Tuesday, February 03, 2004

How Prevalent is HIV in Africa?

It seems reporter Rian Malan, grandson of apartheid creator Daniel F. Malan and author of My Traitor's Heart, has been doing some investigative work on the way in which South Africa's HIV/AIDS statistics are being collected, and has come to the conclusion that the statistics being bandied about for Southern Africa are "grotesquely exaggerated". Rather than rely entirely on the Wired article (which, despite conceding Malan's point about the ASSA 2000 model, and further corroborating his criticisms by citing statistics about the results of a Kenyan AIDS survey, still manages to give an image of Malan as some sort of paranoid AIDS-denier), I thought it best to read his article myself. Following is some of what Malan has to say about the real situation:

It was an article from The Spectator describing the bizarre sex practices that contribute to HIV’s rampage across the continent. ‘One in five of us here in Zambia is HIV positive,’ said the report. ‘In 1993 our neighbour Botswana had an estimated population of 1.4 million. Today that figure is under a million and heading downwards. Doom merchants predict that Botswana may soon become the first nation in modern times literally to die out. This is Aids in Africa.’

Really? Botswana has just concluded a census that shows population growing at about 2.7 per cent a year, in spite of what is usually described as the worst Aids problem on the planet. Total population has risen to 1.7 million in just a decade. If anything, Botswana is experiencing a minor population explosion.

There is similar bad news for the doomsayers in Tanzania’s new census, which shows population growing at 2.9 per cent a year. Professional pessimists will be particularly discomforted by developments in the swamplands west of Lake Victoria, where HIV first emerged, and where the depopulated villages of popular mythology are supposedly located. Here, in the district of Kagera, population grew at 2.7 per cent a year before 1988, only to accelerate to 3.1 per cent even as the Aids epidemic was supposedly peaking. Uganda’s latest census tells a broadly similar story, as does South Africa’s.

Now, there are many criticisms that can be levelled against Malan's statements here, not the least important being that the sources of error in these projections he dismisses may have been some other parameters than the HIV prevalence rate. Nevertheless, Malan's statistics, if they can be trusted, do indeed suggest that something is very wrong with the models in use. Malan lays the blame on the use of a simulation called Epimodel:

In 1985, a science journal estimated that 1.7 million Americans were already infected, with ‘three to five million’ soon likely to follow suit. Oprah Winfrey told the nation that by 1990 ‘one in five heterosexuals will be dead of Aids’.

We now know that these estimates were vastly and indeed deliberately exaggerated, but they achieved the desired end: Aids was catapulted to the top of the West’s spending agenda, and the estimators turned their attention elsewhere. India’s epidemic was likened to ‘a volcano waiting to explode’. Africa faced ‘a tidal wave of death’. By 1992 they were estimating that ‘Aids could clear the whole planet’.

Who were they, these estimators? For the most part, they worked in Geneva for WHO or UNAIDS, using a computer simulator called Epimodel. Every year, all over Africa, blood would be taken from a small sample of pregnant women and screened for signs of HIV infection. The results would be programmed into Epimodel, which transmuted them into estimates. If so many women were infected, it followed that a similar proportion of their husbands and lovers must be infected, too. These numbers would be extrapolated out into the general population, enabling the computer modellers to arrive at seemingly precise tallies of the doomed, the dying and the orphans left behind.

Because Africa is disorganised and, in some parts, unknowable, we had little choice other than to accept these projections. (‘We’ always expect the worst of Africa anyway.) Reporting on Aids in Africa became a quest for anecdotes to support Geneva’s estimates, and the estimates grew ever more terrible: 9.6 million cumulative Aids deaths by 1997, rising to 17 million three years later.

Or so we were told. When I visited the worst affected parts of Tanzania and Uganda in 2001, I was overwhelmed with stories about the horrors of what locals called ‘Slims’, but statistical corroboration was hard to come by. According to government census bureaux, death rates in these areas had been in decline since the second world war. Aids-era mortality studies yielded some of the lowest overall death rates ever measured. Populations seemed to have exploded even as the epidemic was peaking.


In the year 2000, Timaeus joined a team of South African researchers bent on eliminating all doubts about the magnitude of Aids’ impact on South African mortality. Sponsored by the Medical Research Council, the team’s mission was to validate (for the first time ever) the output of Aids computer models against actual death registration in an African setting. Towards this end, the MRC team was granted privileged access to death reports as they streamed into Pretoria. The first results became available in 2001, and they ran thus: 339,000 adult deaths in 1998, 375,000 in 1999 and 410,000 in 2000.

