Article Text
Abstract
Background For over 20 yearss sexually transmitted infections (STIs) have been studied as risk factors for HIV infection. We show how the design of studies in this field has inherent biases due to inappropriate confounder definition, coinfection, and heterogeneity in exposure to HIV. We use herpes simplex virus 2 (HSV2) to illustrate these biases and show that use of a serodiscordant couple design can remove them. Such findings are timely given the interest in using HPV vaccination for HIV prevention based on similar data.
Methods We developed an individual based model (IBM) using published data from an existing STI IBM (STDSIM). This model permits simulation of multiple cohort studies to show the direction and magnitude of these biases. The model is written in Matlab. Analyses were performed using Cox regression in SAS.
Results We identified four causes of bias. (1) While confounding by sexual behaviour is widely appreciated it is less well understood that perfect measurement of sexual behaviour will not permit adequate control of confounding as data on the frequency of HIV exposure is missing. We show that this bias can result in upward confounding. 2) As HSV2 is more infectious than HIV we expect HSV2 to be acquired from coinfected partners first followed by HIV. 3) As coinfection increases HIV viral load HSV2 infection may act as a proxy for a partner's elevated infectiousness with HIV. Both of these mechanisms result in upward bias, the magnitude of which depends on the prevalence of coinfection. 4) Between subject heterogeneity in the risk of disease has been shown to attenuate estimates for any risk factor. We show that this bias can result in significant attenuation of the HR and that it depends on the prevalence of HIV among subjects' partners and their sexual behaviour. We show that if HIV serodiscordant couples are enrolled all four biases can be removed see Abstract P1-S4.04 Table 1.
Conclusions The standard design is affected by at least four biases that preclude causal interpretations of all such HSV2-HIV studies performed to date. Use of a serodiscordant couple study design can remove these biases. It is impossible to correct previous results as the biases are not all in the same direction and their magnitudes depend on the unknown prevalence and transmissibility of both HSV2 and HIV among partners. These findings are expected to generalise to other STI-HIV risk factor studies and can help inform the decision to test HPV vaccination as an HIV prevention measure.