February 10, 2009
In this interview, Alexander Tsai, M.D., Ph.D., discusses the results of his study exploring the effect of antidepressant use on HIV viral load.
My name is Alexander Tsai. I'm a physician at Langley Porter Psychiatric Institute at UCSF [the University of California-San Francisco], and I work with Sheri Weiser, Maya Petersen, Kathleen Ragland and David Bangsberg, variously of San Francisco General Hospital, UC-Berkeley and Harvard Medical School. The poster presentation -- this research project -- is looking at the effects of antidepressant medication on antiretroviral adherence and HIV-1 RNA viral load.1
The impetus for the project basically stems from research that looks at how depression is known to worsen the course of HIV illness. The mechanism for that is unclear. It might simply just be due to worsened adherence to HAART [highly active antiretroviral therapy], but it might also be due to disturbances in the HPA [hypothalamic-pituitary-adrenal] axis that directly affect immune function.
Alexander Tsai, M.D., Ph.D.
Data for this study came from the REACH cohort, which is a project known as Research on Access to Care in the Homeless, on which David Bangsberg is the primary investigator. We have systematically recruited subjects for this research project to interview, from homeless shelters, free-meal programs and single-room-occupancy hotels in San Francisco. On a quarterly basis, they are administered structured questionnaires and blood collections. This gives us a rich set of data -- not only at baseline, but also on a quarterly basis -- looking at sociodemographics, alcohol and drug use, health services and utilization of health status.
For this specific research study, we only looked at the HIV-infected participants who were on highly active antiretroviral therapy. We had about 1,330 person-years of follow-up on 418 participants -- on average, about three years of follow-up on each participant.
The analytic conundrum when doing research in this area is that if you simply use a conventional method of adjusting, the crude estimate with no control for confounding is going to ignore the fact that depressive severity actually increases the probability of antidepressant medication treatment. However, controlling for baseline severity also ignores the fact that the worsening of depression increases the probability of antidepressant medication treatment. Finally, if you just control for time-updated values of depressive severity, you're going to ignore the fact that antidepressant medication acts through lowering depressive severity.
We used marginal structural modeling to estimate the effect of antidepressant medication on antiretroviral adherence in HIV-1 RNA viral load, while estimating for a host of demographic characteristics, alcohol and drug use, depressive severity, and other potential confounders. What falls out of our statistical methods is that, using the marginal structural model, the estimate of the effect of antidepressants on log viral load is -.8 -- which, if you compare to randomized trials, is not a huge effect size. In most randomized trials, you're probably looking at -2 logs or so.
What is also interesting is that, when we adjust for adherence, we find that the treatment estimate goes away. That suggests that the effect of antidepressant medication on viral load is acting through adherence, and not necessarily through some sort of other direct mechanism.
In other words, if they are on antidepressants, they're more likely to adhere to meds, and therefore their viral load will go down.
Yes. There's this outlying hypothesis that perhaps SSRIs [selective serotonin reuptake inhibitors] might have some sort of direct effect on neurotransmitters and immune function, but we're not seeing that in our project. We're really seeing that the effect on viral load is mediated primarily through adherence.
Have other studies shown that antidepressants could have a role?
From what I'm aware of, there have been one or two other observational studies hinting at the possibility. But nothing is really definitive. Neither is this definitive. You really need a randomized study to do that well.
OK. Thanks very much.
This transcript has been lightly edited for clarity.