Secondary Logo

Journal Logo

What Really Is a Concentrated HIV Epidemic and What Does It Mean for West and Central Africa? Insights From Mathematical Modeling

Boily, Marie-Claude PhD*; Pickles, Michael PhD*; Alary, Michel MD, PhD; Baral, Stefan MD; Blanchard, James MD, PhD§; Moses, Stephen MD, MPH§; Vickerman, Peter DPhil; Mishra, Sharmistha MD, PhD*,¶

JAIDS Journal of Acquired Immune Deficiency Syndromes: March 1, 2015 - Volume 68 - Issue - p S74–S82
doi: 10.1097/QAI.0000000000000437
Supplement Article
Free
SDC

Background: HIV epidemics have traditionally been classified as “concentrated” among key populations if overall HIV prevalence was below 1% and as “generalized” otherwise. We aimed to objectively determine the utility of this classification by determining how high overall HIV prevalence can reach in epidemics driven by unprotected sex work (SW) and how estimates of the contribution of SW to HIV transmission changes over time in these epidemics.

Methods: We developed a deterministic model of HIV transmission specific to West and Central Africa to simulate 1000 synthetic HIV epidemics, where SW is the sole behavioral driver that sustains HIV in the population (ie, truly concentrated epidemics), and it is based on a systematic extraction of model parameters specific to West and Central Africa. We determined the range of plausible HIV prevalence in the total population over time and calculated the population attributable fraction (PAF) of SW over different time periods.

Results: In 1988 and 2008, HIV prevalence across the 1000 synthetic concentrated HIV epidemics ranged (5th–95th percentile) between 0.1%–4.2% and 0.1%–2.8%, respectively. The maximum HIV prevalence peaked at 12%. The PAF of SW measured from 2008 over 1 year was <5%–18% compared with 16%–59% over 20 years in these SW-driven epidemics.

Conclusions: Even high HIV-prevalence epidemics can be driven by unprotected SW and therefore concentrated. Overall, HIV prevalence and the short-term PAF are poor makers of underlying transmission dynamics and underestimate the role of SW in HIV epidemics and thus should not be used alone to inform HIV programs.

*Department of Infectious Disease Epidemiology, Imperial College, London, United Kingdom;

Centre de recherche du CHU de Québec, Département de médecine sociale et préventive, Université Laval, Laval, Quebec, Canada;

Center for Public Health and Human Rights, Department of Epidemiology, Johns Hopkins School of Public Health, Baltimore, MD;

§Centre for Global Public Health, Department of Community Health Sciences, University of Manitoba, Winnipeg, Manitoba, Canada;

School of Social and Community Medicine, University of Bristol, Bristol, United Kingdom; and

Division of Infectious Diseases, Department of Medicine, St. Michael's Hospital, Li Ka Shing Knowledge Institute, University of Toronto, Toronto, Ontario, Canada.

Correspondence to: Marie-Claude Boily, PhD, Department of Infectious Disease Epidemiology, School of Public Health, LG26, Norfolk Place, St. Mary's Campus, Imperial College London W2 1PG, London, UK (e-mail: mc.boily@imperial.ac.uk).

The authors have no funding or conflicts of interest to disclose.

M.-C.B and S.M. contributed equally. M.-C.B. formulated the research question. M.-C.B. and S.M. conceived of and designed the study. S.M. designed, developed, and analyzed the mathematical model and conducted the systematic and comprehensive reviews and the data syntheses. M.-C.B. and S.M. wrote the article. M.P. contributed significantly to the study and model design and development. M.A., S.M., and J.B. contributed original data to the data syntheses. All authors provided critical intellectual input into the interpretation of results and edited the article. PV also provided feedback at the design stage.

Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal's Web site (www.jaids.com).

Back to Top | Article Outline

INTRODUCTION

The term “concentrated” epidemic has been commonly used in the HIV public health literature to describe epidemics where HIV prevalence is disproportionately higher (>5%) among key subgroups of the population [including female sex workers (FSWs), men who have sex with men, and people who inject drugs] than in the general population (HIV prevalence <1%)1 (Table 1). Accordingly, “generalized” epidemics have traditionally been defined as those where HIV prevalence exceeds 1% in the general population, which are often represented by antenatal clinic (ANC) attendees. This dichotomy has been used to inform the design of HIV surveillance systems, the allocation of HIV prevention resources, and to decide which risk groups should be prioritized.1,2 For example, it is recommended that key populations (KPs) are prioritized in “concentrated” HIV epidemics, whereas in “generalized” epidemics, the recommendations have focused on HIV programs supporting all segments of the population, assuming that resources would be accessible to KPs without tailored KP programs.6,7 Although this traditional classification is not necessarily wrong when used to describe HIV epidemic patterns, it may be misleading when used to inform the design and content of HIV surveillance and prevention programs, and resource allocation.8–11 Studies suggest that the traditional descriptive classification system does not reflect the underlying transmission dynamics of the epidemic, especially in higher prevalence settings.8–10,12,13 This is because HIV prevalence reflects HIV acquisition and does not provide information on relative risks of onward transmission to and from subpopulations.3,14

