But debate about the accuracy of the 1998 figure has led some statisticians to cast doubt on the government’s interpretation of the spread of infection. Government refusal to allow outside analysts access to the raw data from the antenatal survey, or to provide full details on how their calculation is done, has meant such concerns cannot be confirmed or allayed.
Professor Robert Dorrington, director of the Centre of Actuarial Research at the University of Cape Town and convener of the ASSA Aids committee, believes the 1998 estimate is too high and this brings into question the view that the infection rates are coming under control.
“One big problem is the 1998 figure. The department tends to look at the pattern over the last three years, but the 1998 figure is too high. This indicates we are still on an upward trend. If 1998 was wrong, then it is not certain what we are seeing … it depends on where the errors in the 1998 figure come from,” says Dorrington.
He agrees with the government that the rate of infection is indeed slowing down, but questions the degree. “We have certainly passed the exponential stage but certainly haven’t plateaued.”
This viewpoint is echoed by Dr Debbie Bradshaw of the Medical Research Council: “It is useful to build an awareness that an epidemic does flatten out, probably to do with the fact that people are starting to die off, and the pool of susceptible people is getting saturated. What’s critical is to know the new cases; at this stage monitoring the prevalence is not sufficient.”
The Department of Health warns about interpreting estimates for the general population on the antenatal survey data, acknowledging some of the problems inherent in the assumptions it makes.
The department survey was based on a sample of 16607 women from 400 sites around the country. Extrapolating these figures to the entire population gives an estimate of 4,7-million infected individuals. Of these the government study expects the majority (2,5-million) to be women, with about 110000 defined as “babies”.
However, there is contention about how effectively the sample data was extrapolated, particularly the assumptions that lie behind the analysis.
For example, the survey assumes that the prevalence of HIV infection among pregnant women is the same as among all women in that age group. This is open to debate, especially since it is known that women with HIV have declining fertility and so are less likely to attend antenatal clinics. This would mean that the study is likely to underestimate the prevalence of the virus among older women and hence among the population as a whole, since the estimates of male HIV carriers are drawn from the female numbers.
The department study assumes that the male infection rate is 85% that of the female that 85 men are infected for every 100 women. Dorrington says this ratio is probably fairly accurate, fitting with international and local studies of heterosexual transmission of HIV. However, he points out that over time the number of males tends to converge to that of the females.
Roughly 35% of the extra 700000 people estimated to have HIV using the ASSA model comes from adaptations to the analysis to account for this downward bias in the antenatal figures due to the lower fertility of HIV-positive women.
The 4,7-million number is only the “best” figure within a range of estimates that takes into account a 5% statistical margin of error. For last year the range or “confidence interval” in statistical jargon is from 4,1-million to 5,3-million. Thus the upper end of the confidence interval overlaps with the ASSA projections.
The variation in estimates indicates how difficult it is to plan for an entire population on the basis of antenatal surveys. The Department of Health has started implementing other surveillance mechanisms that should, over time, help refine the accuracy of its modelling.
The drawbacks of relying on statistics can also be shown in the government report in that only two provinces showed “statistically significant” increases in HIV prevalence. This means that these are the only two provinces where the increase falls outside the margin of error of the previous estimate and thus are highly likely to represent a change in prevalence.
While, for example, an increase of 1,1% in the Northern Cape may not be statistically significant, in real terms it still represents a greater number of people infected with the virus. It could be possible to have a string of non-significant increases over a number of years on a year-by-year basis. Yet since there is an upward trend in the figures the trend over the whole time period might be significant.
Although as precise information as possible is needed for the government and civil society to deal with the HIV/Aids epidemic, in one way the exact numbers are not that important. Bradshaw points out that the government figures drive home the size of the epidemic, even if the figures may not be totally accurate. “They are both very big numbers,” she says.