Malaria, HSV-2 and TB contribute significantly to the global burden of infectious disease, with high prevalence areas geographically overlapping with HIV endemic regions. Here, dual infection occurs commonly, making the role of co-infections and their treatment in HIV transmission salient questions. We identified 45 studies of the effect of these co-infections on HIV viral load (6 for malaria, 20 for HSV-2 and 19 for TB). We found significant increases in HIV PVL associated with acute malaria (0.67 log10
copies/mL; 95% CI: 0.15, 1.19), HSV-2 (0.18 log10
copies/mL; 95% CI: 0.01, 0.34) and active TB (0.40 log10
copies/mL; 95% CI: 0.13, 0.67). Overall treatment for malaria appeared to have limited impact on viral load (−0.25 log10
copies/mL, 95% CI: −0.59, 0.10), however when the study with less than 10 days of follow-up was excluded, viral load decreased significantly with malaria treatment (−0.37 log10
copies/mL; 95% CI: −0.70, −0.04). These results are consistent with an earlier study suggesting that the full impact of malaria treatment on HIV VL does not occur within 4 weeks post treatment [35
], Kublin and colleagues chose 8 weeks after malaria treatment as the follow-up time necessary to see a return of HIV viral load to baseline [36
], suggesting that a 9 day follow-up period would be too short. Suppressive treatment for HSV-2 produced significant decreases in HIV viral load (−0.28 log10
copies/mL, 95% CI: −0.36, −0.19), however there was no significant impact of either episodic HSV-2 treatment (−0.08 log10
copies/mL, 95% CI: −0.28, 0.11) or TB therapy (−0.03 log10
copies/mL, 95% CI: −0.19, 0.13) on HIV viral load. Collectively these studies suggest that co-infections increase HIV viral load, and that some of these effects may be reversed by treatment using current treatment regimen.
We found that TB treatment did not significantly reduce viral load, an unexpected finding because active TB was associated with a 0.40 log10 copies/mL increase in VL. There are at least two possible reasons for this finding. First, the long course of TB treatment (6 to 9 months) and variable follow-up times make comparisons challenging since both the outcome of interest (VL) and its main modifier (CD4) may change over the course of treatment. However, excluding studies with less than 6 months of follow-up did not change our results. Second, standard TB treatment regimens may not address the underlying mechanisms that trigger increased PVL. Assessing the impact of treatment on chronic co-infection will require careful study designs to account for modifiers of effect that change over time.
Our results add to those in a recent review of the impact of co-infection treatment [64
]. Our findings for impact of malaria and HSV-2 treatment support those of the review, but our TB results do not. The Modjarrad review included two studies of malaria treatment, two of TB treatment and 6 of HSV-2 treatment. In contrast, we identified 6 studies of malaria treatment, 12 of TB treatment and 8 of HSV-2 suppressive treatment, as well as studies of co-infection prior to treatment. There are several differences between the reviews, including the inclusions/exclusion criteria, methods of analysis, and data extracted. For example, the Modjarrad review required studies to include a control group to adjust for natural history changes in VL over time, but most of the additional studies included in our review were cohort studies conducted over relatively short periods during which VL would not be expected to change substantially in the absence of co-infection or intervention. We therefore included pre-post studies without control groups. Further, the Modjarrad review used the standardized mean difference (SMD) as the measure of effect, rather than the actual viral load, which tends to exaggerate differences between studies because the dimensionless scale of the SMD is not directly interpretable. Finally, for the two TB studies that were included in both this and the Modjarrad review [19
], the data extracted in the Modjarrad review used a subgroup of patients, while we included data from all patients.