TABLE 1

TABLE 1

A key question relevant for HIV programming in sub-Saharan Africa (SSA) is the extent to which unprotected SW, and unprotected sex or needle exchange among other KPs, contributes to the HIV epidemic. The traditional classification of epidemics and output from the UNAIDS Modes of Transmission model8,10,15 may have been misinterpreted and thus underestimated the role of SW on overall HIV transmission in SSA and potentially devalued the importance of focusing HIV prevention on FSWs and their clients.11,16 In West and Central Africa (WCA), HIV prevalence in FSWs and the total population ranges from 7% to 52%, and from 0.1% to 15%, respectively, based on subnational data collected between 2006 and 2009.16–20 Thus, many of these regional epidemics would be classified as “generalized.” However, recent studies suggest that even in settings where HIV prevalence exceeds 1%, epidemics may be truly concentrated,10,12,16 that is, in the absence of transmission during SW, the HIV epidemic would never have occurred and/or it could not be sustained (Table 1). For example, in Cotonou, Benin, it is estimated that more than 93% (range, 84%–98%) of all HIV infection between the start of the epidemic and 1993 may be attributable to unprotected SW, despite an estimated HIV prevalence of 3.3% and 3.4% overall among males and females in 1998, respectively.12,21–23 It has also been estimated that the SIDA1/2/3 program focused on FSWs in Benin since 1993, and extended to clients in 2000, may have averted 33% (range, 20%–46%) of all HIV infections between 1993 and 2008 in the overall population, highlighting the relevance of a tailored KP intervention even in some “generalized” epidemic settings.12 Although these studies suggest that SW could be the main behavioral driver of some HIV epidemics with prevalence exceeding 1% in the general population, it remains unknown how big an HIV epidemic driven only by SW can get. Answering this question could improve our understanding of the epidemiology of HIV, our interpretation of HIV prevalence patterns in WCA, and what we can infer about underlying transmission dynamics from overall HIV prevalence.

The objectives of this study are to (1) develop a dynamic mathematical model of HIV transmission informed by the systematic extraction of most of the relevant behavioral and epidemiologic parameters specific to WCA and (2) use this model to simulate a large family of realistic, data-driven, synthetic epidemics to determine how high overall HIV prevalence can reach in epidemics driven solely by SW (ie, “truly” concentrated HIV epidemics) and how the contribution of SW to HIV transmission (the population attributable fraction, PAF) changes over time in these concentrated HIV epidemics (Table 1).

Back to Top | Article Outline

METHODS

Approach

We generated 1000 synthetic HIV epidemics using the best available epidemiologic and sexual behavior data from WCA (for list of countries, see Appendix 1, Supplemental Digital Content, http://links.lww.com/QAI/A596). For clarity, we conducted the study before wide-scale antiretroviral treatment (ART) access to better answer the research question because effective ART can increase HIV prevalence through increased survival of persons with HIV and on ART. ART coverage remained below 28% in West Africa by 2008,24 and thus, we generated synthetic epidemics by drawing on data collected on or before 2008.

Back to Top | Article Outline

Dynamic Mathematical Model

We developed a dynamical deterministic mathematical model of HIV transmission by extending a previously described dynamic model of commercial SW (CSW).10 The simulated population is stratified into 2 age groups (<24 and ≥24 years) and by sexual activity into low-activity individuals who engaged in multiple partnerships (high volume and low volume), high-volume FSWs/clients, low-volume FSWs/clients, and former FSWs/clients and individuals who used to engage in multiple partnerships (former high-activity class whose behavior is now assumed to be the same as others in the low-activity class). The model is represented by a set of ordinary differential equations detailed in Appendix 1 (see Supplemental Digital Content, http://links.lww.com/QAI/A596). Individuals enter the simulated population at the onset of sexual activity into the low-activity or multiple-partnership classes; from which they can enter SW (as FSWs or clients) at a rate dependent on the fraction likely to enter SW and replace those who exit SW. Individuals in the higher activity classes (including those engaged in SW) retire into the former high-activity classes. Individuals enter the simulation as susceptible (HIV uninfected) and may become infected with HIV with a force of infection dependent on partnership type, partner change rate, HIV prevalence of partners and the disease/infectivity stages of infected partners, sexually transmitted infection, number of sex acts, probability of transmission per sex acts, male circumcision, and condom use. Individuals then progress through 4 stages of untreated HIV reflecting CD4 decline and differential HIV infectivity. The model includes Herpes simplex virus 2 (HSV-2) coinfection at a stable prevalence, wherein HSV-2 increases HIV infectiousness and susceptibility, per sex act,25–29 and was included to enable further transmission heterogeneity across activity classes. The model includes baseline male circumcision, which reduces HIV susceptibility in males by 60%30,31 and is assumed to remain stable over time.

