Adherence to antiretroviral therapy for HIV/AIDS in Latin America and the Caribbean: Systematic review and meta‐analysis

Abstract Introduction Optimal adherence to antiretroviral therapy is closely related with suppression of the HIV viral load in plasma, slowing disease progression and decreasing HIV transmission rates. Despite its importance, the estimated proportion of people living with HIV in Latin America and the Caribbean with optimal adherence has not yet been reported in a meta‐analysis. Moreover, little is known of the factors leading to poor adherence which may be setting‐specific. We present a pooled estimate of adherence to antiretroviral therapy (ART) of people living with HIV in Latin America and Caribbean, report the methods used to measure adherence and describe the factors associated with poor adherence among the selected studies. Methods We electronically searched published studies up to July 2016 on the PubMed, Web of Science and Virtual Health Library (Latin America and the Caribbean Regional Portal); considering the following databases: MEDLINE, LILACS, PAHO and IBECS. Two independent reviewers selected and extracted data on ART adherence and study characteristics. Pooled estimate of adherence was derived using a random‐effects model. Risk of bias in individual studies was assessed independently by two investigators using the Risk of Bias Assessment tool for Non‐randomized Studies (RoBANS). Results and discussion The meta‐analysis included 53 studies published between 2005 and 2016, which analysed 22,603 people living with HIV in 25 Latin America and Caribbean countries. Overall adherence in Latin America and Caribbean was 70% (95% CI: 63–76; I 2 = 98%), similar to levels identified by studies conducted in high‐income regions. Self‐report was the most frequently used method to measure adherence. Subgroup analysis showed that adherence was higher for the shortest recall time frame used, as well as in countries with lower income level, Gross National Income (GNI) per capita and Human Development Index (HDI). Studies reported diverse adherence barriers, such as alcohol and substance misuse, depression, unemployment and pill burden. Conclusions Our study suggests that adherence to ART in Latin America and Caribbean may be below the sufficient levels required for a successful long‐term viral load suppression.


