HPTN 071 (PopART): a cluster-randomized trial of the population impact
of an HIV combination prevention intervention including universal testing
and treatment: Mathematical model
Anne Cori1, Helen Ayles2,3, Nulda Beyers4, Ab Schaap2,5, Sian Floyd5, Kalpana Sabapathy5,
Jeffrey W. Eaton1, Katharina Hauck6, Peter Smith6, Sam Griffith7, Ayana Moore7, Deborah
Donnell8, Sten H. Vermund9, Sarah Fidler10, Richard Hayes5, Christophe Fraser1* and the
HPTN 071 (PopART) study team.
1MRC?? Centre?? for?? Outbreak?? Analysis?? and?? Modelling,?? Department?? of?? Infectious?? Disease??
Epidemiology,?? Imperial?? College?? London,?? London,?? UK.??
2?? ZAMBART,?? University?? of?? Zambia,?? School?? of?? Medicine,?? Ridgeway?? Campus,?? Lusaka,?? Zambia??
3Department?? of?? Clinical?? Research,?? London?? School?? of?? Hygiene?? and?? Tropical?? Medicine,??
4?? Desmond?? Tutu?? TB?? Centre,?? Department?? of?? Paediatrics?? and?? Child?? Health,?? Stellenbosch??
University,?? ?? South?? Africa??
5?? Department?? of?? Infectious?? Disease?? Epidemiology,?? London?? School?? of?? Hygiene?? &?? Tropical??
Medicine,?? London,?? UK??
6?? Business?? School,?? Imperial?? College?? London,?? South?? Kensington,?? London,?? UK??
7 FHI?? 360,?? Research?? Triangle?? Park,?? North?? Carolina,?? USA??
8?? Vaccine?? and?? Infectious?? Disease?? Division,?? Fred?? Hutchinson?? Cancer?? Research?? Center,??
Seattle,?? Washington,?? USA??
9Vanderbilt?? Institute?? for?? Global?? Health?? and?? Department?? of?? Pediatrics,?? Vanderbilt??
University,?? Nashville,?? Tennessee,?? USA??
10Department?? of?? Medicine,?? Imperial?? College?? London,?? UK??
* Corresponding author: Professor Christophe Fraser, MRC Centre for Outbreak Analysis
and Modelling, Department of Infectious Disease Epidemiology, School of Public Health,
Imperial College London, St Mary’s Campus, Norfolk Place, London W2 1PG, UK. Tel: +44
20 7594 3397. Fax: +44 20 7402 3927. Email:?? firstname.lastname@example.org.
The HPTN 052 trial confirmed that antiretroviral therapy (ART) can nearly eliminate HIV
transmission from successfully treated HIV-infected individuals within couples. Here, we present the
mathematical modeling used to inform the design and monitoring of a new trial aiming to test
whether widespread provision of ART is feasible and can substantially reduce population-level HIV
Methods?? and?? Findings??
The HPTN 071 (PopART) trial is a three-arm cluster-randomized trial of 21 large population clusters
in Zambia and South Africa, starting in 2013. A combination prevention package including home-
based voluntary testing and counseling, and ART for HIV positive individuals, will be delivered in
arms A and B, with ART offered universally in arm A and according to national guidelines in arm B.
Arm C will be the control arm. The primary endpoint is the cumulative three-year HIV incidence.
We developed a mathematical model of heterosexual HIV transmission, informed by recent data on
HIV-1 natural history. We focused on realistically modeling the intervention package. Parameters
were calibrated to data previously collected in these communities and national surveillance data.
We predict that, if targets are reached, HIV incidence over three years will drop by >60% in arm A
and >25% in arm B, relative to arm C. The considerable uncertainty in the predicted reduction in
incidence justifies the need for a trial. The main drivers of this uncertainty are possible community-
level behavioral changes associated with the intervention, uptake of testing and treatment, as well as
ART retention and adherence.
The HPTN 071 (PopART) trial intervention could reduce HIV population-level incidence by >60%
over three years. This intervention could serve as a paradigm for national or supra-national
implementation. Our analysis highlights the role mathematical modeling can play in trial
development and monitoring, and more widely in evaluating the impact of treatment as prevention.
