Thursday, September 24, 2009

Recent AIDS vaccine trial results - cause for celebration?

I am all in favour of AIDS vaccine trials. It's pretty much clear to me that safe sex (not to talk about abstinence) just isn't happen in sufficient numbers and consistency as to prevent the pandemic from happily continuing on its destructive way. During the last year or two there have been off-the-record discussions among AIDS activists and clinicians about whether HIV preventive vaccine trials should be abandoned altogether, seeing that until now all trials ended up in utter failure. Instead people played with the idea that we should simply get on with large-scale HIV testing and with putting infected people immediately on AIDS medication. The rationale behind this idea is that it turns out to be the case that infected people who are on AIDS medicines become after a short period of time non-infectious (their viral load undetectable). Mathematical modelling suggests that the pandemic could be brought to heel within a generation or two by using this strategy. It turns out to be the case that this test-and-treat strategy is also the most efficient means to keep infected people alive and kicking. On my reading of the literature AIDS would turn from a terminal illness into a serious chronic illness that can be efficiently dealt with by means of medical care. The later you go on treatment the higher your risk to die of AIDS.

Anyhow, this was the world of AIDS until today, at least if you believed the media. As part of the self-fulfilling prophecy (it's just a matter of time till vaccine research will yield results) we're told in today's newspapers and news programs that a vaccine trial conducted in Thailand resulted in a 31% reduction of likelihood of infection. Most people, of course, do not read beyond the headline, and that is unfortunate. The 'successful' vaccine is a combination of two vaccines that crashed and burned previously. There is no mechanism that would explain why a combination of two failed vaccines makes a successful vaccine. The numbers seem to be speaking for themselves - 31%!!! - but do they? I doubt it. Here's the baseline as reported toward the end of the Los Angeles Times article: 'New infections occurred in 51 of the 8,197 given vaccine and in 74 of the 8,198 who received dummy shots. That worked out to a 31 percent lower risk of infection for the vaccine group.' The 23 additional infections in the placebo arm - out of 16400 overall participants - doesn't seem seem to be a figure that's statistically significant by any stretch of the imagination, no matter how hard the study's spin doctors (it cost 105 mio US$ to conduct the study) try to make it look otherwise. There might be good reason to undertake the same study with a much larger number of participants across the world, but until then, I doubt we have anything to celebrate at all.

For starters the number of people infected in both arms is fairly small (roughly 125 out of >16400, 51 in the active agent arm, 74 in the placebo arm - I doubt that is statistically significant, might be just a fluke). Anyhow, the thing is that it could possibly be said that all other things being equal it might have prevented some people from becoming infected. The trouble is, of course, things are not being equal. There's all sorts of behavioural changes among trial participants, with some behaving like they would have anyhow, some increase their risk taking behaviour, others reduce it (overall there's likely a reduction in risk-taking behaviour, but some participants would almost certainly increased their risk-taking behaviour). So it is possible that among the infected people (in both arms) are people who increased their risk-taking behaviour as a result of trial participation.

The whole story is also yet another example of science by press release... the actual study will only be published in late October 2009...


  1. They're giving us the numbers for the 2x2 table, so we could just test it for significance (e.g. here). I'm getting a two-tailed p-value of .039 or .048, depending on the test. So it's statistically significant at the .05 level, but not by much, and the 95% confidence interval on that 31% drop is very wide (2% - 52%).

  2. This comment has been removed by a blog administrator.


Note: only a member of this blog may post a comment.

Sokal on steroids: Why hoax papers submitted successfully to academic journals proves nothing (ok, little)

A trio of authors has, during a 12 months period, submitted - by their own account - 20 manuscripts to academic journals they broadly identi...