The model includes 5 types of sexual partnerships to reflect the variability in partnership dynamics from the sexual behavior data from WCA. Each type of partnership has a different number of sex acts/year and levels of condom use: SW with occasional clients, SW with regular/repeat clients, transactional (financially motivated but not formal SW32,33), casual sex, and main partnerships. Table 2 summarizes the different risk groups, types of partnerships, and with whom those partnerships are formed. Condom use within each partnership type increased linearly from zero at the start of the epidemic to the first estimates obtained from the data syntheses (range in time period 1991–2000), after which condom use increased through a logistic function to saturate at the most recent estimate (data from 2005 to 2008), remaining stable thereafter.

TABLE 2

TABLE 2

Back to Top | Article Outline

Data From WCA for Model Parameterization

To ensure that the simulated synthetic epidemics were plausible, we comprehensively reviewed and extracted behavioral data (including condom use), HIV and HSV-2 prevalence data for SW, and other sexual partnerships in WCA to parameterize the model. Because most HIV programs are implemented at the province or state level (usually with adult population size >250,000–500,000), we extracted data at the subnational level where available to more adequately capture heterogeneity between locales. The details of the systematic searches and comprehensive review are described in Appendix 1 (see Supplemental Digital Content, http://links.lww.com/QAI/A596). In short, we performed a 3-stage data synthesis to extract parameter values in the 2 main domains: biological (such as male circumcision and HSV-2 prevalence) and sexual behavior. The first stage involved an expansion of a previous systematic review16 to obtain data on sexual behavior within commercial sex, HSV-2, and HIV prevalence on FSWs from 1985 onward. The second stage involved extraction of raw data from the demographic health surveys34 to obtain sexual behavior data on noncommercial partnerships. The third part involved a gray literature search from the UNAIDS country reports for data on overall HIV prevalence across provinces/states and for reports on “noncommercial” multiple partnerships, supplemented by drawing from published systematic reviews of parameters relevant to noncommercial partnerships and populations. Parameters, which were assumed constant between locales within WCA—such as the biological transmission probability of HIV per sex act, and untreated HIV progression—were drawn from the literature. The ranges of extracted parameters used in the model and their sources are shown in Appendix 1 (see Supplemental Digital Content, http://links.lww.com/QAI/A596), with parameters related to partnership type shown in Table 3.

TABLE 3

TABLE 3

Back to Top | Article Outline

Sampling and Plausibility Checks

We used Latin hypercube sampling with a uniform distribution of the parameter range. As much as possible, correlations between parameter values were accounted for by using ratios and relative risks (such as with HSV-2 prevalence and contact rates within specific risk groups). Generating relative risks and ratios was governed by data availability for the same province/state or within the same source (publication or survey). To ensure that we sampled enough parameter sets without SW (to mimic locales where the data might truly suggest there is no CSW), we set 15% of sampled parameter sets to have zero FSWs (and thus, values of zero for all SW parameters). This was also performed to ensure that we did not bias the study toward all synthetic epidemics including some (even if very small networks of) SW. We then conducted plausibility checks of sampled parameter combinations as follows:

  • The relative size of risk groups remained relatively stable; that is, they did not vary by more than 15% of their value at the start of the epidemic.
  • Client population size would not exceed 35% of the male population. This was based on the largest subnational estimate from the systematic review,16 using the indirect method of estimating client population size.35
  • The total population did not exceed a 5% annual growth rate.36
  • The balancing of partnerships did not produce large changes in the partner change rates (of more than 15% of their input value) of a given partnership type in each activity class.
  • FSW incidence would not exceed 50% in the first 2 years of seeding HIV. This was based on pre-2002 HIV incidence measurements of 10%–30% per year.37 These empirical estimates were measured among women who had already been in SW for >2 years and estimated after 1994; thus we used 50% as our upper bound for feasibility checks.
  • An epidemic established when all condom use was set to zero from the start of the epidemic. That is, each synthetic epidemic satisfied the following condition in the absence of condom use: total HIV incidence exceeded 1 per 1000 people per year (as per Granich et al38 working definition for local elimination) at 50 years from HIV seeding.
Back to Top | Article Outline