HIV, universal testing and treatment, clinical trial, mathematical modeling, treatment as
In 2011, the HPTN 052 trial (HPTN: HIV Prevention Trials Network) reported that early
antiretroviral therapy (ART) reduces HIV-1 transmission amongst serodiscordant couples by
96% . This finding, obtained in a closely monitored individually-randomized trial,
corroborated the results of earlier studies [2,3] and opened new and exciting perspectives for
HIV prevention and control: expanding HIV testing and treatment could reduce sexual
transmission of HIV close to zero . A recent observational study in South Africa
demonstrated that in fact, the ART coverage in the population immediately surrounding an
individual was highly predictive of his/her risk of HIV acquisition . In this context,
several trials have been designed in order to test the feasibility of large scale HIV
combination prevention strategies including universal HIV testing with immediate
antiretroviral treatment for HIV-positive persons, and to measure their impact at the
population level [6-10].
HPTN 071 (PopART, Population effects of Antiretroviral Therapy to reduce HIV
transmission) is the largest of these trials, co-funded by the Office of the US Global AIDS
Coordinator (OGAC), the US National Institutes of Health, and the Bill and Melinda Gates
Foundation. It is planned to start in 2013, with annual follow-up until 2016, and analyses and
results reported in 2017 [11-14].
In brief, it is a cluster-randomized trial consisting of 21 communities in Zambia and South
Africa, covering approximately 1.2 million people. Each community, delimited as the
catchment population of a health facility delivering ART, will be randomized to one of three
arms. Interventions in arms A and B will include home-based voluntary testing (HBT) and
counseling, male circumcision, prevention of mother to child transmission (PMTCT)
services, treatment of sexually transmitted infections (STIs), condom distribution, and ART
for HIV positive individuals. ART will be offered universally (regardless of CD4 count) in
arm A and according to national guidelines (currently CD4<350 cell count per µL of
peripheral blood) in arm B. Arm C will serve as a control arm with health system
strengthening activities to ensure that standard of care services (voluntary testing and
counseling, male circumcision, PMTCT, treatment of STIs, and ART for HIV positive
individuals) are delivered according to national guidelines. The inclusion of three arms will
allow separate assessment of the benefit of enhanced home-based voluntary testing,
counseling and linkage to care, under national guidelines for treatment, and the additional
prevention benefit of treatment regardless of CD4 count. The primary end-point will be
cumulative HIV incidence over 3 years, measured in cohorts of 2,500 adults randomly
selected in each of the 21 communities (total cohort size 52,500). ??
Mathematical modeling is an essential tool to assess the impact of interventions on HIV
epidemics  because of the indirect benefit to members of the population not receiving the
intervention. Also, mathematical modeling allows analyzing in a single framework the effect
of multiple interventions, and thus takes into account synergistic (or interfering) effects
between components of a combination prevention package. Therefore, over the last years,
mathematical models have been increasingly used to provide insights in the potential long-
term impacts of different interventions [4,16,17] and to assist with the post-hoc
interpretation of trials and observational studies [18,19]. It has also become clear that
mathematical modeling could be used more extensively within clinical trials, to assist trial
design, to inform monitoring and evaluation as a trial progresses, and finally to interpret and
extrapolate the trial results .
Mathematical modeling was a key part of designing the HPTN 071 (PopART) trial: we
developed a deterministic compartmental model of HIV transmission specifically conceived
to assist the trial design. We focused on realistically describing the intervention package to
be delivered during the trial. Model parameters were calibrated based on data collected
during previous studies in the study communities as well as routine national surveillance
In the following, we describe this mathematical model and present the predicted impact of
the intervention package that will be delivered during the trial. Most importantly, we present
an extensive uncertainty and sensitivity analysis to quantify the influence of process
variables (such as the uptake of testing) on the relative reduction in population level
cumulative incidence in the intervention arms compared to the control arm. This analysis
pinpoints the key variables that drive the magnitude of the reduction in incidence, and could
therefore affect success or failure of this intervention package. Monitoring those variables
during the trial will enhance evaluation of its progress, as will feeding values back into the
model to obtain revised interim predictions.
