By Rian Malan

Business Day (South Africa) 2 Nov. 2001

SA in a unique position to determine accuracy of AIDS data and help rest of Africa

This is dedicated to Robert S Cooper, a wise US epidemiologist who has done service in west Africa, attempting to quantify the suffering caused by various diseases in regions where administration is weak and recordkeeping barely existent.

"The idea that the sum of humanity's misery is now calculated and published is seductive," he concludes. In Cooper's estimation, disease statistics for sub-Saharan Africa are often just "guesstimates", and potentially dangerous ones at that. Numbers drive policy, and if the numbers are wrong the policies are liable to harm, rather than help, desperate people.

Cooper wasn't writing about AIDS, but he would have been amused by the furore surrounding SA's AIDS statistics, generally presented as a conflict between the righteous (the Medical Research Council) and the stupid (Thabo Mbeki's government). The truth is considerably more complicated.

Nobody knows with absolute certainty how many South Africans die as a result of HIV infection. The statistics we read in newspapers are computer projections based almost entirely on sample surveys at a small number of antenatal clinics.

Once a year, researchers remove left-over blood samples from these sites and have them screened for traces of the HI virus. The results are entered into computer models which use knowledge of "the well-known natural course of HIV infection" to produce estimates. If so many pregnant women are HIV-positive, the formula says, then a certain percentage of all adults and children must be infected, too.

The trouble with such modelling exercises is that they are capable, as UNAIDS chief epidemiologist Bernard Schwartlander recently phrased it, of producing "complete bullshit". Consider, for instance, the staggering array of AIDS death estimates for the year 2000 presently on offer in SA. Preliminary estimates from Stats SA put the toll at 109000. The Metropolitan Doyle model says 120000. The ASSA 2000 model, used by the MRC, says 164000. An earlier ASSA model says 190000. The software used by UNAIDS says 305000 or thereabouts, and the United Nations Population Division predicted AIDS deaths in the region of 400000.

Which of these numbers is right? It seems wise to have a debate about AIDS estimates and factors that influence their accuracy. To my mind, the most crucial of these are the accuracy of blood tests and that of death registration figures.

When President Mbeki raised questions relating to blood test accuracy last year, he was laughed at. "These tests are 99,99% accurate," French researcher Francoise BarreSinoussi told the SABC. This may well be true in laboratory settings, where the tests are applied in batteries, with each positive result confirmed as many as five times. What happens if you use just one test, unconfirmed by anything?

This is the rule observed in the antenatal screenings on which SA's AIDS statistics are based one test only. Evidence from elsewhere in Africa indicates that the results may not be entirely reliable. In Uganda the "positive predictive value" of a single test has proved to be in the region of 66%. When French army researchers in Mozambique subjected hospital patients to a single test, between 40% and 70% of the positives were false.

In its report, the MRC praises government for making "extensive efforts" to improve all forms of statistics-gathering. Stats SA concurs. At issue is when this improvement took place. The MRC believes completeness of death registration surged from 54% in l990 to 85% in l996, and has since remained stable.

Stats SA finds this illogical, pointing out that the efforts apparently responsible for this huge improvement were conceived in l996 and fully implemented only in 1998.

Here's why this is important. Birth registrations are up 36% over the past five years, even though the real-life birth rate is declining. Death registrations are up 24% in the same period. The MRC attributes the rise in deaths largely to AIDS. Government believes that rising registration is a factor. This is a legitimate dispute.

Probability of transmission, or PT, estimates the likelihood of acquiring the AIDS virus in a single unprotected sexual encounter with an HIV-infected person. This is measured by long-term studies of couples who are "HIV discordant" he has it, she doesn't, or vice-versa. In the west, the probability of transmission between monogamous heterosexuals has been pegged at about one in 1000. In Africa, PT was presumed to be very much higher, but the true figure remained "not known" until April, when The Lancet published the continent's first PT study.

AIDS researchers were in for a surprise: PT was one in 909 for both sexes, seven times lower than the assumption on which the MRC has been operating. Change just this one parameter in the MRC's computer model, and the number of estimated AIDS cases falls radically. MRC chief Prof Malegapuru Makgoba has denounced such "tinkering" as nonsensical, on the grounds that the outcome cannot be reconciled with HIV prevalence figures derived from pregnancy clinic surveys. But what if those figures are skewed by unconfirmed testing?

The point is: in a world of limited resources, AIDS estimates are fraught with awesome consequence. If the highest estimates are accurate, we should heed the calls of AIDS activists and devote every available cent to combating the disease. But what if the numbers are wrong? Our government might wake up five years hence to discover that billions have been squandered to save people from AIDS, who are more seriously threatened by poverty and ordinary disease.

It is against this backdrop that Stats SA and other government departments have queried the MRC report. An open-minded reading of the evidence suggests that government's position is reasonable, although not necessarily right. It may yet turn out that the MRC's estimates are on target, but we need to think carefully about what this will have proved. If the MRC is right, WHO/UNAIDS is wrong, and the most commonly cited figures for AIDS deaths in Africa might be inflated by at least a factor of two.

All told, there is much more at stake here than a denialist conspiracy to suppress the truth. SA is the only country on the continent that has the capacity to detect discrepancies between computer projections and recorded deaths. Let's stop vilifying each other, get our numbers right, and then export our knowledge to the rest of Africa.

Journalist Rian Malan has spent most of the past year researching African AIDS statistics for an article in Rolling Stone magazine.