Generating Plausible Synthetic HIV Epidemics Based on Data From WCA

Each simulation that passed the plausibility checks was run to 50 years postseeding, and the epidemic type (ie, concentrated or not) was determined for each simulation. To determine whether or not the epidemic was concentrated, we simulated the same parameter set with the following counterfactual: where HIV transmission during SW was set to 0 in both directions (ie, from clients to FSWs and from FSWs to clients, during SW partnerships). In this counterfactual, condom use within transactional sex, casual sex, and main partnerships were also set to zero to ensure we did not incorrectly label a nonconcentrated epidemic as concentrated because of increasing condom use in non-SW partnerships. An epidemic was considered concentrated if in the counterfactual, the annual HIV incidence in the total population was less than 1 case per 1000 people per year38 at 50 years after seeding. The 50-year time frame was chosen after checking the sensitivity of the classification to assessment postseeding.

Although the goal was not to fit the model to a specific locale within WCA, we restricted model outputs to the observed epidemics in WCA. Thus, we bounded the simulations by imposing constraints using the upper bound of the documented FSW and province/state level overall or antenatal clinics population HIV prevalence ranges for each region by time period (<1990, 1990–1995, 1996–2000, 2001–2005, 2005–2008). We used the overall HIV prevalence data to bound HIV prevalence in young females (excluding FSWs) but did not bound client HIV prevalence. Parameters were resampled and the above steps repeated until 1000 CSW-driven synthetic HIV epidemics were obtained.

Back to Top | Article Outline

Analyses: Estimated Maximum Epidemic Size

First, we estimated the maximum HIV prevalence in the total population, measured at any time after the start of the epidemic and before 2009, from each synthetic epidemic. These results provide a theoretical estimate of how big an epidemic driven by sex work can get. We also conducted a sensitivity analysis where the general population females (females excluding FSWs) were used. Because results were not very different (less than 10% relative difference in the maximum), we present the overall HIV prevalence (in the total population).

Back to Top | Article Outline

Analyses: Population Attributable Fraction of SW to Concentrated HIV Epidemics

The cumulative PAF of SW is a measure of the contribution of SW to direct and indirect HIV transmission.5,9,12,13,39,40 We measured the cumulative PAF of SW in the synthetic epidemics by “turning off” (ie, setting to zero) HIV transmission during occasional and regular/repeat commercial sex partnerships from time t0 onward and comparing the relative difference in the cumulative number of new HIV infections acquired in the total population over x years. This approach was exactly the same as that taken for determining if epidemics were concentrated, with the following exception: for the PAF, condom-use in other (non-SW) partnerships was allowed to rise. The cumulative PAF of SW is measured from the model outputs as where the superscripts SW and no SW indicate the total cumulative number of new HIV infections in presence and absence of SW transmission, respectively, over the relevant time period:

We measured the PAF1988+x and PAF2008+x for different values of x, to determine how the cumulative PAF in concentrated HIV epidemics varies over the course of the epidemic [early in the HIV epidemic (1988) and later in the epidemic (2008)] and how it changes over time (from x = 1 to x = 20 years). We then explored the relationship between condom-use levels (weighted average of condom use with occasional and regular clients) achieved by 2008, stratified by overall HIV prevalence, with the cumulative PAF over 20 years.

Back to Top | Article Outline

RESULTS

Figure 1A shows the predicted HIV prevalence over time from the 1000 simulated epidemics. Figure 1B shows the range of empiric estimates of consistent condom use with clients (and used in simulations) in the previous 30–90 days as reported by FSWs, and which increases over time. Figure 1A shows that in the absence of HIV transmission during SW, no HIV epidemic would occur (blue line at 0% prevalence), reflecting that SW is the main driver of all the simulated epidemics under study.

FIGURE 1

FIGURE 1

HIV prevalence in the total population in concentrated HIV epidemics varies from 0.1% to 4.2% (5th–95th-percentile range) in 1988 and reaches a maximum of 12% HIV prevalence (Fig. 1A). HIV prevalence ranges between 0.1% and 2.8% (5th–95th-percentile range) in 2008 even if condom use during SW has reached a relatively high level in most synthetic epidemics in 2008 (median of the weighted average for occasional and regular commercial sex, 74%). These results suggest that HIV epidemics can be sustained by SW even if overall HIV prevalence is as high as 12%. No epidemic exceeds 15% HIV prevalence.