| INTRODUCTION
Latin America and the Caribbean (LAC) consists of 33 sovereign countries which cover an area that stretches from the northern border of Mexico to the southern tip of Chile, including the Caribbean. It has an area of over 20 million km 2 , as of 2017, its population was estimated at approximately 650 million (~9% of the world population), being predominantly urban (80%) [1]. LAC is mostly a developing region which had a combined nominal gross domestic product (GDP) of 5,5 trillion USD and a GDP purchasing power parity (PPP) of 9.7 trillion USD in 2017 [2]. By 2015, the region's HDI was 0.731 (high), varying from 0.493 (Haiti, low) to 0.847 (Chile, very high) [3]. Antiretroviral therapy (ART) revolutionized the treatment of people living with HIV (PLHIV) by dramatically decreasing their morbidity and mortality [4]. In LAC region there were 2.1 million PLHIV and more than 1.1 million PLHIV on ART by the end of 2016 [5]. Indeed, studies have indicated promising results in the region with regards to the HIV Care Continuum: clinical retention, ART use and viral suppression significantly improved from 2003 to 2012 (63 to 77%, 74 to 91% and 53 to 82% respectively; p < 0.05, each), though disparities for vulnerable groups, such as female sex workers, people who inject drugs, gay men and other men who have sex with men, remain [6].
Argentina, Bolivia, Brazil, Costa Rica, Mexico, Paraguay, Uruguay and Venezuela, in Latin America; and Antigua and Barbuda, Bahamas, Barbados, Saint Vincent and the Grenadines, and Trinidad and Tobago, in the Caribbean, have now adopted the World Health Organization recommendation of initiation of ART for all PLHIV irrespective of CD4 cell count [7]. ART coverage reaches 58% (95% confidence interval [95%CI] 42-72%) and 52% (95%CI 41-60%) of all PLHIV in LAC, respectively [7]. At the country level, treatment coverage was 70% in Cuba, followed by 64% in Argentina, 62% in Trinidad and Tobago and 60% in Brazil and Mexico. On the other hand, Bolivia had only 25%, Paraguay had 35% and Guatemala had 36% of PLHIV accessing treatment in 2016 [5]. From 2010 to 2016, the number of deaths has not dramatically decreased (12% of decrease in Latina America and 28% in the Caribbean) and, most significantly, the number of new infections remained relatively stable (0% of decrease in Latin America and 5% decrease in the Caribbean) [5]. This stability hides big differences between countries. In Latin America, although new infections have decreased by more than 20% in Colombia, El Salvador, Nicaragua and Uruguay, they increased significantly in Chile (34%), followed by Guatemala (23%), Costa Rica (16%), Honduras (11%) and Panama (9%); and slightly in Argentina and Brazil (3%). In the Caribbean, the majority of new infections occurred in Cuba, where estimated numbers of new HIV infections more than doubled between 2010 and 2016 [7], despite of the dramatic increase in treatment coverage [5].
The Political Declaration on Ending AIDS by 2030 [8] established specific goals for LAC region, which included reducing the number of new infections in LAC from 100,000 to 40,000 and increasing the number of PLHIV on ART from 1.1 to 1.6 million by 2020. Achievement of these goals will be challenging, requiring continued efforts from governments and international agencies.
Specifically, the HIV epidemic cannot be ended without containing the new infections, and adherence to ART plays an important role in this process. ART adherence is closely linked to suppression of the HIV viral load in plasma [9,10] which leads to immune reconstitution and also decreases onward HIV transmission [11]. The optimal adherence level to achieve viral suppression is unclear, though the 95% threshold established by Paterson et al. has largely been used as a goal [9]. More recently, other authors demonstrated that high levels of viral suppression could be obtained with adherence levels below 95% when in use of newer ART regimens [12][13][14][15][16]. The monitoring of ART adherence is highly recommended by health organizations [17] and the main methods are: selfreport by interview, pill counts, pharmacy refill and medication event monitoring system (MEMS). Monitoring patient's ART adherence is a challenging but critical way to identify those with poor adherence.
Even though there is a huge difference in socio-cultural characteristics across countries in LAC, there are still similar inequalities and traditional values that may act as barriers for HIV treatment, which may impact adherence to ART. This is more pronounced among high-risk populations, who are more vulnerable to social inequalities, discrimination and violence. However, little is known of the factors leading to poor adherence which may be setting-specific. Although ART coverage has increased in the region, the fact that the number of new infections remained relatively stable may be related to the lack of adherence to ART.
In this meta-analysis, we synthesize the published peer reviewed literature, generating a pooled estimate of adherence to ART of PLHIV in LAC. In addition, we present the adherence proportion according to the country's income level, Human Development Index (HDI) rank, and Gross National Income (GNI) per capita, as well as other factors as detailed in the Methods section. We qualitatively synthetize the methods used to monitor adherence and describe the factors associated with poor adherence among the selected studies. Greater knowledge of ART adherence levels of PLHIV in LAC may provide means to improve patient care and could help Governments and regional institutions to accomplish the goal of ending AIDS by 2030.

| Information sources and search strategy
Articles were identified through searches conducted on 14 July 2016 on PubMed, Web of Science and Virtual Health Library (Latin America and the Caribbean Regional Portal) considering the following databases: MEDLINE, LILACS, PAHO and IBECS. The search combined terms derived from four domains: (a) adherence; (b) HIV; (c) antiretroviral (d) countries of Latin America and the Caribbean (see Additional file 1 for the full PubMed search strategy). Citations were inserted into the study database when the four domains were jointly present in the title, abstract, MeSH terms or keywords. No limits were applied for language or publication date in the search. A reference manager (Zotero) was used to collect and organize search results and for duplicate removal.

| Study selection
Two investigators (JMC, TST) reviewed all abstracts and fulltext articles independently, according to the eligibility criteria. Discrepancies were adjudicated by an independent third investigator (PML).

| Data collection process
Data extraction was performed independently by two investigators (JMC, TST) using a predefined extraction form. Each paper was coded for publication characteristics (authors, publication year, full title, journal and language), study characteristics (years when data was collected, country(ies) where the study was performed, study design, sample size, recruitment setting and number of study centres), participants characteristics (age, sex and race/ethnicity), adherence monitoring characteristics (method for adherence measurement, cutoff of optimal adherence, proportion of adherents, time frame used to measure adherence), and the factors significantly associated with adherence (p < 0.05) on multivariate modelling. Discrepancies in extracted data were adjudicated by an independent third investigator (PML).

