HSV-2 as a biomarker of HIV epidemic potential in female sex workers: meta-analysis, global epidemiology and implications

This study investigated herpes simplex virus type 2 (HSV-2) seroprevalence utility as a predictor of HIV epidemic potential among female sex workers (FSWs) globally. We updated and analyzed a systematically-assembled database for paired HSV-2 and HIV seroprevalence measures among FSWs. The study identified 231 paired HSV-2/HIV prevalence measures from 40 countries. The pooled mean HIV prevalence using meta-analysis increased from 3.7% (95% CI 0.3–9.9%) among populations of FSWs with HSV-2 prevalence < 25% to 18.7% (95% CI 14.1–23.8%) among those with HSV-2 prevalence 75–100%. HIV prevalence was negligible in FSWs with HSV-2 prevalence ≤ 20% suggesting a threshold effect. Multivariable meta-regressions explained > 65% of HIV prevalence variation, and identified a strong positive HSV-2/HIV association. Compared to populations of FSWs with HSV-2 prevalence < 25%, adjusted odds ratios (AORs) of HIV infection increased from 2.8 (95% CI 1.2–6.3) in those with HSV-2 prevalence 25–49%, to 13.4 (95% CI 6.1–29.9) in those with HSV-2 prevalence 75–100%. HSV-2 is a strong predictor of HIV epidemic potential among FSWs. HSV-2 prevalence of 25–49% indicates potential for intermediate-intensity HIV epidemics, with higher levels indicative of large epidemics. HSV-2 surveillance could inform HIV preparedness in countries where HIV prevalence among FSWs is still limited or at zero-level.


Results
Search results and scope. The systematic search identified a total of 3386 citations, which after removing duplicates and screening, yielded 78 eligible reports (Fig. 1). Hand searching of the reference lists of eligible reports and reviews yielded three additional articles, and one comprehensive country-level public health report from India 25 that replaced three other full-texts [26][27][28] . Two reports were further excluded after consulting with Professor Rhoda Ashley-Morrow, an expert advisor in HSV-2 diagnostics, because the reliability of HSV-2 serologic testing could not be confirmed 29,30 . In total, 77 reports comprising 231 paired HSV-2 and HIV prevalence measures among FSWs, from 40 countries, were eligible for inclusion. These contributed to the database generated through our earlier systematic review 20 a total of 63 additional paired HSV-2 and HIV prevalence measures from 17 recent reports. Identified measures dated from 1988-2018 and are tabulated in Table S1 of Supplementary Information (SI) based on WHO regional classification [Region of the Americas (AMRO), African Region (AFRO), EMRO, European Region (EURO), South-East Asia Region (SEARO), and Western Pacific Region (WPRO)].
As the focus of this work is on examining the association between the two infections, it was pre-decided to restrict the analysis to settings where HIV has been introduced; we therefore excluded 37 paired measures with zero HIV prevalence from further analysis. After excluding measures with zero HIV prevalence, analysis was performed on a total of 194 paired measures from 33 countries (Fig. S1 of SI). India contributed the largest number of measures (n = 58; 29.9%), followed by China (n = 37, 19.1%), then Peru (n = 19; 9.8%). The distribution of measures across world regions is illustrated in Fig. 2A,B. The highest data contribution was for SEARO (n = 71; 36.6%), followed by AFRO and AMRO (each with n = 41; 21.1%), WPRO (n = 38; 19.6%), and lastly EURO (n = 3; 1.6%). There were only four studies from EMRO, all of which reported zero HIV prevalence, and thus were excluded from analysis.

Association of HSV-2 with HIV prevalence.
The multivariable models, whether considering HSV-2 prevalence as a categorical variable (Model 1) or as a linear variable (Model 2), both showed strong evidence for an association with HSV-2 and WHO region (p value ≤ 0.05). Some evidence for an association, that is a p value between 0.05 and 0.1, was found for consistent condom use, but no evidence (p value > 0.1) was found for year of data collection. Models 1 and 2 explained, respectively, 65.3% and 70.6% of the variation in HIV prevalence.