Figure 2 shows the

of cumulative HIV infections because of SW from t0 = 1988 onward or t0 = 2008 onward over different time periods x. Because all the synthetic concentrated HIV epidemics are driven by SW, the PAFstart of epidemic + large x of cumulative HIV infections from the start of the epidemic is, by definition, 100%. Despite this, the

measured from 1988 or from 2008 varies substantially when measured, over different time periods (x years) and across simulated epidemics. The long-term PAF1988+x are larger than the 1 year PAF1988+1; the latter fails to account for chains of secondary transmission from FSWs and their clients to subsequent partners (and vice versa). The median PAF1988+x increases from less than 5% over 1 year, to 33% over 5 years, and to 58% over 20 years (Fig. 2A). The

is also larger when measured earlier in the HIV epidemic (PAF1988+20 = 58%) than later (PAF2008+20 = 32%) because condom use during SW was always higher later in the epidemic (Fig. 1B). There is a wide variation in the PAF across epidemics. For example, PAF2008+20 ranged from 16% to 59% (Fig. 2B). Figure 2C shows a weak negative association of the PAF2008+20 on the level of condom use achieved by 2008 during SW. PAF2008+20 can be very different across epidemics of similar HIV prevalence and even condom-use levels. The PAF2008+20 can also be very similar in epidemics of very different prevalence levels, reflecting the influence of other elements of the underlying transmission chains (ie, other determinants and model parameters that were varied in the simulations).

FIGURE 2

FIGURE 2

Back to Top | Article Outline

DISCUSSION

We conducted a mathematical modeling exercise using the best available data from WCA to explore the potential size of concentrated HIV epidemics (driven by SW) and to examine the properties of the PAF of SW to HIV epidemics. The study provides key insights (Box 1) and adds new knowledge to previous studies by suggesting that HIV epidemics driven and sustained by SW could achieve higher overall HIV prevalence than typically assumed.10,12,13 To the best of our knowledge, this is the first modeling study that draws on the best available data across a wider region of SSA to clarify and explicitly demonstrate that concentrated HIV epidemics can achieve an HIV prevalence exceeding 1%, and which may be as high as 12%. None of the synthetic concentrated HIV epidemics exceeded HIV prevalence of 15%, suggesting that it may be much less likely that SW could be the sole behavioral driver in regions experiencing such high HIV prevalence. However, as alluded to in previous studies, unprotected SW could still contribute substantially to onward HIV transmission in regions with HIV prevalence exceeding 15%.3,4,10,41 Although it has recently been recommended to restrain the use of the 1% threshold for classifying epidemics because of its rigidity and potential for confusion, it is still often used as a descriptor of epidemics.42 The results here provide a scientific rationale for abandoning the 1% threshold for inferring underlying transmission dynamics and guiding prevention. Our results have important implications for improving our understanding of the epidemiology of HIV in WCA.

Back to Top | Article Outline

BOX 1 Key Insights From Our Analysis Cited Here...

  1. HIV prevalence of epidemic driven by SW in WCA can easily exceed 1% and be as high as 12%
  2. We cannot infer that SW does not drive an HIV epidemic solely based on a high overall HIV prevalence
  3. The traditional epidemic classification based on the HIV prevalence criteria of 1% is a very poor proxy of underlying transmission dynamics (and thus, epidemic type)
  4. The cumulative PAF of SW can vary substantially across epidemics even if they are all driven by SW. Epidemics of similar HIV prevalence can have different PAF. Epidemics with different HIV prevalence can have similar PAF
  5. The short-term PAF is a poor marker of epidemic type and cannot identify the real behavioral drivers of HIV epidemics
  6. Traditional classification of epidemics based on the HIV prevalence criteria should only be used to describe HIV epidemics because it does not provide insight into the transmission potential of each risk behaviors and their importance in sustaining the HIV epidemic in the long term
  7. A more meaningful definition of a concentrated epidemic should better reflect the vulnerabilities and risk behaviors in the minority of a population that can sustain HIV epidemics in a region
  8. There is a need to revisit the contribution of SW to WCA epidemics to determine the appropriate scale and content of HIV prevention, treatment, and care programs for different populations at risk of acquiring and transmitting HIV

Our results on the short-term and long-term PAF demonstrate its subtlety. The

measured after the start of the epidemic as typically performed in many studies40,43 varied greatly across concentrated HIV epidemics, where by definition, the PAFstart of epidemic + large x =100%. The PAF can be difficult to interpret because different estimates can be obtained from epidemics with similar HIV prevalence and even similar condom-use levels. The short-term PAF (eg, over 1 year typically used in studies13,16,40,44) remains consistently lower that the long-term PAF because the former fails to account for longer chains of secondary transmission from (and to) FSWs, their clients, and subsequent partners. Like the overall HIV prevalence, the short-term PAF is a poor marker of epidemic type and cannot be used to infer the behavioral drivers of HIV epidemics—that is, those risk behaviors that enable HIV to establish and spread in the population and in the absence of which the epidemic would eventually fade out.5

What is a concentrated epidemic? Our results suggest that a more meaningful definition of a “truly” concentrated epidemic would reflect which behaviors (eg, SW) are necessary for HIV to establish and persist in a locale, such that it is necessary to focus on these behaviors to control the local HIV epidemic (Table 1). In other words—and going back to early HIV epidemic theory4,45–47—concentrated epidemics should be defined as 1 where vulnerabilities and risk behaviors in a small fraction of a population can lead to a disproportionate amount of HIV transmission and sustain onward transmission. According to this new definition, our results suggest that it would be theoretically possible for concentrated epidemics driven by SW in WCA to be as large as 10%–12%.

