| |
Pregnancy May Not Protect Against HIV Disease Progression
|
| |
| |
There are 2 letters below in the April JID expressing concerns regarding the findings from a study published in JID suggesting that pregnancy does not accelerate HIV and may slow it down. At the bottom of this report is link to that study. Jules
Effect of Pregnancy on HIV Disease Progression during the Era of Highly Active Antiretroviral Therapy
CORRESPONDENCE
The Journal of Infectious Diseases April 1 2008;197:1075-1076
Pai-Lien Chen, Mark Weaver, Cynthia Kwok, and Charles S. Morrison
Family Health International, Research Triangle Park, North Carolina
Potential conflicts of interest: none reported.
To the Editor-A recent article by Tai et al. [1] described the relationship between pregnancy and HIV disease progression in women during the era of highly active antiretroviral therapy (HAART). The authors' results suggested that pregnancy was associated with a lower risk of HIV disease progression. Two possible reasons for this finding were provided: either (1) women who became pregnant during the study had a healthier immune status from the start or (2) HIV-infected women may have benefitted from an interaction between pregnancy and HAART. The discussion in the article, in conjunction with the accompanying editorial commentary by Dr. Anastos [2] in the same issue of the Journal, provides readers with the strong impression that the latter explanation is the more likely. For example, in her commentary, Dr. Anastos cited the study findings and stated, "Not only was pregnancy not found to be associated with more rapid progression of HIV disease, it in fact had a marked protective effect" [2, p. 971].
However, we have serious concerns about the study design and data analysis, and we believe that the study findings could be misleading. The authors treated pregnancy as a baseline exposure when, in fact, pregnancy was a time-varying exposure that occurred during study follow-up. This treatment resulted in 2 potential sources of bias, neither of which was adequately accounted for in the analyses. First, according to the results presented in table 1 in the article by Tai et al. [1], the women who became pregnant were younger, healthier (with a lower viral load and a higher CD4+ lymphocyte count), and stayed in the study longer than women who did not become pregnant. This observation provides clear evidence of selection bias. Second, although "nonpregnant" women were at risk for an event at any time during the study, the "pregnant" women were only at risk after they became pregnant. Figure 1 in the article by Tai et al. [1] shows that study end points were reached by more than 20% of the nonpregnant women before the study end point was reached for the first time in the group of pregnant women, providing evidence of such length bias. These results provide reasonable doubt regarding the possibility of a causal association between pregnancy and delayed HIV disease progression.
In the study by Tai and colleagues, the use of the propensity score approach to provide balance between the pregnant and nonpregnant groups at baseline, in terms of important prognostic factors for AIDS, resulted in pregnancy being treated as a baseline exposure variable. This is inappropriate, because pregnancy occurred during follow-up. Thus, we are not surprised that the inclusion of the propensity score in the Cox regression models (table 4 in [1]) had very little impact on the estimated effect of pregnancy. Alternatively, one should consider the use of other adjustment methods, such as risk set matching approaches [3, 4], in which each HIV-infected woman who was pregnant at time t is matched to another woman with a similar history of HIV infection up to time t but who was not pregnant.
In addition, the "before-pregnancy event" analysis that used one-to-one matched-pair data seems peculiar. There should have been few or no study end points for the pregnant group, because the event would have had to occur in the relatively short interval after the woman became pregnant but before the pregnancy event. The statistically significant result suggesting a reduced risk of prepregnancy events among women in the pregnant group provides further evidence that, from the start, women in the pregnant group were healthier than the women in the nonpregnant group. The analysis of an "after-pregnancy event" intuitively seems to be more reasonable. However, the analysis population no longer appears to allow a one-to-one match, because some nonpregnant women would have experienced AIDS-defining events (and thus would have been censored) before their paired partner experienced a pregnancy event.
As with most studies of HIV-infected persons, important information about the duration of HIV infection is unavailable in this study. Previous studies have indicated that HIV-infected women with greater immunosuppression have lower fecundity [5], so the study analyses could be confounded by the differing disease stages noted among women in the pregnant and nonpregnant groups.
Clearly, the issue of a possible association between HIV disease progression and pregnancy is important-especially for the millions of women living with HIV infection in sub-Saharan Africa. We believe that more research needs to be done to assess the relationship between pregnancy and HIV outcomes in the HAART era before it can be concluded that we have "good news for women living with HIV" [2]. Basing the conclusion that pregnancy provides a protective effect on a single, potentially flawed analysis could not only be misleading but harmful. Attempts by HIV-infected women to become pregnant in an effort to improve their own health could result in incidents of mother-to-child or woman-to-partner transmission of HIV that would not have otherwise occurred.
