Though the mechanisms by which HAART exerts its effect on HZ incidence are not yet clear, accumulating evidence suggests that HAART might have different effects on HZ incidence depending on how long HAART has been used. Many studies demonstrated that HZ occurs at a high level shortly after initiation of HAART [4–15], which might reflect dysregulated immune responses against pre-existing varicella zoster antigens at the beginning of immune restoration, rather than profound immunosuppression [4, 27]. However, like a few other observational studies [2, 16], we did not observe such a phenomenon in this study. Possible reasons for this discrepancy might include: (1). HAART users in our study had higher CD4+ T cell counts (372 cells/μl) before starting HAART; (2). In this short term analysis, our study might not have sufficient power to detect minor difference in HZ incidence due to relatively small sample size and rarity of HZ events; (3). Participants were followed every 6 months and the exact dates of HAART initiation could not be determined.
Since HAART became available in 1996, a steady decrease in HZ incidence over time has been observed among HIV-infected women in the WIHS coincident with their increasing access to HAART. Like many other studies [18–21, 28], ours showed a protective effect of long-term HAART use on HZ incidence. However, a statistically significant effect was achieved only when a HAART adherence level of ≥95% was attained. Thus, a further examination of the role of HAART adherence level might shed light on the lack of associations between HAART use and HZ incidence as reported by some earlier studies as no such information was reported . In addition, we found that the long-term effect of HAART on HZ incidence was mainly mediated through its effects in reducing HIV viral load and improving host immune status. Enhanced immune status and decreased shedding of varicella zoster virus antigen by HAART  might explain the protective effect of HAART on HZ incidence as observed in our study.
In this study, we found that acyclovir use was surprisingly associated with higher HZ incidence in univariate analysis. Possible explanation for this lies in that prior acyclovir use might be an indication of other herpes infections that were closely related to HZ incidence due to common cause - immunosuppression by HIV infection, rather than be used to prevent occurrence of HZ [30, 31]. In addition, lower overall QOL was also associated with higher incidence of herpes zoster, which is consistent with the role of psychological stress in the development of HZ .
Our study had some advantages over previous studies. First, we used a propensity score matching method to balance background covariates. In this way, the indication bias for HAART initiation was dramatically decreased. Second, previous studies focused on HAART use but did not take into account HAART adherence level. Our study showed the protective effect of HAART use was most clear when a ≥95% HAART adherence level was attained, suggesting that HAART adherence level should always be considered in future similar studies. However, some limitations of our study should also be mentioned. First, self-reported HZ was not documented by a health professional; however, a previous study has ascribed high validity to self-reports of HZ , suggesting that HZ misclassification in our study was likely to be very low. Second, we used self-reported HAART adherence as an exposure variable. However, this concern might be mitigated by the fact that the relationships between the WIHS participants and their interviewers have been built on trust and respect over many years, and previous WIHS research has shown that self-reported medication use is consistent with objective measures of HIV outcomes, such as CD4+ T-cell count, HIV viral load, and self-reported physical functioning . Lastly, our study only included women, and the results might not be generalizable to men.