| Study definitions
Adherence was estimated for each study by dividing the number of individuals with optimal adherence by the number of individuals evaluated. This implies that our overall pooled adherence was based on the adherence threshold adopted in each study. When a study examined the effect of an intervention on ART adherence, only the adherence result at baseline was considered. In case there was no baseline assessment, only the first adherence assessment of the control group was extracted and analysed. As adopted in prior studies [20,21] when more than one adherence measurement was reported, the most objective method was chosen for the analysis (e.g. medication event monitoring system (MEMS) > pill count > pharmacy refill > self-reported adherence in the past week > self-reported adherence in the past month). When an optimal adherence threshold (e.g. ≥80 or ≥95%) was not explicitly defined in the study and adherence was categorized in levels (for example, low, regular and strict; or 65-84%, 85-94% and 95-100%), the highest adherence category was considered.
For subgroup analysis, adherence recall time frame was categorized in four periods: 3-4 days, 7 days, 30 days and 90 days. Location/country was classified by geographical area and categorized as: Brazil, South America (Chile, Colombia, Peru), Central America and Caribbean (Cuba, Dominican Republic, Guatemala, Haiti, Jamaica), North America (Puerto Rico and Mexico) and multi-region (included countries from more than one region). Time period when study was conducted was categorized as ≤2005, 2006-2010 and ≥2011. Study design was categorized as cross-sectional, longitudinal (non-RCT) and RCT. Income group was categorized by low/ lower middle, upper middle, high and mix, following the World Bank definitions for 2017 [22]. HDI and GNI per capita data were extracted from the Human Development Reports of the United Nations Development Programme [3]. The HDI is a composite index measuring average achievement in three basic dimensions of human development -a long and healthy life, knowledge and a decent standard of living [3]; while the GNI per capita reflects the average income of a country's citizens [3]. HDI ranking and the GNI per capita, were classified in two categories each (HDI: <0.754 and ≥0.754; GNI per capita in USD: <14,145 and ≥14,145). When a study involved multiple countries, the lower HDI or GNI value was considered, though we repeated the analysis considering the highest HDI and GNI values and the results changed minimally. Number of study sites was categorized as single-site, multi-site and online (participants accessed and completed survey via an online portal). Treatment experience was categorized as na€ ıve (included individuals who had just recently started ART, usually within six months), experienced (included individuals who had been using ART for some time) and na€ ıve and experienced. Instrument used to measure adherence were categorized as self-report, MEMS, self-report+drug refill and self-report+pill count. Adherence threshold was categorized as <94% (when threshold was ≥64%, ≥80%, ≥85% or ≥90%), 95% (when threshold was ≥95%), 100% and not reported. Finally, the presence of statistical models evaluating factors associated with adherence (YES/NO) was evaluated in subgroup analysis.

| Risk of bias in individual studies
The quality of the included studies was assessed by two investigators (PML, LEC) using the Risk of Bias Assessment tool for Non-randomized Studies (RoBANS) [23]. RoBANS includes criteria for judging the risk of bias for each domain. The risk of bias in a study was graded as low, high or unclear based on the following study features: selection of participants (selection bias), consideration of confounding variables (selection bias), measurement of exposure (detection bias), handling of incomplete outcome data (attrition bias) and selective outcome reporting (reporting bias).

| Data synthesis
Logit transformation of the proportions and their standard errors were calculated to achieve a normal distribution which is required for the pooling of data [24]. Pooled adherence proportion was calculated using the DerSimonian-Laird method [25] assuming a random-effects model. Heterogeneity between studies was initially evaluated by visual inspection of forest-plots. The proportion of true heterogeneity to total variance was calculated by the Higgins I 2 statistic [26].

| Additional analyses
We conducted subgroup analyses and estimated the pooled adherence proportion according to adherence recall time frame, location/country, time period when study was conducted, study design, country's income level, HDI rank, GNI per capita, sites, treatment experience, instrument to measure adherence, adherence threshold, and presence of statistical models evaluating factors associated with adherence.