With subnational HIV prevalence ranging between 0.1% and 15% in the overall population, many HIV epidemics in WCA may be concentrated. However, overall HIV prevalence alone will not be helpful to identify “truly” concentrated epidemics, whereas the design of HIV surveillance, prevention/care programs, and resource allocation should be aligned with the best possible characterization of the underlying transmission dynamics and behavioral epidemic drivers.2,3 Thus, it remains critical to identify in which locales, HIV epidemics are driven by SW (or other KP behaviors). Such information would tell us that in these locales, it is necessary, sufficient, and more efficient (especially under constrained resources) to focus HIV prevention and treatment efforts on FSWs and clients to control the local HIV epidemic.5 Further work is required to identify combination of epidemiologic parameters that would help identify SW driven epidemics and develop new and validated tools to classify epidemics in a way that identify the real, and programmatically meaningful, behavioral epidemic drivers and better reflect the underlying transmission dynamics of the epidemic, especially in higher prevalence settings.3,4

Major strengths of our analysis include developing synthetic epidemics based on the best available epidemiologic and behavioral parameters representative of WCA and taking into account various SW partnerships types. The study was designed to theoretically and objectively explore the discordance between traditional epidemic classifications and the underlying transmission dynamics (Table 1), rather than to specifically replicate or make direct inferences about specific subnational HIV epidemics in WCA. Despite using the best data, our conclusions on the maximum size of SW driven epidemics should be interpreted with some caution, in the light of the mathematical model used, which did not include duration of partnerships or long-term concurrency and may have over- or underestimated the epidemic size, respectively. Although our model did not include ART, it is unlikely to alter our conclusions, which were based on data before 2008 (when ART coverage was still low). This needs to be taken into account when interpreting future empirical HIV prevalence estimates. Our results tell us the theoretical maximum size of HIV epidemics. However, they cannot be used alone to determine whether an epidemic of a given size is concentrated or generalized. In other words, if HIV prevalence is below 12% in WCA, this does not necessarily mean that SW drives them but that it is plausible. Such inference requires detailed modeling analysis with sufficient high-quality data in the locale of interest to estimate the PAFstart of epidemic + large x because of specific behaviors, as performed in a recent analysis for Cotonou, Benin.12 Future work includes detailed and multivariate sensitivity analyses of key parameter combinations that influence overall HIV epidemic size and an exploration across other behaviors and KPs.

In conclusion, we cannot make inferences about the underlying transmission dynamics of HIV epidemics based on overall HIV prevalence alone. A more meaningful definition of a concentrated epidemic would capture the transmission dynamics of HIV (Table 1). HIV epidemics in WCA that exceed 1% (up to 12%) HIV prevalence may still be driven and sustained by SW and warrant a revisit to more appropriately characterize the contribution of SW (and other KP behaviors) to overall HIV transmission. Only then will governments have the information needed to determine the appropriate scale and content of HIV prevention, treatment, and care programs for different populations at risk of acquiring and transmitting HIV.

Back to Top | Article Outline

ACKNOWLEDGMENTS

This study was undertaken as a KEYPOP Partnership project. The authors thank Romain Silhol (Imperial College London) for technical advice. This study was conducted as part of a PhD thesis (S.M.), which was supported by a Canadian Institutes of Health Research Fellowship and a Royal College of Physicians and Surgeons of Canada Detweiler Travelling Fellowship.