Materials and Methods
The model was designed with the intention to be simple but capable of representing different
scenarios explored in trial design, and to represent a relative consensus of existing
approaches to modelling the dynamics of generalised HIV epidemics. Its structure was
particularly inspired by the models of Granich et al. , Hallett et al. , and Bezemer et
al. [21,22]. The model describes the generalised HIV epidemics in Zambia and South Africa,
the two countries where the HPTN 071 (PopART) trial will take place.
The model is a deterministic compartmental model describing heterosexual transmission of
HIV in the population aged 15 and over, specified by ordinary differential equations for the
time-evolution of the number of individuals in different states. Our model is not age-
structured and we therefore do not distinguish between the intervention, which is universal,
and the measurement of incidence, which is in a cohort of adults aged 18 to 44. Our choice
of age group was motivated by the availability of national prevalence estimates to which we
calibrate our model.
A full description of the model structure, equations and parameterization, is presented in the
supporting information (see in particular Figures S1 and S2 in File S1 and Tables S1, S2, S3
and S4 in File S1 for model structure and Tables S5, S6, S7, S8 and S9 in File S1 for
definitions and values of model parameters).
Individuals are classified by sex (female/male), infection status (susceptible/infected), and
sexual risk propensity (high/medium/low). The susceptible and infected stages are further
stratified to represent the clinical progression of HIV and the intervention delivered in each
arm of the trial. The model includes temporal delays between different steps of the
intervention (such as testing and treatment). Susceptible males are classified as
uncircumcised, uncircumcised planning circumcision (following a negative HIV test),
circumcised in the wound healing period, and circumcised (see Figure 1A). Infected
individuals are classified as untreated, untreated waiting for treatment (following a positive
test), treated but not virally suppressed, and treated and virally suppressed. Infected
individuals who are untreated are further classified in one of five disease stages: acute/early
HIV, followed by four stages defined by the CD4 count (stage 1 corresponds to CD4≥500
cells/µL peripheral blood, stage 2 to 350≤CD4<500, stage 3 to 200≤CD4<350, and stage 4 to
CD4<200, see Figure 1B). Upon ART initiation, infected individuals enter an ART category
corresponding to the CD4 count level at which they initiated treatment, such that persons
initiating treatment at higher CD4 levels have a better prognosis. A schematic description of
the model for infected individuals is presented in Figure 2A.
Modelling?? testing,?? treatment,?? and?? circumcision??
We separately modelled a background level of HIV-related care for adults in all arms that
would be presumed to occur in the absence of the trial activities, and an additional process,
specific to interventions implemented in arms A and B during the trial.
Background?? testing,?? treatment?? and?? circumcision??
Background HIV testing was not modelled explicitly. Instead, we modelled the rate at which
individuals initiate ART, encompassing both testing and successful linkage to care. We
assume that only individuals with CD4<350 could initiate treatment. The rate at which they
do so was modelled as a smooth function gradually ramping up from 2004 onwards, with a
greater rate for individuals with CD4<200.
We assumed that, in all arms and both countries, a certain proportion of males are
circumcised prior to entry into the modelled population at age 15. We assume these are fully
circumcised. We did not model any adult circumcision outside of that offered as part of the
intervention package in arms A and B.
During the trial, community HIV care providers teams (CHiPs) will offer, in arms A and B,
?? Additional?? testing,?? treatment?? and?? circumcision?? in?? arms?? A?? and?? B?? during?? the??
home-based testing in annual rounds in all intervention communities. These intervention
rounds are scheduled to last 6 months: here we model these taking place from 1st July to 31st
December, from 2013 to 2015. Both the schedule and start date are subject to adjustment,
with a likely 4 to 5-month delay from this modelled schedule. These annual rounds of testing
were modelled by a constant number of tests offered each day by CHiPs. Following testing
by CHiPs (which, when offered, is only accepted by a proportion of individuals), a fraction
of men testing negative will decide to get circumcised and a fraction of individuals testing
positive will decide to link to care (and start ART if eligible). Those individuals will go
through the “waiting” stages (awaiting circumcision or awaiting treatment) before becoming
circumcised or initiating treatment. Those stages were modelled to account for delays from
the time that elapses between HIV testing and presentation at the health facility for
circumcision or treatment initiation.
