Since I reviewed, back in April, the paper by Ashraf, Field, and Lee on the effect of providing vouchers for injectable contraceptives to women in reducing unwanted pregnancies in Lusaka, Zambia, I had been worrying about the use of these modern, convenient, and reliable technologies in those parts of the world in which HIV is highly prevalent. My thinking was that if there is a causal relationship between adopting modern contraceptives and risky sexual behaviors (such as unprotected sex), then the same technology that reduces pregnancies can also increase HIV transmission. And, in areas with high HIV prevalence and among certain sub-populations (such as uninfected individuals in serodiscordant couples), it is not clear whether this trade-off favors promoting family planning over concerns with HIV.
It turns out, however, that I may have been right to worry for the wrong reason: injectable contraceptives might indeed be causing an increase in HIV transmission, but the channel is biological rather than behavioral. This paper (gated), published in The Lancet Infectious Diseases last month, suggests that women on hormonal contraceptives (mainly injectables) were twice as likely to become infected with HIV and also twice as likely infect their male partners with it.
Before we go any further, a few words about limitations of the study and its generalizability are in order – although I have to admit that the most media outlets that covered this story, such as the NYT or the NPR did a decent job of mentioning these. If I had one tiny quibble with the media coverage, it was the sensationalist titles, such as “Contraceptive Used in Africa May Double Risk of HIV.” This is because, one of the study’s limitations is that it uses a sample of serodiscordant couples that were recruited for another study on the effectiveness of suppressing HSV-2 in reducing HIV transmission (which, itself, did not have any effect). So, external validity is a concern because while serodiscordant couples are an important part of the overall HIV transmission, they are not the universe. For many people, the news that an injectable, such as Depo-Provera or a generic version, may increase HIV risk will have exactly zero impact on their lives because they are not worried about HIV. Put it another way, if we increase the use of these contraceptives by 10% in the general population, we will likely see a substantially smaller effect on HIV incidence than the rate suggested in this paper: the findings here are applicable to those partnered with (about 75% of who were married to) HIV-infected individuals.
Then there is the issue of observable and unobservable differences between those using contraceptives and those who are not. As the authors are using a study not designed to answer this question, they are limited in seeking exogenous sources of variation in the adoption of injectables, and make no attempt to do so (as an aside, the fact that biomedical researchers only rely on either conditional correlations or RCTs is strange. Tools that help with identification, such as instrumental variables or propensity score matching, have been available for a while now). They do control for some self-reported sexual behavior measures as well as biological markers (more on these below), but it is hard to rule out differences between these two groups that may independently cause differences in HIV seroconversion. This is especially true for the transmission from women to men, where the sexual behavior of the initially uninfected men (measured at follow-up) is significantly different among those with a partner on a hormonal contraceptive than those with a partner not using any hormonal contraception.
What gives the results plausibility, however, is the fact that the researchers have access to data in the infected partner on CD4 counts, HIV concentration in the plasma, and endocervical concentrations of HIV-1 RNA. Using these, they can show that HIV-infected women who are using hormonal contraceptives have higher concentrations of HIV in their genital tracts even though the HIV concentration in their plasma is not higher. These results hold even when the researchers control for CD4 counts. This potential biological mechanism of transmission from women to men supports the argument that there is something about the hormonal contraceptives causing these changes rather than selection bias (or behavior change) entirely responsible to these findings. The authors, in the discussion section, provide citations suggesting that hormonal contraceptives might have other physiological actions beyond pregnancy prevention, including possible bone-density loss and cervical cancer.
The question is what to do with these findings. NYT is reporting that the World Health Organization (WHO) has called for a meeting of experts in January to see whether it should add warnings to hormonal contraceptives. I found this part of the discussion, both in the paper and in the media, much more confused in contrast with the evidence emerging from the study that is carefully presented and qualified. Perhaps the first issue that emerges from this study is whether a further study is needed. While the evidence presented in this paper seems to be sufficient to cause substantial worry (and might well lead to health warnings, especially to certain sub-populations, in the short run), I don’t see how we can avoid having an RCT designed to dig deeper into this question. The authors say that such a study (i.e. random assignment of hormonal vs. non-hormonal contraceptives) might be difficult to implement. Sure, but random assignment is not the only tool we have available to us: we can consider an encouragement design, where women are randomly offered incentives to take up hormonal contraceptives compared to a control group with no incentives. The experiment in Zambia (cited above) increased uptake by 15 percentage points with a simple voucher to couples. A study design that included incentives, given directly (and individually) to women, who want to avoid becoming pregnant over the next few years might well increase take up by a lot more than that. Yes, this would mean that the sample size of the follow-up RCT would have to be much larger than an RCT that randomizes assignment, but it would also be much more practical. (Q: can we get injectable manufacturers to pay for the follow-up RCTs?)
Furthermore, as I mentioned above, perhaps the data used in this study can be combined with richer data on the characteristics of the individuals and the areas they live in (such as distance to clinics with a reliable supply of injectables, or data on exogenous injectable shortages, etc.) and re-analyzed to see whether the findings are robust: this would be much cheaper than the more than $30 million that would have to be spent on the RCT. If the size of the effect was smaller, the conclusions about future public health policy (even while waiting for the results of an RCT) might be different.
Second, while the authors are on firm ground when they suggest that their findings argue for policies to counsel women about the potential for increased HIV-1 risk with hormonal contraceptive use, it’s harder to agree with the call to combine contraceptive counseling with HIV counseling and testing. We have discussed here earlier that clinics being even perceived to require HIV testing for women in order to receive antenatal care may actually deter them from seeking such important care. The issues around combining HIV and family planning, individual vs. couples counseling (both for HIV as well as for FP), and the role of women’s empowerment, are complex enough to make most heads spin. We need much more research in this area.
Finally, it would behoove the researchers conducting the follow-up study to sort out the biological mechanism much more clearly. The answer may well lie in further technological change by developing easy-to-use hormonal contraceptives that do not increase the risk of HIV transmission. It would be a real shame if a follow-up study confirmed the results of the current study but still could not sort out whether (or how much of) the increased HIV risk was due to behavior change in the treatment group vs. biological reasons.
I had intended to cover a few more recent papers on family planning when I started writing, but they will have to wait for another post in the upcoming weeks.