Back to Top | Article Outline

REFERENCES

1. UNAIDS/WHO Working Group on Global HIV/AIDS and STI Surveillance. Guidelines for Second Generation HIV Surveillance. Geneva, Switzerland: World Health Organization and Joint United Nations Programme on HIV/AIDS; 2000. Available at: http://www.who.int/hiv/pub/surveillance/pub3/en/index.html. Accessed December 1, 2013.
2. Wilson D, Halperin DT. “Know your epidemic, know your response”: a useful approach, if we get it right. Lancet. 2008;372:423–426.
3. Mishra S, Sgaier SK, Thompson LH, et al.. HIV epidemic appraisals for assisting in the design of effective prevention programmes: shifting the paradigm back to basics. PLoS One. 2012;7:e32324.
4. Moses S, Blanchard JF, Kang H, et al.. AIDS in South Asia: Understanding and Responding to a Heterogenous Epidemic. Washington, WA: The World Bank; 2006.
5. Mishra S, Pickles M, Blanchard J, et al.. Validation of the modes of transmission model as a tool to guide HIV prevention: a comparative modeling study. PLoS One. 2014;9:e101690.
6. UNAIDS. Report on the Global AIDS Epidemic. Geneva, Switzerland; 2013. Available at: http://www.unaids.org/en/resources/campaigns/20121120_globalreport2012/. Accessed December 1, 2013.
7. World Health Organization. Prevention and Treatment of HIV and Other Sexually Transmitted Infections for Sex Workers in Low- and Middle-income Countries. Geneva, Switzerland; 2012. Available at: http://www.who.int/hiv/pub/guidelines/sex_worker/en/index.html. Accessed March 1, 2013.
8. Shubber Z, Mishra S, Vesga J, et al.. The HIV Modes of Transmission model: a systematic review of its findings and adherence to guidelines. J Int AIDS Soc. 2014;17.
9. Mishra S, Pickles M, Blanchard J, et al.. Distinguishing the source of HIV transmission from the distribution of newly acquired HIV infections: why is it important for HIV prevention programming?. Sex Transm Infect. 2013. doi: 10.1136/sextrans-2013-051250.
10. Mishra S, Pickles M, Blanchard JF, et al.. Validation of the modes of transmission model as a tool to prioritize HIV prevention targets: a comparative modelling analysis. PLoS One. 2014;9:e101690.
11. Pettifor A, Rosenberg N, Behets F. The need to focus on sex workers in generalized HIV epidemic settings. Sex Transm Dis. 2011;38:324–325.
12. Williams JR, Alary M, Lowndes CM, et al.. Positive impact of increases in condom use among female sex workers and clients in a medium HIV prevalence epidemic: modelling results from Project SIDA1/2/3 in Cotonou, Benin. PLoS One. 2014;9:e102643.
13. Vickerman P, Foss AM, Pickles M, et al.. To what extent is the HIV epidemic in southern India driven by commercial sex? A modelling analysis. AIDS. 2010;24:2563–2572.
14. Blanchard JF. Populations, pathogens, and epidemic phases: closing the gap between theory and practice in the prevention of sexually transmitted diseases. Sex Transm Infect. 2002;78:i183–i188.
15. Prudden H, Watts CH, Vickerman P, et al.. Can the UNAIDS modes of transmission model be improved? A comparison of the original and revised model projections using data from Nigeria. AIDS. 2013;27:2623–2635.
16. Mishra S, Moses S, Boily MC, et al.. Characterizing the contribution of sex work to HIV epidemics in Sub-Saharan Africa: a systematic review, meta-analysis, and mathematical modelling study. 2014. Submitted.
17. Papworth E, Cessay N, An L, Thiam-Niangoin M, et al.. Epidemiology of HIV among female sex workers, their clients, men who have sex with men and people who inject drugs in West and Central Africa. J Int AIDS Soc. 2013;16(suppl 3):18751.
18. UNAIDS. Ungass 2012 Country Progress Reports. Geneva, Switzerland; 2013. Available at: http://www.unaids.org/en/dataanalysis/knowyourresponse/countryprogressreports/2012countries/. Accessed September 1, 2013.
19. World Bank. World Bank Database. 2013. Available at: http://data.worldbank.org/indicator/NY.GDP.DEFL.KD.ZG. Accessed October 31, 2013.
20. Baral S, Beyrer C, Muessig K, et al.. Burden of HIV among female sex workers in low-income and middle-income countries: a systematic review and meta-analysis. Lancet Infect Dis. 2012;12:538–49.
21. Auvert B, Buve A, Ferry B, et al.. Ecological and individual level analysis of risk factors for HIV infection in four urban populations in Sub-Saharan Africa with different levels of HIV infection. AIDS. 2001;15(suppl 4):S15–S30.
22. Buve A, Carael M, Hayes RJ, et al.. The multicentre study on factors determining the differential spread of HIV in four African cities: summary and conclusions. AIDS. 2001;15:S127–S131.
23. Behanzin L, Diabate S, Minani I, et al.. Decline in HIV prevalence among young men in the general population of Cotonou, Benin, 1998-2008. PLoS One. 2012;7.