Treatment?? failure?? and?? drop-‐out??
Individuals on ART were assumed to stop receiving treatment (e.g. due to dropping out or
treatment failure) at a rate of 10% per year (an assumption varied in sensitivity analysis).
They were then assumed to go back to the “untreated” stage. They may then be re-started on
treatment at a later time at the same rate as treatment-naïve persons.
Clinical?? progression?? on?? and?? off?? treatment??
Upon becoming infected, all persons first enter a period of acute HIV infection lasting for a
mean of 2.9 months, after which infected persons may enter any of the CD4 cell count
categories. The proportion entering each category and the rate of progression to the next
lower CD4 count category were calibrated to recent clinical cohort data from the large
multinational CASCADE collaboration  (see supporting information). Compared to
previous models, which assumed all individuals start with a post-seroconversion CD4 count
≥500 and progress through each CD4 stage, the new model better captures heterogeneity
between individuals observed in the clinical seroconverter data.
Upon treatment initiation, following the approach of Granich et al. , individuals enter a
‘treatment’ compartment mirroring the CD4 stage from which they initiate treatment. They
then progress through stages of treatment half as fast as untreated patients. This simple
model allows capturing the improved prognosis for patients initiating treatment at higher
CD4 cell counts [24-28]. Sensitivity analyses show that the short-term predicted
epidemiological impact is not strongly dependent on assumptions about the rate of
progression of individuals on treatment (see supplementary material). However, a more
mechanistic representation of viral suppression and CD4 reconstitution  could be
important for capturing long-term predictions of epidemiological impacts, costs, and clinical
benefits. Interpretation of current clinical data from generalised epidemics in sub-Saharan
Africa has proven difficult because of the confounding of mortality and loss to follow up
; improving these estimates will be an important feature of the analysis of trial outcomes
in HPTN 071 (PopART), albeit with a relatively short time horizon.
Contact?? patterns,?? relative?? susceptibility?? and?? relative?? infectivity??
We use a model of assortative heterosexual sexual mixing between three sexual risk groups.
Individuals in our model form partnerships at different rates according to their risk group.
We assume that individuals in the low and middle risk groups have on average one new
partner every ten years and one partner every year, respectively. The average number of
partners per year for individuals in the high risk group is calibrated to fit national HIV
prevalence estimates. We assume partnerships are made preferentially within the same risk
group, with a level of assortativity which is calibrated by fitting the model to national HIV
prevalence estimates. We also assumed that 5% of partnerships are formed with partners
from outside the study community, thereby allowing for “contamination” of the intervention
communities. We further assume that within a partnership, unprotected sexual acts occur at
an instantaneous rate which depends on the risk groups of the two individuals: it is the same
for all partnerships between individuals of different risk groups, as well as partnerships
between two mid-risk individuals; it is twice as high for partnerships between two high-risk
individuals and twice as low for partnerships between two low-risk individuals.
During the trial, we aim to collect data for better parameterisation of this component of the
We assumed that circumcision decreases male susceptibility by 60% [31-34]. We assume
infectiousness is greater during acute/early and late stage infection, and reduced for
individuals on ART (see Figure 1B, Figure 2C and Table S7 in File S1). Men in the wound
healing period following circumcision are assumed to have decreased sexual activity, but an
increased susceptibility and infectiousness per sex act, in balance leading to an overall
reduced susceptibility and infectiousness during the same period than had they not
undergone circumcision [35,36]. We assume no difference in infectiousness for circumcised
and healed infected males and uncircumcised infected males.
In a sensitivity analysis, we also investigate potential consequences of reductions in
unprotected sexual activity (modelled as lower susceptibility and infectivity levels) for
individuals in the “waiting” stages due to the HIV counselling and condom distribution.
The non HIV-related death rate was calculated dynamically to constrain the population size
and the birth rate to match national demographics data since 1978 (see supporting
The HIV epidemic was calibrated to match HIV prevalence estimates reported by UNAIDS
 by varying the basic transmission rate (λ0, the rate at which an untreated infected
individual with CD4≥350 not in acute infection transmits to a partner, assuming they are
both in the mid-risk group), the time of seeding of the epidemic, the proportion of
individuals in each risk group, the rate of sexual contacts in the high risk group, and the
assortativity. The background rate of ART initiation was modelled as a Hill function
increasing from 2004 onwards to achieve the ART coverage data reported during the
ZAMSTAR trial [38,39].