This was grimly consistent with predictions of rising mortality, but the scale was problematic. Epimodel estimated 250,000 Aids deaths in 1999, but there were only 375,000 adult deaths in total that year — far too few to accommodate the UN’s claims on behalf of the HIV virus. In short, Epimodel had failed its reality check. It was quietly shelved in favour of a more sophisticated local model, ASSA 600, which yielded a ‘more realistic’ death toll from Aids of 143,000 for the calendar year 1999.

At this level, Aids deaths were about 40 per cent of the total — still a bit high, considering there were only 232,000 deaths left to distribute among all other causes. The MRC team solved the problem by stating that deaths from ordinary disease had declined at the cumulatively massive rate of nearly 3 per cent per annum since 1985. This seemed very odd. How could deaths decrease in the face of new cholera and malaria epidemics, mounting poverty, the widespread emergence of drug-resistant killer microbes, and a state health system reported to be in ‘terminal decline’?

But things get more interesting still, as model replaces model, with the number of AIDS deaths declining sharply with each revision:

Towards the end of 2001, the vaunted ASSA 600 model was replaced by ASSA 2000, which produced estimates even lower than its predecessor: for the calendar year 1999, only 92,000 Aids deaths in total. This was just more than a third of the original UN figure, but no matter; the boffins claimed ASSA 2000 was so accurate that further reference to actual death reports ‘will be of limited usefulness’. A bit eerie, I thought, being told that virtual reality was about to render the real thing superfluous, but if these experts said the new model was infallible, it surely was infallible.

Only it wasn’t. Last December ASSA 2000 was retired, too. A note on the MRC website explained that modelling was an inexact science, and that ‘the number of people dying of Aids has only now started to increase’. Furthermore, said the MRC, there was a new model in the works, one that would ‘probably’ produce estimates ‘about 10 per cent lower’ than those presently on the table. The exercise was not strictly valid, but I persuaded my scientist pal Rodney Richards to run the revised data on his own simulator and see what he came up with for 1999. The answer, very crudely, was an Aids death toll somewhere around 65,000 — a far cry indeed from the 250,000 initially put forth by UNAIDS.

There is a lot more to this article, but I've quoted more than enough of it already, so I'll just say that if one grants that everything Malan says is true, it gives grounds, not just for extreme scepticism about the scale of the AIDS crisis in Africa, but also for cynicism about all statistical modelling that is based on simple extrapolations of current trends. This is an issue I've mentioned before, but the issues Malan raises are as vivid a real-life illustration of what I was going on about as one can get. When one has researchers saying things like the following

"The nature of statistics is that we don't know," said Mary Crewe, director of the Centre for the Study of AIDS at the University of Pretoria. "Modeling is to some extent guesswork ... and in a way it doesn't matter if you're working on a figure of 10 percent or 20 percent of the population. It's still an appalling number of people who are dying."

one has a serious problem on one's hands, as the danger with crying wolf is that people stop believing you even when you're speaking the truth. Yes, a 10 percent figure is atrocious, but there is a difference between 10 and 20 percent, and it does matter. As Malan points out, AIDS is far from being the only preventable cause of illness and death in Africa, but diseases like malaria and tuberculosis get nowhere near the funding AIDS treatment does. It makes no sense to spend $400 on anti-retroviral drugs to keep a single person alive when 20 other lives could be saved with exactly the same amount of money.

Unlike Rian Malan, I'm not willing to attribute the tendency to overstate the HIV epidemic entirely to self-seeking advocates obsessed with their cause to the exclusion of all else - though I do think this is a major issue, as with all advocacy. I think Malan's numbers are at least partly wrong, and that there really is an incipient epidemic occurring in Southern Africa, even if not quite on the scale most sources have made it out to be. The danger with the sort of position Malan is pushing is that a lot of people who are bent on denying that there even is an AIDS problem (like, say, Thabo Mbeki) will take what he has to say as vindication of their beliefs, rather than as the criticism of simplistic extrapolations that Malan meant it to be, but even if this comes to pass, the blame will still have to rest primarily on the shoulders of those who relied on sloppy guesswork to make overblown claims. Scientists shouldn't be in the business of perpetuating falsehoods, even if they are doing so for what seem to be noble causes.