| Study Characteristics
The flow diagram of study selection is shown in Figure 1. Fifty-three studies, composed of 22,603 participants in ART from 25 LAC countries, met eligibility criteria for the systematic review (Table 1)  . The median number of participants in ART per study was 201 [range: 13- Adherence was most commonly self-reported via structured interviews (96.2%). Forty-nine studies (92.4%), enrolling 21,974 participants, provided a self-reported adherence proportion. Forty-seven studies (88.7%) used self-report instruments only and one used MEMS only. Five studies (9.4%) used a combination of patient self-report, MEMS, pill count and drug refill. The following standardized instruments were used to measure self-reported adherence: the AIDS Clinical Trials Group (ACTG) adherence instrument [80]; the CAT-VIH -Cuestionario de adherencia al tratamiento para el VIH/SIDA [81]; the CEAT-VIH -Cuestionario para la Evaluaci on de la Adhesi on al Tratamiento Antirretroviral [82]; the Morisky, Green & Levine Medication Adherence Scale [83]; the PMAQ -Patient Medication Adherence Questionnaire [84]; the SMAQ -Simplified Medication Adherence Questionnaire [85] and the VPAD-24 -Variables psicol ogicas y comportamientos de adhesi on [86]. Twenty-five studies (47.2%) did not report the instrument used or the instrument was designed for the study or adapted from other studies.
Two studies combined two different adherence measures reporting the overall optimal adherence proportion: Teixeira   [56], assessed adherence using the ACTG questionnaire (5 items) and the adherence index, but only the results for the ACTG questionnaire were considered in this meta-analysis. Though as a general rule we opted to use reported data from the most objective methods, a few exceptions were made. Campbell et al. (2010) [37] used both self-report (Visual Analogue Scale [VAS]) and pill count for measuring adherence. Although pill count was the most objective measure, the time frame information for it was not available, thus only data from the seven-day recall self-report measure (VAS) was considered in the analysis. In Souza et al.
, only the self-reported adherence measure was considered because the adherence as measured from medication dispensing data addressed a period greater than three months (the whole study period). Studies used different thresholds to define optimal adherence (range: ≥64% to > 100%). The most common definitions used for optimal adherence were higher than 95% and 100% of prescribed doses (54.7%). Adherence recall time frames varied between the last three days and the last 90 days. Seventeen studies (32.1%) did not clearly report the time frame used.