Uncertainty?? and?? Sensitivity?? analysis??
Uncertainty and sensitivity analyses were conducted to assess whether the predicted
reduction in HIV incidence in the intervention arms was strongly influenced by the
parameter values chosen to best fit the UNAIDS national prevalence estimates, and to
analyse the impact that the “process” parameters, such as the uptake of circumcision during
the intervention, would have on the estimated reduction in HIV incidence.
Influence?? of?? parameters?? calibrated?? to?? prevalence?? curves??
For each country, we used a Latin hypercube sampling scheme  to simulate epidemics
for range of values for the parameters described in Table 1, and selected the 9 parameter sets
(out of 9000) with best fits to the prevalence. For each of these, we then ran an optimization
routine, starting from this parameter set, to obtain a neighbour parameter set with an
improved fit to HIV prevalence. Because we were using a local optimisation algorithm, this
did not converge on the global optimum. We compared the predicted reduction in incidence
under these 9 final parameter sets to the original best-fit parameter combination.
Influence?? of?? process?? parameters??
To explore the influence of process parameters that could potentially be controlled during
the intervention implementation, we defined four scenarios ranging from best to worst case
(most optimistic, optimistic, central and most pessimistic), with corresponding parameters
shown in Table 1. For each country, we generated, using a Latin hypercube sampling scheme
, a set of 1000 parameters drawn uniformly within the range defined by the worst and
best cases, and examined the resulting variability in the predicted 3-year cumulative HIV
incidence in each arm. In order to assess the main drivers of this variability, we used a linear
model exploring the relationship between the reduction in 3-year cumulative HIV incidence
in intervention arms and the process parameters. The relative impact of each process
parameter on the reduction in incidence was assessed by examining the proportion of the
variance explained by each predictor (see supporting information).
The projected HIV prevalence and incidence for each country and in each arm are shown in
Figure 3, demonstrating a good fit to the UNAIDS national prevalence estimates used for
calibration. The saw-tooth pattern in incidence in the intervention arms projections reflects
the six-monthly rounds of the intervention. In a sensitivity analysis, we found that rounds of
6 months for the CHiPs intervention would be preferable to rounds of 9 or 12 months, as
they would lead to a greater reduction in HIV incidence over 3 years (see supporting
information, in particular Figure S7 in File S1). The predicted relative reductions in HIV
incidence for both countries are shown in Table 2. Under the central target, we estimated a
reduction in 3-year cumulative incidence of 61% (Zambia) and 62% (South Africa) in arm A
and 25% (Zambia) and 26% (South Africa) in arm B respectively, compared to standard of
care (arm C), with an effect increasing from one year to the next.
These results were based on parameter values that yielded HIV epidemics most closely
matching UNAIDS prevalence estimates. Exploring a variety of parameter sets which fitted
those relatively well, we found that very different combinations of parameter values relating
to the contact structure in the population could match the prevalence data (see Figures S3 to
S5 in File S1). This suggests that these data alone are not very informative about the
structure of contacts between the three risk groups, or the characteristics of those groups.
Despite those differences, we found that the predicted reduction in HIV incidence over three
years was relatively stable regardless of the parameter set chosen (see Figure S6 in File S1).
However, the reduction in 3-year cumulative incidence was highly dependent on the value of
process parameters such as the uptake of circumcision or testing, as illustrated in Figure 4.
We found a very strong linear dependence of the relative reduction in 3-year cumulative
HIV incidence on process parameters (adjusted R-squared >97% in both arms and both
countries). The coefficients of the regression were strikingly similar between countries,
although interestingly, a stronger influence of parameters related to circumcision was found
in Zambia, where we assumed only 13% of men are circumcised during adolescence, than in
South Africa, where we assumed 76% of men are circumcised during adolescence, as
measured in ZAMSTAR [38,39] (see Table 1).
