Reading HIV/AIDS numbers

A Top Story at AIDSMEDS this week was:  “Tenofovir HIV prevention gel to be fast-tracked for FDA approval”. The success of the gel had been one of the most publicized highlights of the XVIIIth International AIDS Conference in July in Vienna: male-to-female HIV transmission had been cut by 39 percent.
Biased as I am to disbelief, I looked for reasons to cast doubts, and did not have far to look.
In the first place, of course, there is the question of what testing “HIV-positive” even means. Since the demography of “HIV” demonstrates that it isn’t infectious (see my book and many subsequent blog posts), any changes in apparent transmission need to be explained in other ways; however and unfortunately, the mainstream rarely reports the variables needed to construct such alternative explanations. One can then only deconstruct mainstream claims on their own terms. In the present instance, even the researchers acknowledge that their trial (CAPRISA 004) — a “proof of concept” study, not a full-fledged clinical trial of effectiveness — was too small, conducted in too few places, and had too low adherence (Quarraisha Abdool Karim et al., Science 329 [2010] 1168-74).
The gel was tested on 445 women while another 444 received a placebo gel. The total study cumulated to 1341 women-years, so an average of only about 18 months per woman. The “HIV infection” rates were 5.6/100 women-years using the tenofovir gel and 9.1/100 women-years using placebo. In other words, incidence rates of 5.6% and 9.1% respectively!
Even the study’s authors remark that this is extraordinarily high. Their only explanation, however, is that the study was carried out in the epicenter of the HIV/AIDS epidemic where prevalence rates are around 30%. But this entails absurd improbabilities: it would have taken only a bit over 3 years to reach that prevalence (9.1% being the normal rate, the control no-prevention rate in the study). Why is the prevalence then as low as 30%? These sexually active women had a median age of about 24 and a median age of “sexual debut” of 17.5; at 9% annual infection rate, why is the prevalence only 30%? And furthermore, why is there any population left alive, since this epidemic has been raging for more than a decade in absence of effective prevention or widespread antiretroviral treatment?
These numbers are simply not compatible with the theory that testing “HIV-positive” means the active presence of a deadly infectious agent.
The study also displays a “too good to be true” aspect in its details. The 889 enrolled women were randomized into treatment and control groups. Miraculously, those in the two groups behaved indistinguishably (dashed and continuous lines in the Figure below) as to use of condoms, adherence to use of gel, and even the number of sex acts per month and the rate at which this declined — from about 7.2 to about 3.1 — during the two years of observation:

That two groups of women should behave so similarly in 4 distinct respects seems most unlikely, doesn’t it? Would they also have such similar smoking and shopping habits, say?
And what about that marked decline in frequency of intercourse within a couple of years? The authors don’t discuss it. The pregnancy rate was 4%, far too low to explain such a drastic decrease in frequency of intercourse. The most obvious possible explanation, then, is that the women or their spouses found use of the gel — be it tenofovir-laced or placebo — so displeasing that they had sex less frequently. That would not augur well for the prospects of bringing something like this gel into general use.

Quite a few aspects of this study, then, do not inspire confidence, and seem at considerable variance from the ballyhoo that greeted the announcement of this preliminary and flawed little study. Though the plaudits and ballyhoo are substantively unwarranted, they are of course understandable from a psychological and sociological viewpoint, given that more than two decades have brought no success to mainstream ventures as to vaccines, microbicides, or effective and tolerable (non-toxic) treatment. Any slight sign of success is naturally welcomed with profuse and unrestrained overstatements.

On the other hand, the study’s introductory literature review is unexceptionable:
“The search for new technologies to prevent sexually transmitted HIV infection over the past three decades has had limited success. Only five of 37 randomized controlled trials, which tested 39 HIV prevention strategies, have demonstrated protection against sexual transmission of HIV infection . . . . The successful trials tested medical male circumcision . . . , sexually transmitted infection (STI) treatment in Tanzania . . . , and a HIV vaccine combination in Thailand . . . . Hence, HIV prevention technologies that women can use and control remain a pressing priority . . . . Over the last 20 years of microbicide research, none of the 11 effectiveness trials of six candidate products have demonstrated meaningful protection against HIV infection”.
Unexceptionable, that is, except for:
1. “Limited success” is euphemistic for “no success”.
2. The cited successes have not been replicated, and perhaps for that reason they have not been adopted, as any proven effective prevention surely would be.
3. Five successes out of 37 trials might well mean no actual successes. The criterion typically used, 95% confidence intervals or p ≤ 0.05, means that at least 1 in 20 apparent successes will be artefacts, occurring by chance; 2 successes in 37 trials would certainly be no better than chance. A Bayesian statistical approach, however, reveals that p ≤ 0.003 rather than p ≤ 0.05 is required to make the odds of being wrong only 1 in 20 (Matthews, “Significance levels for the assessment of anomalous phenomena”, Journal of Scientific Exploration, 13 [1999] 1-7).  So in 37 trials, one would reasonably expect significantly more than 2, perhaps something like 5 or even more, to be successful purely by chance, if the studies used the typical frequentist statistical analysis as a criterion rather than a Bayesian analysis. For a discussion of the flaws underlying frequentist p-value analyses that is accessible for non-specialists, I recommend very highly Robert A. J. Matthews, “Facts versus Factions: The use and abuse of subjectivity in scientific research”, European Science and Environment Forum Working Paper (1998), reprinted (pp. 247–282) in J. Morris (ed.), Rethinking Risk and the Precautionary Principle, Butterworth, 2000.