Discussion
Motivated by the concept of using current HSV-2 prevalence in a population as a proxy biomarker of future HIV prevalence in that population 8,9,20 and its relevance to HIV preparedness, this study assessed the utility of HSV-2 as a predictor of HIV epidemic potential among FSWs through a global systematic analysis of empirical paired HSV-2 and HIV prevalence measures. We found strong evidence for an association between HIV and HSV-2 prevalence, even after accounting for potential confounders such as region, temporal trend, and condom use (Tables 2 and 3). HIV prevalence was negligible at HSV-2 prevalence ≤ 20% (Fig. 2), but increased steadily with higher HSV-2 prevalence suggesting a threshold effect-the odds of HIV infection doubled with a 25% increase in HSV-2 prevalence (Tables 1 and 2). These findings demonstrate that in populations where HIV prevalence is still limited, but has potential to grow, HSV-2 prevalence can be used to provide a prediction of future HIV prevalence.
The hierarchy of HIV prevalence among FSWs was evident even in the context of Africa's general population HIV epidemics (Table 1). Outside the African Region, HSV-2 prevalence among FSWs of 25-49% was indicative of the potential for intermediate-intensity HIV epidemics with an HIV prevalence of ~ 5% or less. For FSW populations with HSV-2 prevalence ≥ 50%, HIV prevalence was higher and often exceeded 10%. Our findings based on analysis of empirical data substantiate mathematical modeling analyses predicting quantitatively such an association 8,9 , which also appears to exist for other populations 20 . The modeling analyses simulating HSV-2 and HIV propagation along diverse sexual networks demonstrated that HSV-2 prevalence ≥ 50% is indicative of substantial sexual risk behavior, sufficient to sustain large HIV epidemics in a sexual network 9 . In contrast, HSV-2 prevalence < 20% in a sexual network is indicative of low sexual risk behavior that is not likely to sustain an epidemic (a "threshold effect") 8 . Both of these modeling predictions were confirmed in the present study through analysis of actual empirical data (Table 1).
After decades of virtually zero HIV prevalence 2 , EMRO has recently seen emergence of HIV epidemics among FSWs in several countries 5 . However, and despite an apparently rapid epidemic growth, HIV prevalence in FSWs remains overall at low levels 5 . It is unfortunate that there were too few HSV-2 prevalence measures among FSWs in this region to predict HIV epidemic potential (Table S1 of SI) 24 . Available measures indicated also relatively low HSV-2 prevalence, often below 20% (Table S1 of SI) 24 , the apparent threshold for a significant Table 1. Results of meta-analyses on studies reporting HIV prevalence among female sex workers stratified by HSV-2 prevalence levels. CI, confidence interval; HSV-2, herpes simplex virus type 2. a Excluding 37 studies with zero HIV prevalence. b Meta-analysis not possible for a single study. c Q: the Cochran's Q statistic is a measure assessing the existence of heterogeneity in effect size (here, HIV prevalence) across studies. d I 2 : a measure assessing the magnitude of between-study variation that is due to differences in effect size (here, HIV prevalence) across studies rather than chance. e Prediction interval: a measure estimating the 95% interval of the distribution of true effect sizes (here, HIV prevalence).  (Fig. 2). HSV-2 prevalence in the general population in EMRO also appears to be low, and overall lower than that in other regions 8,31 . Indeed, a recent global assessment 32 estimated HSV-2 prevalence among women in the general population at 7.6% in EMRO, 9.6% in SEARO, 10.7% in EURO, 14.6% in WPRO, 24.0% in AMRO, and 43.9% in AFRO, whereas median HSV-2 prevalence among FSWs in our study was > 50% in all regions aside from EMRO. This suggests that HIV prevalence may not grow to reach considerable levels in many FSW populations in EMRO, and possibly will persist at levels close to zero HIV prevalence. Having said so, this region could largely benefit from integrating testing for HSV-2 in HIV surveillance activities. However, much more data on HSV-2 prevalence are needed before we can assess HIV epidemic potential among FSWs in this region with meaningful confidence.   www.nature.com/scientificreports/ Several other findings emerged from this study. There was regional variation in HIV prevalence that could not be captured by HSV-2 prevalence, especially so for the African Region (Table 2), but also outside Africa (Table 3). This finding suggests that other factors may differentially impact each of HSV-2 and HIV prevalence, and that these should be accounted for to better describe the HIV/HSV-2 association. This is also supported by modeling analyses that demonstrated that, while some sexual network statistics affect HSV-2 and HIV transmission similarly, others can affect them differentially 9 . A plausible explanation relates to HIV having lower infectiousness and shorter acute infection duration, therefore facing more difficulty in propagating within sexual networks compared to HSV-2 9 . For instance, while concurrency (mean number of current sexual partners) is a strong predictor of both HSV-2 and HIV prevalence, clustering within a sexual network (or high exposure within specific circles), provides a higher chance for HIV to spread, but limits HSV-2 from reaching farther nodes in the wider sexual Table 3. Results of meta-regression analyses assessing the association between HIV prevalence and HSV-2 prevalence among female sex workers globally but excluding the African Region. Adj, Adjusted; AMRO, Region of the Americas; AOR, adjusted odds ratio; CI, confidence interval; EURO, European Region; FSWs, female sex workers; HSV-2, herpes simplex virus type 2; OR, odds ratio; SEARO, South-East Asia Region; WHO, World Health Organization; WPRO, Western Pacific Region. Adjusted R 2 is 58.2% in the multivariable model 1, and 64.1% in the multivariable model 2. a Factors with p value ≤ 0.2 were eligible for inclusion in the multivariable analysis. b Factors with p value ≤ 0.05 and those with 0.05 < p value ≤ 0.1 in the multivariable model were considered as showing, respectively, "strong" and "some" evidence for an association with HIV prevalence. c Analysis of the association with HSV-2 prevalence as a linear term excluded three measures with HSV-2 prevalence ≤ 20% in light of the observed threshold effect. d Missing values for year of data collection were imputed using data for year of publication adjusted by the median difference between year of publication and median year of data collection for studies with complete information.  www.nature.com/scientificreports/ network 9 . Meanwhile, higher degree correlation, that is broad connectivity between sexual partnerships, appears to favor HSV-2 spread, but not HIV 9 . This suggests that, despite the strength of the association, HSV-2 cannot be used as the sole predictor of HIV epidemic potential. Our findings indicated only a small role for self-reported condom use in predicting HIV prevalence (Tables 2  and 3), suggesting that such self-reported behavioral measures may not carry meaningful explanatory power, and affirming documented issues in self-reported measures 11,12,33 .

