Unlike the United States, the UK has bothered to track HIV infection among large, representative samples of drug injectors (both current injectors and those in methadone and other treatment programmes) since close to the start of the epidemic. Part of this effort involved testing anonymous samples of left over blood for HIV — the samples were usually taken from treatment or diagnostic purposes and are stripped of all but the most basic demographic and risk information (age, sex, length of time injecting, recent needle sharing) before being tested with HIV. The results, shown separately for London and the rest of England and Wales are shown below.

(Click to enlarge)

Yes, prevalence for the whole of England and Wales (including London — Scotland has its own system and reports separately) has risen by over 77% in the last decade. But still, fewer than one injector in 60 is infected with HIV. If you draw the graph using a normal percentage scale, you’ll see something close to the true level of infection — still too high, of course, but not exactly an overwhelming prevention failure when compared with data from any city or country that doesn’t have needle exchanges. The graph compares what happened in the UK with what happened in Jakarta, just because I happened to have the Indonesian data handy. But it would look just the same with data from Bangkok or Moscow or even New York in the years before the city (with no help from the federal government) began to hand out needles.

(Click to enlarge)

Since I’m just a numbers nerd and obviously don’t have a great visual imagination, perhaps someone could help me out here: how can you conclude from these pictures that safe injecting programmes fail to prevent HIV?

Leave aside the fact that they used Respondent Driven Sampling (RDS), a fashionable but deeply unsatisfactory method that could more honestly be named GSS — Glorified Snowball Sampling. [An aside for nerds: if you want to know why RDS is a bad idea even for drug injectors read this fabulous paper by Greg Scott. You've just got to love the young lady who has organised a training scheme to help non-injectors fool study staff into letting them participate. Ah, the land of enterprise...] Leave aside the fact that they inexplicably excluded from analysis the 8% of the sample who knew they were HIV infected. (This is the agency that invented Positive Prevention, for God’s sake. They ought to be supremely interested in the ongoing risk behaviours of people who are already infected.) Leave aside the three year time lag between collection of routine public health surveillance data and publication of the results –a delay that borders on the unethical. Even if those problems didn’t exist, you’d still have almost nothing of interest to write home about, because the analysis is so uninformative.

The headline figure is needle sharing — close to a third of injectors who think they are HIV negative used a needle that might already have been used by another injector at some time in the preceding 12 months. We’re talking about smack and crack users here, people who inject toxic substances into their bodies perhaps as often as three times a day. In most countries, we ask injectors to think back over the last week, and even then we get some dodgy results. How US injectors can remember what they did 11 months earlier I can’t imagine.

Then there’s information about sex. For some reason, they only report vaginal sex, although anal sex between men carries a far higher risk of HIV infection in the United States than any other activity. We’re told, wide-eyed, that 62% of injectors had unprotected vaginal sex in the last year, and that 42% had more than one partner in that time. But we don’t know if the sex with multiple partners was protected, we don’t know whether the unprotected sex was with injecting partners. Despite the CDC’s pronouncement that “a substantial proportion of IDUs are at risk for acquiring HIV infection through their sexual behavior in addition to their drug use”, we actually have no way of knowing whether the sex added to their risk or not.

I could rant on with more examples of missed opportunities for useful analysis, but I guess I’ve made my point. I do want to congratulate CDC, though, on their “honesty box”. They don’t try to fudge data loss and missing values:

“During May 2005–February 2006, a total of 13,519 persons were recruited to participate; of these, 1,563 (11.6%) were found ineligible, and 46 had missing recruitment information. Among the remaining 11,910 participants, data for 1,609 were excluded: 881 who already knew they had HIV infection, 334 whose data were lost during electronic upload, 288 initial participants (whose responses were excluded as part of the respondent-driven sampling methodology), 68 who could not be identified as either male or female, and 38 who gave responses with questionable validity.”

Interesting that CDC doesn’t allow for transgender injectors. Interesting that they provide “technical assistance” on HIV surveillance to countries such as Cambodia and Vietnam who have in the past made a far better fist of the work than CDC does at home.

Last week the Centers for Disease Control and Prevention (CDC) suddenly declared that around 55,000 Americans are getting infected with HIV each year (the estimate for the most recent year, 2006, is 56,300). That’s a whopping 40% more people infected each year than CDC previously thought. BUT, says CDC, it doesn’t mean the epidemic is getting worse. For all the years that CDC has trotted out the “40,000 new HIV infections this year” line, and that’s as many years as I can remember, there have actually been another 15,000 or so that didn’t hit the estimates.