CORRESPONDENCE
Pregnancy and HIV Disease Progression: Methodological Concerns
Potential conflicts of interest: D.W. receives funding from an unrestricted educational training grant from the University of North Carolina-GlaxoSmithKline Center for Excellence in Pharmacoepidemiology and Public Health.
Daniel Westreich and Aaron Kipp
Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hil
To the Editor-We thank Tai et al. [1] for their recent article in the Journal. We read the article with great interest. The authors noted a substantial reduction in the risk of AIDS or AIDS-related death associated with pregnancy among HIV-positive women attending clinics in Tennessee. Although we recognize the substantial effort expended to control for confounding in this study, several methodologic issues urge caution in interpreting the otherwise welcome findings of these authors.
First, we are troubled by the definition and classification of exposure in this study. In these analyses, all person-time contributed by a woman who became pregnant at any point during follow-up-including person-time contributed both before and after the period of actual pregnancy-was considered to be time exposed. As a result, "exposed" person-time is, in fact, a mix of exposed and unexposed person-time. Furthermore, the unexposed person-time included as "exposed" time is in fact "immortal person-time," because these "exposed" women are necessarily alive and free of disease progression until the beginning of pregnancy. Although the authors note this latter concern in their discussion, and although this misclassification should not influence analyses in which follow-up began after pregnancy, other analyses may well have been affected.
An additional concern is the authors' decision to designate CD4+ lymphocyte counts obtained up to 365 days after the initial clinic visit as "baseline" measurements. CD4+ lymphocyte counts can change dramatically within a year of follow-up, depending on whether or not individuals are receiving antiretroviral therapy (ART); therefore, such counts are a potentially inaccurate measurement of the true CD4+ lymphocyte count at baseline.
An analytic approach that could help to address both of these concerns is to allow exposure and the CD4+ lymphocyte count, as well as other key confounders, to vary throughout follow-up. Because the CD4+ lymphocyte count and other confounders (including ART received) may vary with time, exposure, and each other, the use of marginal structural models (MSMs) may be necessary to obtain an unbiased estimate of effect. Regardless of whether MSMs are required, controlling for time-varying exposure and confounders may yield dramatically different results. For example, in a study analogous to that of Tai et al. [1], Hernan et al. [2] reported rate ratios of 2.3 (1.9-2.8) in a model controlling only for baseline confounders, but they reported rate ratios of 0.4 (0.3-0.5) in a model accounting for time-dependent confounding and 0.7 (0.6-1.0) in the appropriate MSM.
We also are concerned about the inclusion criteria used for the study. Given the authors' focus on the highly active antiretroviral therapy (HAART) era, it is not immediately clear why 75 women who received non-HAART ART and 144 nonpregnant women (23% of that population) who received no ART at all were included in the analysis. The beneficial effect of HAART with regard to prevention of AIDS events and death is well established; differences between the 2 populations may therefore be partly attributable to differences in ART received rather than to any effects of pregnancy.
A related issue is that the women who became pregnant and the women who did not become pregnant are different enough that they are not interchangeable-that is, comparisons between them cannot form the basis for valid causal inference. Table 1 in the article by Tai et al. [1] shows large, presumably clinically significant differences between the 2 groups with regard to a number of key variables known to affect time to development of AIDS; these variables include receipt of ART (see above), subject age (median age, 25 years for pregnant women vs. 36 years for nonpregnant women), and CD4+ lymphocyte count (median count, 450 cells/mm3 for pregnant women vs. 352 cells/mm3 for nonpregnant women). Similarly, in the matched analysis, only 81 (58%) of pregnant women were successfully matched to a nonpregnant woman, despite a wide CD4+ matching range (±100 cells/mm3), again suggesting limited exchangeability between the groups. It would be helpful, given the inclusion of a propensity score analysis in their report, if the authors could provide the propensity score histograms for the 2 study groups. In a qualitative sense, if these curves overlap substantially, then there is some evidence for exchangeability between pregnant and nonpregnant women; if the curves do not overlap, then the basis for counterfactual inference is more questionable.
We believe that the aforementioned concerns urge caution in the interpretation of the important findings of Tai and colleagues. However, we wish to emphasize that we are advocating skepticism, not cynicism; the remarkable consistency seen across subanalyses in this study urges deeper investigation of these issues.
Pregnancy Does Not Accelerate HIV In Women Good News for Women...
In lower-resource settings, pregnancy was associated with faster progression in some studies [8, 9], but it may have represented non-HIV pregnancy-related...
www.natap.org/2007/HIV/090407_09.htm
|
|
| |
| |
|
|
|
