Research design determines what can be known about modes of HIV transmission

Abstract

We commend Becker and colleagues [1] for assessing one type of potential exposure to HIV, medical injections, rarely examined in HIV epidemiology in India and other developing nations. However, their interpretation of their results is at odds with the evidence they presented, and their research design severely limits the inferences about transmission that can be drawn from the study. Remarkable proportions of their respondents reported receiving medical injections in the past year (68%) and ever (95%). Becker and colleagues [1] observed a dose-dependent association between the number of medical injections received and prevalent HIV infection (e.g. adjusted odds ratio for 10+ versus no injections in the past year 2.3, 95% confidence interval 1.2–4.3). Clearly, this is a meaningful relationship, and it is consistent with other research in India [2,3]. However, Becker and colleagues [1] concluded that ‘… data from this study do not provide evidence for an association between HIV infection and medical injections’ (p. 1335). Becker and colleagues [1] dismissed the association by suggesting that it represents HIV-infected individuals seeking more medical care, and thus injections, as a consequence of HIV-related illnesses. Whereas this is possible [4], Becker and colleagues [1] presented no data to support the contention. In the absence of identified incident infections and measures of why respondents received injections at particular points in time, there is no empirical basis for interpreting the temporal order of this association. Given the prevailing assumption that most HIV transmission in India and other poor countries is through penile–vaginal sex, it is important for Becker and colleagues [1] to show results for analogous analyses of sexual behavior as a correlate of HIV infection. The appropriate benchmark with which to compare their results on medical injections is the association between penile–vaginal sex (ever in lifetime, frequency in the past year, and number of opposite sex partners in the past year) and prevalent HIV infection, including adjustments for the same demographic covariates used in their published analyses and the corresponding measure of medical injections. The measures of penile–vaginal sex should not reflect condom use or partner type, because their measures of medical injections did not capture the riskiness of those exposures or the settings in which they occurred [5]. As we have noted previously [6,7], many other types of parenteral exposures (e.g. dentistry, tattooing, scarification, injection drug use, surgery, blood transfusion, phlebotomy, etc.) are important to consider for HIV transmission; Becker and colleagues [1] apparently did not assess any of these. To our knowledge, no study on HIV epidemiology in India has yet included a thorough inventory of both sexual and parenteral exposures. Other barriers to understanding the modes of HIV transmission in India and sub-Saharan Africa are the lack of any systematic analysis of HIV contact tracing or risk networks (especially with respect to incident infections), investigation of the place and time dimensions of transmission, and sequencing of HIV DNA in relation to any of these kinds of data [8]. Without such fundamental evidence on the structure and context of transmission, it is not possible to determine the important modes of transmission with confidence.