| Risk of Bias
The results of the risk of bias assessment for each study included in the meta-analysis are shown in Figure 2. The risk of selection biases due to the inadequate selection of participants was high in 43 studies, low in three and unclear in seven. The risk of selection biases due to the inadequate confirmation and consideration of confounding variables during the design and analysis phases was low in 36 studies, but high in the remaining 17. The risk of performance biases caused by inadequate measurements of exposure was low in 50 studies, high in two and unclear in one. The risk of attrition biases caused by the inadequate handling of incomplete outcome data was low in 47 studies and high in the remaining six. The risk of selective reporting bias was low in all studies.
Our results show that, when assessing adherence, depending on the time-frame for recall, different results might be achieved. Our results point to higher adherence (80%) in the shortest time frame and lower adherence (55%) in the longer time-frame. These findings are consistent with the meta-analysis of adherence among HIV-positive drug users, conducted by Malta et al. (2010) [91], where the pooled estimate for the shortest period was higher (71%) than the pooled estimated for the intermediate period (54%). Moreover, and again, similarly to Malta et al. (2010) [91], the shortest time frame also yielded a less heterogeneous estimate of adherence. Taken    together, these findings suggest that a shorter time frame might yield estimates that are less prone to recall bias and thus more accurate. Conversely, a longer time interval increases the chances of adherence issues, what may not be observed using a shorter time frame. That said, which adherence time frame best predicts virological failure is less known, with a few studies suggesting that the impact of the time frame might be minimal [21,96]. There were no significant differences in the pooled adherence among different optimal adherence thresholds (<94%, 95%, 100%), which was similar to the findings of   [90]. Moreover, in a recent meta-analysis conducted by Bezabhe et al. (2016) [21], there were no significant differences in the pooled odds ratios for virological failure among different optimal adherence thresholds (OR ≥98-100% = 0.54, 95% CI: 0.29-1.00, I 2 = 85%; OR ≥95% = 0.34, 95% CI: 0.24-0.47, I 2 = 92; OR ≥80-90% = 0.34, 95% CI: 0.23-0.51, I 2 = 0%) showing that irrespective of the cutoff point, optimal adherence to ART was associated with positive clinical outcomes.
As per Marmot, health inequalities are perhaps the most damning indictments of social and economic inequalities [97]. Similar social gradients in health can be observed if we stratify the population by region, country or income. Social determinants of health can greatly affect adherence to ART. The proportion of people who reported optimal adherence to ART varied according to the country's income level, HDI and GNI per capita. We found that in studies from lower income countries the pooled proportion of adherence was higher than in studies from middle income countries. HDI and GNI per capita followed the same trend. Studies in countries with a lower HDI and GNI per capita had higher proportions of adherence than studies in countries with a higher HDI and GNI per capita. These findings are consistent with previous meta-analysis: Uthman et al. (2014) [98] found that the proportion of PLHIV who achieved good adherence was significantly higher in lower-income countries (86%) compared to higher-income countries (67.5%; p < 0.05). Although our findings suggest that less developed countries can achieve the same or higher level of adherence than more developed ones, huge differences between populations and availability of healthcare may have been included in each setting (selection bias). It is possible that in lower-income countries participation in a research
Suggesting that although adherence may be similar or higher, the expected effect of adherence on virologic suppression is lower in low/middle income countries. A possible explanation might be the lack of consistent virological monitoring [99] with patients adhering to non-suppressive treatment. Measures of adherence include individual self-report, pharmacy records, pill counts, electronic measurement devices, therapeutic drug concentrations and clinical outcomes. The easiest and therefore most frequently used method is patient self-report. Self-report questions consider the number or frequency of doses missed/taken, or simply asks individuals to rate their adherence level, always considering a specific time period. Questionnaires frequently use a Likert-type scale as the response format (often five-point ordered response ranging from the most positive to the most negative response to a statement). Although individual self-report can be inexpensive, easy to administer, and accurately identify medication-taking behaviour [100], they may also overestimate adherence due to social desirability (i.e. respondents answer questions in a way that will be viewed favourably by clinicians) and recall biases [101]. These biases may have impacted our results, because patient's self-reported adherence was the most common method for assessing adherence in this meta-analysis. Self-report questionnaires, which have a reasonable predictive power, are useful for resource-limited clinical settings. The ACTG Adherence Questionnaire was the most extensively used instrument. It is a 5-item self-report measure, but frequently, only the first item (four-day recall of how many doses have been missed) is used in clinical setting. The CEAT-VIH was the second most used self-report instrument. The CEAT-VIH is a short (20 items  across studies that is due to heterogeneity rather than chance; s 2 , tau-squared is an estimate of the between-study variance; p, p-value. validate adherence measures such as patient's self-report or pharmacy database (medication withdrawal or refill data), the cost of this device substantially impairs its widespread use. Clinical outcomes, is considered by some researchers as one of the best measures of a patient's adherence behaviour, but the use of clinical outcomes as a proxy of adherence can always be biased by the presence of any patient-or diseaserelated factor. Each adherence measurement strategy has strengths and weaknesses. The best measurement strategy for clinical practice should take in consideration the setting, the population, and most importantly have acceptable reliability and validity. We found that the assessment of the quality of the published studies was sometimes challenging. Many studies did not report relevant methodological details about the assessment of adherence, making it difficult to judge the strength of their findings. The absence of these data may have introduced a possible bias in the results (data retrieval bias). To promote improvement in the quality of measurement of medication adherence in research Williams et al. [102] have proposed a set of best practices for conducting adherence measurement. For studies using self-report, for example, it is recommended the use of an instrument and method of administration that demonstrate both concurrent and predictive validity. When using a new instrument, it needs to be validated in a pilot test. When a scale is used to measure adherence, it needs to be culturally sensitive, worded clearly, and subjects need to know how to respond to the scaling response options with little difficulty. In addition to their recommendations, and to improve also the reporting of adherence measurement, we suggest that researchers clearly identify the instrument used, whether it was validated for use in the study population, when the assessment was carried out (date), the adherence recall time frame used, the adherence definition used (i.e. no. of pills taken/prescribed or instrument score), the optimal adherence cutoffs or thresholds adopted and ART used, which can greatly affect adherence to ART. Accurate assessment of adherence behaviour is essential for treatment planning while accurate reporting of adherence studies is essential for further advancement of the subject.
Identifying specific barriers for patients and implementing appropriate interventions to overcome them is extremely necessary to improve adherence. The social support was one of the factors associated with adherence in this systematic review and meta-analysis. Social network and social influence, can provide a powerful approach for health behaviour change [103]. Alcohol use was associated with nonadherence to ART in many studies. This association has been observed previously, where alcohol abuse can be higher in PLHIV than in the general population, and may lead to medical and psychiatric complications, poor adherence and poorer treatment outcomes [104]. In addition, alcohol use is associated with intravenous drug use and risky sexual behaviour, major modes of HIV transmission [104]. Unemployment was another barrier to nonadherence in LAC countries. This association exists globally, and was recently estimated with a pooled odds ratio of 1.27 (95% CI: 1.04-1.55) in a meta-analysis carried out by Nachega et al. (2015) [105]. Reporting of traditional barriers to ART such as toxicity and pill burden has reduced over time since current ART regimens are simpler and better tolerated. Consequently, the primary individual barriers to adherence have changed. In a recent meta-analysis conducted by Shubber et al. (2016) [106], individual barriers most frequently reported by patients included forgetting, being away from home, and a change to daily routine. Depression, alcohol/substance misuse, stigma, feeling sick, health service-related barriers (i.e. distance to clinic) and stock outs were less frequently reported. In this systematic review and meta- When the number of studies in each group was ≤2, meta-analysis was not performed. d Used two methods to measure adherence, self-report or medication withdrawal, to calculate study proportion of participants in optimal adherence. e Used two methods to measure adherence, self-report and pill count, to calculate study proportion of participants in optimal adherence. f p-value for the Q statistic hypothesis test of whether there is heterogeneity, a p-value <0.05 implies a rejection of the null hypothesis that the studies are homogeneous. analysis studies reported many traditional barriers, in particular those related with alcohol and substance misuse. The presence of so many traditional barriers may be indictments that regimens in use in LAC are not aligned with WHO, PAHO and other agencies efforts to optimize drug regimens. In 2012, 4% of the patients of the region, approximately 30,000 patients out of the 725,000 that receive ART, were being treated with obsolete or inappropriate ARVs [107]. Also, in many countries the number of regimens in use still exceeds 15-20 [108].
This study has important limitations worth noting. Metaanalysis of observational studies, unlike randomized controlled trials, are prone to biases including confounding. The high risk of selection biases within included studies may be a function of their observational design. Moreover, we found high heterogeneity among the studies indicating that adherence varied significantly across studies, possibly due to different populations, different thresholds, different time frames and methods of measurement. Accordingly, random effects models were chosen as our analytical framework to better accommodate the heterogeneity since it assumes that each study was drawn from populations that differ from each other in ways that could impact the proportion of adherents. Heterogeneity was not entirely explained by subgroup analysis. A metaregression could help investigate the sources of the heterogeneity across studies by studying the relationship between study-level characteristics and adherence to ART, but it was not performed in this study due to discrepancies associated with reporting of participants' characteristics. For example, we were not able to evaluate a possible relationship between adherence and different regimens because this information was rarely available among the included studies. Another limitation is that most of the studies included in this meta-analysis used a cross-sectional design, making it difficult to determine causal relationships between level of adherence and other factors. Lastly, we did not include grey literature and therefore may have missed studies that were relevant to our research question during the literature search. However, the inclusion of grey literature may itself introduce biases if the studies found were not representative of all unpublished studies. Though, our results must be interpreted with caution, considering our assumptions and limitations, meta-analyses are still the preferred methodology in providing a qualitative interpretation of the results [109].