Impact on HIV transmission
Although our review found only modest changes in PVL associated with co-infections and their treatment (0.2 log10
to 0.7 log10
for co-infections and −0.3 log10
to −0.4 log10
for treatment), modeling studies have reported that small changes in mean viral load at population level could translate to significant reductions in HIV incidence – a decrease in PVL of 0.3 log10
was estimated to decrease HIV transmission by 20% and HIV progression by 25% [65
]. Lingappa and colleagues produced similar estimates in their analysis of PVL among 108 genetically linked HIV transmission events, suggesting that a 0.3 log10
reduction in PVL would reduce HIV transmission risk by 25% [66
Fraser and colleagues proposed that HIVevolves to reach a balance between sufficiently high viral load for successful transmission with each exposure and sufficiently low viral load for a protracted asymptomatic period to maximize transmission opportunities [67
]. We posit that co-infections help HIV achieve that “sweet spot” viral load through immune activation [68
], increasing the cumulative time spent at the optimum viral load for both asymptomatic disease and HIV transmission. From a population perspective, this may accelerate HIVepidemic trajectory by increasing the aggregate person-time at viral loads at which the risk for transmission is sufficient to maintain the reproductive rate above 1 in the population. ART may be able to supersede other factors in forcing VL back below these levels. However, for the vast number of HIV-infected individuals who are in the pre-ART window or who do not have access to ART despite being eligible, interventions to prevent and treat co-infections may be an important strategy to tip the balance in favor of the host not only by limiting progression and delaying ART initiation, but also by maintaining community viral load at levels that limit transmission and help curb epidemic trajectory. It is noteworthy that the potential impact of co-infections on total community viral load over time is a function not only of the magnitude of the increase in HIV VL observed in co-infected individuals, but also of the prevalence of the infection among HIV-positives and the average duration or proportion of time during an interval such as one year that co-infection is likely to exist. When the results of our meta-analysis are examined from this perspective, the potential impact of HSV-2 co-infection on total community viral load over a year is likely to markedly outstrip that of malaria and TB, even though the increase in HIV VL in co-infected individuals with malaria is more than three times that observed with HSV-2, and the increase with TB is more than twice that with HSV-2.
Although treatment of co-infections has been demonstrated to delay HIV disease progression and ART initiation, it has yet to be shown to reduce HIV transmission in intervention trials. While an RCT of HSV-2 suppressive therapy found a 16% reduction in HIV progression as measured by CD4<200, ART initiation or death (HR 0.83 (0.71–0.98)), there was no effect on HIV transmission [47
]. This may be due to persistent immune activation, including HIV target cells in the genital tract as much as two months after ulcer healing on standard HSV suppressive therapy [71
]. Hence, while HSV-2 suppressive therapy decreases PVL sufficiently to help reduce the impact of HIV disease on infected individuals, further work is needed to fully understand the mechanisms by which HSV boosts HIV VL, better define the VL reductions needed to curb transmission, and design HSV treatment regimens that reliably achieve these targets. One recent model indicates that at least a 0.75 log10
reduction in HIV VL is needed to decrease transmission by 50% [66
]. Trials are needed to evaluate the impact of carefully redesigned treatment regimens for HSVand other co-infections on HIV transmission, as well as on the surrogate of HIV viral load. Evidence of an impact of TB and malaria on HIV progression is limited, with mixed findings for TB [72
] and no effect for malaria [74
]. In evaluating non-ART regimens to reduce HIV disease progression we suggest using outcomes as defined by Lingappa and colleagues: ART initiation, CD4<200 and HIV related death [70
Methodological quality of individual studies was variable. Several studies (especially of malaria and TB) were cross-sectional and thus particularly susceptible to bias. However, nearly all controlled for potential confounders such as CD4 and gender, and results were similar when restricted to these studies. Although we found no evidence of publication bias, we may not have had sufficient power to detect this. Studies included used different diagnostic methods and definitions for acute/active infection, which could affect our findings (although when examined explicitly for malaria [36
], no material difference was found in the results. Further, assays for measuring viral load and CD4 count also varied between studies and standardization of tests would likely narrow the confidence intervals.
We used a random effects model and found that heterogeneity was significant, which we would expect given variability in study design, populations, sample sizes and follow-up times. Follow-up times varied from 9 to 65 days post malaria treatment, 28 days to 24 months for HSV-2 therapy and 2 to 12 months for TB treatment. In addition, differences in treatment regimens contributed to the observed heterogeneity. Thus, our results are interpreted with caution, but provide strong stimulus for further research on the impact of co-infections on HIV viral load and transmission.
Our study focused on plasma viral load, a reproducible quantitative measure of HIV infectiousness. However, because HIV is sexually transmitted in regions bearing the greatest burden of the epidemic, it would be most relevant to determine the impact of co-infections on genital viral load. In a study among 2 521 African HIV serodiscordant couples, Baeten and colleagues demonstrated that higher GVL was associated with greater risk of HIV transmission – they found that a 1 log10
increase in GVL was associated with roughly a two-fold increased risk of HIV transmission (2.2 fold, 95% CI 1.60–3.04 for endocervi-cal swabs and 1.8, 95% CI 1.30–2.47 for semen) [76
]. To date, data on GVL are limited in studies of the effect of co-infections and their treatment on HIV viral load. Indeed, GVL was not measured in any of the malaria or TB studies that we reviewed. However, it is plausible that compared to systemic infections, sexually transmitted infections increase GVL more than PVL due to local inflammatory factors in the genital tract. A key area for future research is elucidating the HIV viral load dynamics between plasma and the genital compartment for both systemic and sexually transmitted infections that modulate viral load. Improved approaches to accurately and reproducibly measure GVL will help answer these questions.