In an interview with the wonderous Kaiser Family Foundation, CDC’s HIV epi-boss Kevin Fenton told us not to worry.

“I would like to reassure listeners that the new methods that we have used to calculate incidence are very different to the methods that we use to calculate prevalence and we should not try necessarily to impute one for another.”

Hmm. Because HIV is incurable, there’s a really straightforward relationship between incidence (new infections) and prevalence (total numbers infected at any given point in time). Add up all the incident cases, subtract all the deaths in infected people, and you’re left with the prevalence cases. If you’ve undercounted incidence by 15,000 cases a year for 10 years, then there are 150,000 more people infected with HIV than you previously thought. (Unless people have been dying like flies and you haven’t noticed.) That’s not imputation, that’s arithmetic.

CDC’s last substantial HIV prevalence estimates for the US were based on data up to 2003; it has promised new estimates for a while now, but hasn’t delivered yet. I sympathise with staff who have to try and make national estimates from the mish-mash of data served up by surveillance systems that differ by state (promoters of decentralisation in developing countries take note!). For the record, the new national incidence estimates were extrapolated from data from just 33 states; areas such as California and Washington DC, which contribute disproportionately to the HIV epidemic, do not provide the sort of data that CDC’s Atlanta-based epi-nerds need. No surprises, then, that when the new prevalence estimates do come out they are unlikely to be accompanied by systematic estimates of the number of people in each of the categories at highest risk: gay and bisexual men (who accounted for over half of new infections in 2006), drug injectors and Black Americans with several sex partners. China produces risk population estimates and HIV estimates for every district in the country. So does Indonesia. India, the other country that ranks with the US in the world’s four most populous is trailing, but working on it. Why does the US lag in some of the basics of risk surveillance and HIV prevention planning? (Why it lags in prevention is easier to understand. Even if systems to track risk behaviour and measure infection were delivering the goods, it would be hard to do much about that risk with the money available for prevention: just 4% of federal spending on AIDS within the US went to prevention programmes in 2008.)

CDC’s incidence estimates were broken down only by sex (73% male), risk behaviour (53% gay or bisexual men) and ethnic group (45% black), but not yet by combinations of the above. It is clear that the US has dropped the ball on prevention among gay men. (“Our data suggests that we are seeing sustained increased in HIV incidence in men who have sex with men,” Fenton said). It is equally clear that HIV among Black Americans has been hideously neglected. More on the epidemic in Black America tomorrow.

But a lot of the new HIV in Ireland appears to be imported. The HPSC knows the nationality of 120 of the new cases: 42% of them are in people from sub-Saharan Africa. Only 15% of the heterosexually transmitted HIV was in people born in Ireland, compared with well over 50% of newly identified infections in gay men and drug injectors.

Imported HIV is behind the rise in heterosexual infections reported in the UK, too. Headline writers who don’t look carefully at the data tend to seize on the fact that heterosexual HIV is rising in Britain, but in fact 81 percent heterosexual cases diagnosed in Britain are actually air-freighted in, mostly from Africa, while among gay guys, nearly two thirds is acquired at home. This is another indication that we need to look at our information carefully when planning what to do: Ireland, and Britain too, clearly need to keep their prevention programmes focused on the gay men and drug injectors who are at highest risk of getting infected at home.

Let’s be clear — the lower figures in India are the result of better data, not good prevention programmes. They represent a triumph of surveillance and spreadsheets, not a triumph of clean needles and condoms. Essentially, India has been looking for its epidemic in all the wrong places. Against the advice of just about everybody, including some of its own top public health officials, the Indian government has for years stuck stubbornly to an HIV surveillance system which is entirely inappropriate to its epidemic. It has been measuring infection rates in pregnant women and STI patients rather than in the sub-populations most likely to be infected: drug injectors, sex workers, their clients, and men who have anal sex with one another.

It has taken a hugely expensive survey of over 102,000 households to get closer to the truth. What the Dandona’s don’t mention is that the bare bones of a more sensible surveillance system focusing on high risk populations has been built up over the past two years by the Avahan project, with funding from the Bill and Melinda Gates Foundation. This has provided a much more detailed picture than was previously available of levels of HIV infection and risk behaviour in those most likely to be exposed to the virus, but it only covers a handful of states. India would do well to expand these efforts. This would allow the country to keep better track of its epidemic, and, most importantly of all, to do a better job of providing HIV prevention services to the people who need them most, rather than scattering its efforts across populations who were never at any great risk of infection in the first place.