24. UNGASS 2010 Country Progress Reports. 2011. Available at: http://www.unaids.org/en/dataanalysis/knowyourresponse/countryprogressreports/2012countries/. Accessed September 1, 2013.
25. Freeman EE, Weiss HA, Glynn JR, et al.. Herpes simplex virus 2 infection increases HIV acquisition in men and women: systematic review and meta-analysis of longitudinal studies. AIDS. 2006;20:73–83.
26. Wald A, Link K. Risk of human immunodeficiency virus infection in herpes simplex virus type 2-seropositive persons: a meta-analysis. J Infect Dis. 2002;185:45–52.
27. Barnabas RV, Wasserheit JN, Huang YD, et al.. Impact of herpes simplex virus type 2 on HIV-1 acquisition and progression in an HIV vaccine trial (the STEP study). J Acquir Immune Defic Syndr. 2011;57:238–244.
28. Chen L, Jha P, Stirling B, et al.. Sexual risk factors for HIV infection in early and advanced HIV epidemics in Sub-Saharan Africa: systematic overview of 68 epidemiological studies. PLoS One. 2007;2.
29. Serwadda D, Gray RH, Sewankambo NK, et al.. Human immunodeficiency virus acquisition associated with genital ulcer disease and herpes simplex virus type 2 infection: a nested case-control study in Rakai, Uganda. J Infect Dis. 2003;188:1492–1497.
30. Bailey RC, Moses S, Parker CB, et al.. Male circumcision for HIV prevention in young men in Kisumu, Kenya: a randomised controlled trial. Lancet. 2007;369:643–656.
31. Auvert B, Taljaard D, Lagarde E, et al.. Randomized, controlled intervention trial of male circumcision for reduction of HIV infection risk: the ANRS 1265 trial. PLoS Med. 2005;2:1112–1122.
32. Dunkle KL, Wingood GM, Camp CM, et al.. Economically motivated relationships and transactional sex among unmarried African American and white women: results from a U S national telephone survey. Public Health Rep. 2010;125:90–100.
33. Cooper ML, Barber LL, Zhaoyang R, et al.. Motivational pursuits in the context of human sexual relationships. J Pers. 2011;79:1333–1368.
34. Measure DHS. Publications and Data Search. 2013. Available at: http://www.measuredhs.com/. Accessed October 31, 2013.
35. Cote AM, Sobela F, Dzokoto A, et al.. Transactional sex is the driving force in the dynamics of HIV in Accra, Ghana. AIDS. 2004;18:917–925.
36. The World Bank Databank. 2011. Available at: http://data.worldbank.org/indicator/SP.POP.GROW.
37. Graham SM, Shah PS, Costa-Von Aesch Z, et al.. A systematic review of the quality of trials evaluating biomedical HIV prevention interventions shows that many lack power. HIV Clin Trials. 2009;10:413–431.
38. Granich RM, Gilks CF, Dye C, et al.. Universal voluntary HIV testing with immediate antiretroviral therapy as a strategy for elimination of HIV transmission: a mathematical model. Lancet. 2009;373:48–57.
39. Deering KN, Vickerman P, Moses S, et al.. The impact of out-migrants and out-migration on the HIV/AIDS epidemic: a case study from south-west India. AIDS. 2008;22:S165–S181.
40. Steen R, Hontelez JAC, Veraart A, et al.. Looking upstream to prevent HIV transmission: can interventions with sex workers alter the course of HIV epidemics in Africa as they did in Asia?. AIDS. 2014;28:891–899.
41. Lowndes C, Alary M, Belleau M, et al.. West Africa HIV/AIDS Epidemiology and Response Synthesis: Implications for Prevention. 2008. Available at: www.unaids.org/en/media/.../201003_MOT_West_Africa_en.pdf. Accessed December 15, 2013.
42. WHO/UNAIDS Working Group on Global HIV/AIDS and STI Surveillance, Organization WH. Guidelines for Second Generation HIV Surveillance: An Update: Know Your Epidemic. Geneva, Switzerland: World Health Organization; 2013. Available at: http://apps.who.int/iris/bitstream/10665/85511/1/9789241505826_eng.pdf.
43. Bekker L-G, Johnson LF, Cowan F, et al.. Combination HIV prevention for female sex workers: what is the evidence? Lancet. [published online ahead of print July 21, 2014]. doi:10.1016/S0140-6736(14)60974-0.
44. Prüss-Ustün A, Wolf J, Driscoll T, et al.. HIV due to female sex work: regional and global estimates. PLoS One. 2013;8:e63476.
45. Brunham RC. Core group theory: a central concept in STD epidemiology. Venereol-Interdiscipl Int J Sex Health. 1997;10:34–35.
46. Hethcote HW, Yorke JA, Nold A. Gonorrhea modeling—a comparison of control methods. Math Biosci. 1982;58:93–109.
47. Hethcote H, Yorke J. Lecture Notes in Biomathematics: Gonorrhea Transmission and Control. Levin S, ed. Berlin, Germany: Springer-Verlag; 1984.
Keywords:

concentrated epidemic; West and Central Africa; mathematical modeling; sex work; epidemic driver

Supplemental Digital Content

Back to Top | Article Outline
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.