Articles / Bad stats

Bad Stat of the Month: HIV infection

AIDS is a horrible disease. My family has lost at least one dear friend to it, and the world has lost millions of loving, creative people. It also affects a significant number of transgender people around the world. AIDS needs to stop, and I’m heartened by recent news of medical advances that help people to live full lives with HIV infections, and even suggest the possibility of a cure.

In order to stop the spread of AIDS and help save the lives of people infected with HIV, we need to know what’s going on. We need the best information possible, and we need to avoid overreaching and making unjustified assumptions. That’s why I’m frustrated by April’s Bad Stat of the Month: Worldwide burden of HIV in transgender women by Stefan Baral, Tonia Poteat, Susanne Str√∂mdahl, Andrea Wirtz, Thomas Guadamuz and Chris Beyrer at Johns Hopkins, the Karolinksa Institutet and the University of Pittsburgh.

As I said with the inaugural Bad Stat of the Month, I’m not happy to be doing this. I’m sure that Baral and his associates are all caring professionals who really want to make a difference in the fight against AIDS. I believe that they also want to help transgender people, and they think they’re doing that by spreading these figures. I hope that this post will convince them that this strategy does more harm than good, and that they should concentrate their efforts elsewhere.

So here’s the headline that came blazing across my feed reader earlier this month: Transgender Women 49 Times More Likely To Have HIV, Study Says. “It doesn’t seem like it’s been a priority for global funding entities to care about the needs of transgender communities,” Baral told the Huffington Post’s Meredith Bennett-Smith. Bennett-Smith continues, “Going forward, Baral said he hopes health care providers and advocates will improve the way they target transgender populations and tailor treatment systems and support networks.”

Fair enough. I want global funding entities to improve the way they target transgender populations, because much as I want them to care about my needs, as a middle-class HIV-negative white American I don’t need their funding. What I’m afraid of is that the research of Baral and his associates will obscure that fact and other important patterns in the data, making it more difficult for them to target populations, exactly the opposite of their intentions.

How does this study obscure these patterns in the data? There are multiple ways, and I could probably write a post a day for a month. Tonight I’m going to focus on one specific weakness in the methdology, that I’ve discussed many times before. Baral and his colleagues pooled data from 39 studies across fifteen countries. I’ve looked at a few of these studies, and they all use non-representative samples. It is well known that a non-representative sample cannot be reliably generalized to the population at large, and many of these studies warn against such overgeneralizations. Baral and his co-authors simply ignored these warnings, stating in their summary, “Our findings suggest that transgender women are a very high burden population for HIV and are in urgent need of prevention, treatment, and care services.”

For example, one of the studies that Baral et al. used was a study by Kristen Clements-Nolle, Rani Marx, Robert Guzman and Mitchell Katz of the San Francisco Department of Public Health, who write, “The primary limitation of our research was the use of non-probability sampling. Our findings may not generalize to other urban areas, and there may be threats to internal validity if certain sampling methods were more likely to recruit high-risk individuals. Most traditional random sampling approaches would not produce reliable samples, however, because the transgender population has strong privacy concerns and has never been counted, and because many transgender persons are marginally housed or homeless.”

Despite this warning, Baral et al. went right ahead and generalized Clements-Nolle et al’s results, not only to other urban areas, but to the entire United States, and with a certain weighting, to the world.

As I said, there are more problems with the meta-study, but I’ll have to save those for later posts. I’ll also talk about the implications of these problems for trans people and people at risk for HIV.

It's only fair to share...Share on FacebookShare on Google+Tweet about this on TwitterShare on Tumblr


  1. I agree with much of what you say here–these are pretty bad stats! But we did appreciate this in writing a seriously hefty limitations section which I include at the bottom of the post. And you are right that this manuscript is as much advocacy as it is science. Ie, that highlighting that ignoring the needs of trans communities as it pertains to surveillance, dedicated interventions, support, etc is putting many at risk. Something has to change…and this was our limited contribution to the evidence base arguing for that change.

    There are several limitations with the approach
    used for this meta-analysis. There is probably limited
    generalisability of pooled estimates to represent the rates
    of all transgender women in a country, especially in the
    countries where only small studies have been done.
    Traditional sampling methods such as time-location
    sampling, a method through which the study population
    is sampled randomly from within a sampling frame of
    times and venues such as brothels or clinics for sexually
    transmitted infections, might result in oversampling of
    transgender women who are sex workers or report any
    transactional sex or transgender women who are seeking
    medical care related to a sexually transmitted infection or
    HIV.73,74 Sampling biases could result in overestimation of
    the actual HIV prevalence in all transgender women in a
    country. To address this improved sampling frames are
    needed, such as is being done with the 2011 census in
    Nepal that allowed formal registration of third gender
    people.75 Studies have previously analysed HIV preva lence
    between transgender women who are sex workers and
    those who were not.6 For this article, we noted that unless
    the studies exclusively targeted transgender sex workers,
    the proportion of transgender participants with a history
    of sex work was often not described, rendering subgroup
    analysis by sex work impossible. Transgender women
    who have undergone medical and social tran sition might
    assimilate into the general population and not identify
    themselves as transgender. These women could be less likely to be accrued into epidemiological or prevention
    studies on transgender populations. Inclu sion criteria
    requiring biological testing for HIV excluded a substantial
    number of studies where self-reported HIV prevalence
    was provided. However, in view of the low coverage of
    HIV testing and potential social desirability bias, objective
    evidence of HIV infection was deemed necessary. Thus,
    only data from publications and reports where methods of
    sampling and testing were described in detail were
    included in the analysis. Moreover, pooling hides the
    substantial intracountry spatial variation of the burden of
    HIV in large countries such as Brazil. Random-effects
    models were used to partly address the substantial
    heterogeneity of the HIV prevalence results included in
    the meta-analysis since these are studies from diff erent
    populations of trans gender women completed across
    diff erent settings and contexts. Although a random-eff ects
    model for meta-analysis was deemed more appropriate,
    this approach does tend to equalise the weight of studies
    of diff erent size and precision to the pooled estimate.76

  2. Thanks for your comment, Stefan! Can you say more about which HIV funding agencies you think could have the biggest effect on HIV in trans communities, and how?

Comments are closed.