HSV-2 prevalence a
Our study has limitations. There was variability in the number of paired HSV-2/HIV prevalence measures among FSWs across regions, thus limiting our ability to perform further stratified, region-specific, analyses. For instance, there was an insufficient number of studies from EURO to warrant meaningful analysis and interpretation, and no studies from EMRO. Our regional estimates may have also been biased by some countries having larger data contributions (that is more or larger sample size studies) than others, but meta-regression analyses did not identify an association with study sample size. There was also heterogeneity in HIV prevalence, as commonly seen in observational studies assessing prevalence 5,34 . The latter, however, was (mostly) explained through the meta-regression analyses, which affirmed HSV-2 prevalence as an independent contributor to this heterogeneity (Tables 2 and 3). Only a handful of studies reported age-related data, and these varied immensely in the type of reported measure, thus constraining age inclusion in the analysis.
A number of studies did not report data on condom use among FSWs, and very few reported coverage for other interventions to warrant their inclusion in the analyses. For example, only one study reported antiretroviral therapy (ART) coverage (Table S1 of SI), which presumably could affect the association between HIV and HSV-2 prevalence. This being said, most studies were conducted before the mass scale up of ART (Table 2), and thus ART is unlikely to have affected the observed association in the current analysis but may impact future analyses on future data examining this association. Few studies also reported data on current injecting drug use, a nonsexual mode of HIV transmission, with overall no major differences across regions. The latter however is unlikely to have affected the observed HSV-2/HIV association given that the median fraction of FSWs currently injecting drugs is < 5% (Table S1 of SI). Our findings also showed that even in studies where the proportion of FSWs who inject drugs was ≥ 5%, HSV-2 prevalence was substantial with a median of 72%, likely given the nature of the study population and/or the likelihood of exchanging sex for drugs.
The association between HIV prevalence and HSV-2 prevalence is likely non-linear, although the distribution of measures (Fig. 2) and an earlier mathematical modeling analysis 8 suggested that this association may not be far from linearity (above the threshold effect). This implies that our AOR for the HIV/HSV-2 (linear term) association should be interpreted with caution as an estimate for the average increase in odds of HIV prevalence per 1% increase in HSV-2 prevalence beyond the 20% threshold. While HSV-2 prevalence was probably at endemic equilibrium given infection circulation in human populations for centuries, HIV prevalence may not have been at equilibrium, but we were unable to account for the HIV epidemic phase in the analysis 17 . Despite these limitations, the parsimonious multivariable meta-regression models explained > 65% of the variation in HIV prevalence supporting the inferences drawn in this study.
In conclusion, we demonstrated an association between HSV-2 prevalence and HIV prevalence among FSWs that can be utilized in assessing HIV epidemic potential in this at-risk population. We also demonstrated the relevance of integrating testing for HSV-2 in HIV surveillance activities targeting this population, especially in settings where HIV prevalence among them is still at negligible or low level. Our findings stress the need for HSV-2 testing in future surveillance efforts, notably in IBBSS surveys, as a tool to inform HIV preparedness and resource allocation, particularly in countries where HIV epidemic potential among key populations remains unknown. Such data is essential to avoid the costly implications of emerging HIV epidemics and to ensure that countries are still "on track" towards ending AIDS 35 .

