HIV/AIDS: NUMBERS THAT DON’T ADD UP
Posted by Henry Bauer on 2007/11/29
UNAIDS recently decreased by more than 6 million its estimate of the number of HIV-infected people, putting it now at 33 million as against last year’s 39+ million. The estimated number of new infections was also lowered by 40%. (For useful commentary, see Science Guardian of November 20th.)
Media coverage failed to report clearly that the revision was only of statistically calculated estimates, not of the actual situation those numbers pretend to describe. Thus an editorial on November 25 in the Arizona Republic had the heading, “Turning the corner on HIV is inspiration to keep going”, and the optimistic comment that “The United Nations has revised its HIV estimates downward, correcting statistical flaws that, frankly, should have been addressed earlier. But that shouldn’t obscure the good news: a significant drop in new infections in recent years, especially in hard-hit sub-Saharan Africa. Efforts to fight HIV/AIDS have actually turned the corner. Now is the critical time to keep resources flowing, when it’s clear that prevention and treatment are paying off.”
But there had been no good news, just the bad news–for those who didn’t already know it–that UNAIDS’s numbers are not worthy of attention, let alone belief. In this latest revision, for example, the recalculated infection rate in sub-Saharan Africa for 2001 is given as 5.0% (4.6-5.5); in the 2004 version, the rate for 2001 had been given as 7.6% (7.0-8.5). Naïve consumers of numbers may imagine that when experts state a range like 7.0-8.5, that asserts with great confidence that the true value lies between those bounds. Yet three short years later, we are asked to have great confidence in a considerably lower range, 4.6-5.5, that doesn’t even overlap the earlier one. That should inspire great confidence in this conclusion: these experts do not know what they are doing.
There is no obvious reason to lend any credence to the latest numbers, and sound reason not to. Detailed descriptions of the technicalities of the computer models can make the head spin, but it takes no expertise to recognize that the estimates are an affront to plain common sense. The ranges of uncertainty attached to UNAIDS’s estimates are clearly nonsensical. Furthermore, UNAIDS estimates for the United States differ greatly from the data published by the Centers for Disease Control and Prevention (CDC).
* * * * * *
Computer models are used to project estimates, to make extrapolations beyond actual counts, because so few actual counts are at hand. One would expect the estimates to be most reliable for those regions where reporting of actual counts is most reliable: first because the computer models should be tested against actual counts, and second because every estimate has to use some actual counts as a baseline or starting point for any calculation. Yet UNAIDS and the World Health Organization publish estimates that are said to be better for Africa than for North America! Here, for instance, for people “living with HIV” in 2004:
Sub-Saharan Africa 23.4–28.4 million (25.9* ± 10%)
Western & Central Europe 0.48–0.76 million (0.62* ± 20%)
North America 0.54–1.6 million (1.07* ± 50%)
(* For convenience, the midpoints of the ranges and a single ± percentage are given here instead of the “most probable” values in the cited publication, so that a single number can describe the range of uncertainty instead of different percentages + and – respectively; cited from “The Global HIV/AIDS Vaccine Enterprise: Scientific Strategic Plan”, PLoS Medicine, February 2005, e25, based on data from UNAIDS and the World Health Organization).
How strange that the most accurate estimates, to just ±10%, are for sub-Saharan Africa, where reporting is acknowledged to be far from perfect; while in North America, where reporting is most highly organized and has been done for the longest period of time, the estimates are no better than ±50%.
I had written this a little while ago. The supposedly improved estimates in the “AIDS epidemic update” of December 2007 are, if anything, even less credible. For sub-Saharan Africa the estimated range of uncertainty is now even less, at 7.5%, while for North America the range of uncertainty is now even greater, at 60%!
Similarly incredible are the estimates of new infections (again for 2004):
Sub-Saharan Africa 2.7–3.8 million (3.25 ± 17%)
Western & Central Europe 14,000–38,000 (26,000 ± 45%)
North America 16,000–120,000 (68,000 ± 75%)
In school we were taught that, having completed a calculation, we should stand back and ask ourselves whether the result made sense–in case we had misplaced a decimal point, say. That advice seems not to have been taken by the experts at UNAIDS and WHO. These numbers are outrageous affronts to common sense. They sap any confidence one might otherwise still have that the people responsible for these estimates understand what they are doing. The media are culpable as well, for disseminating these numbers, typically reporting the “best estimates” without the enormous ranges of uncertainty, apparently oblivious to the clear evidence that these numbers should not be taken seriously.
It is even more ludicrous when it comes to deaths from AIDS (again, for 2004):
Sub-Saharan Africa 2.1–2.6 million (range is ± 11%)
North America 8,400–25,000 (range is ± 50%)
We are asked to believe that in the United States deaths from a reportable disease are known to no better than ± 50%, whereas for sub-Saharan Africa, where governance, administration, and infrastructure leave much to be desired, the numbers are known to within ± 11%!
Yet it gets even worse. CDC’s Surveillance Report for 2005 gives the number of AIDS deaths in the USA in 2004 as 17,453–accurate to a single death! Since the UNAIDS/WHO estimate of 8,400–25,000 is for North America, not just the United States, deaths in Mexico and Canada were apparently as few as NEGATIVE 9,053 [8,400-17,453] or as many as 7,547 [25,000-17,453].
For what’s wrong with many other aspects of officially disseminated HIV/AIDS numbers see, for instance, in The Origins, Persistence and Failings of HIV/AIDS Theory:
Estimates much higher than actual counts, p. 224;
Unexplained retroactive reduction by CDC of actually reported AIDS deaths, p. 221;
Number of HIV-positive Americans unchanged for two decades
during a supposedly spreading epidemic, pp. 1-2;
Poor performance of the computer models used by CDC, p. 223;
CDC increasingly disseminating estimates rather than actual counts, pp. 221-2;
Poor performance of the computer models used by UNAIDS/WHO, pp. 135-6 & 204-9;
—and these are far from the only instances.