| CONCLUSIONS
In conclusion, our study suggests that PLHIV in LAC can achieve comparable adherence levels to other populations of developing regions though it may be below the sufficient levels required for successful long-term viral load suppression. Monitoring ART adherence is critical to provide information about possible causes of virological failure in LAC, where viral load testing is often carried out less frequently than regional guidelines recommend [99]. We encourage initiatives for improving ART adherence that consider the social determinants of health inside each context, involving communitybased organizations and social participation to address the huge socio-economic disparities and the health inequities present across and within LAC countries.

C O M P E T I N G I N T E R E S T S
The authors declare that they have no competing interests.

A U T H O R S ' C O N T R I B U T I O N S
All authors contributed extensively to the work presented in this paper. JMC, TST, PML conceived the study. JMC, TST and PML created and performed the literature search strategy, JMC built the data extraction file, JMC and TST performed the data extraction and PML supervised the process. PML and LEC performed the statistical analyses, and all authors interpreted the data. JMC drafted the manuscript, and all the other authors contributed substantially to the writing and revising of the manuscript. All authors have read and approved the final version of the manuscript.

Funding sources
This work was conducted during a scholarship (JMC) supported by CAPES (Brazilian Federal Agency for Support and Evaluation of Graduate Education within the Ministry of Education of Brazil) and FAPERJ (Fundac ßão Carlos Chagas Filho de Amparo a Pesquisa do Estado do Rio de Janeiro) at the Instituto Nacional de Infectologia Evandro Chagas -Fiocruz (Brazilian National Institute of Infectious Diseases). PML acknowledges funding from the National Council of Technological and Scientific Development and FAPERJ.