Methods
Data sources and selection methods. We updated a database of paired HSV-2 and HIV prevalence measures, retrieved through an earlier systematic review 20 , by conducting a new search focused on FSWs, on September 3rd, 2019, using broad MeSH/Emtree and free text terms for "sex work", "women", "HSV-2", and "HIV" (search criteria in Box S1 of SI). Paired measures eligible for inclusion were identified through a systematic review process following Cochrane Collaboration guidelines 36 . Briefly, PubMed, Embase, and the abstract archives of International AIDS Society conferences were surveyed. Citations were screened for duplication, and then for relevance using Endnote (Thomson Reuters, USA). Full-texts of articles deemed relevant or potentially relevant underwent further screening, and paired measures for HSV-2 and HIV antibody prevalence (seroprevalence), based on primary data, were identified and extracted along with key information on study population characteristics, year(s) of data collection, year of publication, country of origin/survey, number tested and number positive for HSV-2 and HIV infections, diagnostic tests used for infections' ascertainment, proportion of FSWs who inject drugs, proportion of infected FSWs on ART, and proportion of FSWs reporting consistent condom use. The latter was assessed primarily using self-reported condom use at last sex with client, or alternatively using self-reported "consistent/regular" condom use or condom use "all the time" during commercials sex acts (extraction list in Box S2 of SI).

Plan of analysis. Descriptive analysis.
Scatterplots were generated to illustrate the distribution of paired HSV-2 and HIV prevalence measures among FSWs across world regions. Countries' regional classification was based on the WHO regional definition (WHO classification in Box S3 of SI) 37 . Maps showing countries' data contribution were generated using Tableau Desktop v.10.1 38 . Studies were classified into four categories based on HSV-2 prevalence level among FSWs (< 25%, 25-49%, 50-74%, and 75-100%). Descriptive statistics of the reported HIV prevalence measures were then calculated stratified by HSV-2 prevalence category.

Scientific Reports
| (2020) 10:19293 | https://doi.org/10.1038/s41598-020-76380-z www.nature.com/scientificreports/ Meta-analysis. Forest plots were used to visualise estimates of HIV prevalence and 95% CIs for each HSV-2 stratum. The pooled mean HIV prevalence and associated 95% CIs were estimated, for different HSV-2 strata, using random-effects meta-analysis. Here, variances of HIV prevalence measures were first stabilized using a Freeman-Tukey type arcsine square-root transformation 39,40 . Prevalence measures were then weighted using the inverse-variance method 40,41 , and subsequently pooled using a DerSimonian-Laird random-effects model 42 to account for sampling variation and true between-study heterogeneity 43 . Heterogeneity across HIV prevalence measures was assessed, with and without considering HSV-2 stratification, using: Cochran's Q statistic to confirm existence of heterogeneity across prevalence measures, I 2 to quantify magnitude of variation that is due to true differences in prevalence across studies rather than chance, and prediction interval to estimate the 95% interval of the distribution of true prevalence measures 43,44 . Additional metaanalyses contrasting the African Region to the rest of world regions were performed, for relevance, as almost all HSV-2 prevalence measures in this region were > 50% (in contrast to the other regions), and considering the unique HIV epidemic history in this part of the world 1 .
Meta-regression. Random-effects meta-regression analyses were conducted to assess whether HSV-2 prevalence can be used as a predictor of HIV prevalence among FSWs. Covariates, considered a priori,  (Table S1 of SI). The proportion of infected FSWs on ART also could not be factored in our analysis as only a single measure was identified (Table S1 of SI). Missing values for year of data collection were imputed using data for year of publication adjusted by the median difference between year of publication and year of data collection (for studies with complete information). Meta-regression analyses were performed using two scenarios including and excluding AFRO. Meta-regressions estimated the odds ratios of HIV infection assuming that the probability of HIV infection for a given population is equal to that of HIV prevalence in this population. Factors associated with HIV prevalence at p value ≤ 0.20 in univariable analysis were eligible for inclusion in the multivariable analysis. Two multivariable models were considered using HSV-2 prevalence as a categorical variable, or as a linear term after excluding HSV-2 prevalence ≤ 20% given observed threshold effect. In the multivariable model, a p value of ≤ 0.05 for any factor indicated strong evidence for an association with HIV prevalence, while 0.05 < p value ≤ 0.1 indicated some evidence for an association with prevalence.