Unsurprisingly, the major driver of the variability in the reduction in incidence was the
magnitude of community-level changes in sexual risk behavior in response to the
intervention activities (ibc), especially in arm B, where over 80% of the variance in the
outcome is explained by ibc. We emphasize that in the model, behavior change refers to an
overall response in the community, not to specific responses in individuals following ART.
While this behavior is an overall modifier that affects all individuals, it will be balanced by
counseling and treatment in those aware of their status. Therefore the principal manner in
which this community change affects incidence is through changed risk behaviors in
individuals who do not know their serostatus. In particular, changes towards more risky
behaviors could jeopardize the success of the trial and lead to an increase in incidence in
intervention arms compared to the control arm. On the other hand, protective changes in
sexual behavior resulting from the trial activities could increase the reduction in incidence
beyond that directly associated with the home-based testing, active linkage to care, ART and
circumcision interventions. However, based on previous experience, we do not expect major
changes in risk behavior during this trial, and our baseline scenarios reflect this assumption
The second driver of the variability in the reduction in cumulative incidence was the relative
infectivity of individuals under ART, which accounted for 34% (Zambia) and 35% (South
Africa) of the variance in cumulative incidence in arm A, and approximately 8% in arm B in
both countries. In fact, in arm A, the relative infectivity of individuals on ART was as
important as the community-level changes in sexual risk behavior in response to the
intervention activities, which accounted for 33% (Zambia) and 34% (South Africa) of the
variability in cumulative incidence.
Other important drivers included the uptake of testing and ART, the proportion of sex acts
with partners from outside of the community and the rate of drop-out from ART. The uptake
of circumcision was found to have little influence on the outcome in South Africa, but a
larger influence in Zambia, especially in arm B.
Although the intervention is planned to run for a 3-year duration, we looked at the impact of
an intervention extended to a 10-year horizon under the central target. We found a reduction
in 10-year cumulative incidence of 63% and 29% in arms A and B respectively in Zambia,
and 64% and 29% in South Africa, compared to standard of care (arm C). The relative
reduction in incidence for year 10 only would be 61% and 30% for arms A and B
respectively in Zambia, and 64% and 31% in South Africa.
We developed a deterministic compartmental model to predict the potential impact of the
intervention activities that will be undertaken in the HPTN 071 (PopART) trial, designed to
explore the potential population effect of universal home-based testing (in arm B) and
universal testing and treating (in arm A) on HIV incidence in large communities in Zambia
and South Africa. Our pre-trial modeling analysis predicts that if intervention targets are
reached, HIV incidence will decrease dramatically in both intervention arms, with a 3-year
cumulative reduction of 61 to 62% in arm A and 25 to 26% in arm B, relative to standard of
care (arm C), in both countries. Our model predicts that the reduction in cumulative HIV
incidence associated with home-based HIV testing (>25% over 3 years in arm B) will be
much greater than that effected by community-based HIV testing in a recent trial in sub-
Saharan Africa and Thailand (14% over 3 years, not statistically significant) .
In addition to projecting the overall impact of the complex intervention package for the
purposes of designing and ensuring adequate power in the trial, understanding, in this pre-
trial phase, the modifiable factors affecting the reduction in incidence is crucial to prioritize
allocation of human and financial resources to areas where the success of the intervention
could be threatened. We showed that the reduction in 3-year cumulative incidence in the
intervention arms compared to the control arm is almost linearly determined by a handful of
process parameters. This linear dependency suggests that monitoring a few parameters
during the course of the HPTN 071 (PopART), and other similar trials, should be enough to
assess its progress in real time and to increase targeted efforts if needed.
Unsurprisingly, we found that important threats to the trial success would be increases in
risky sexual behaviors at the population level in response to the trial activities and
secondarily the uptake of testing and ART as well as non-adherence to treatment. We
hypothesize that continued counseling, facilitated by the annual visits of the CHiP teams in
all households will be important to prevent increased risk behaviors and promote adherence.
The uptake of circumcision appeared to be a relatively important factor in determining
changes in incidence in Zambia, where the current circumcision levels are low, but less so in
the South African trial sites, located in the Western Cape region where circumcision is more
We found that once uptake of the intervention and adherence are ensured, minimizing delays
in linkage to care would favor the trial success, but might not be as crucial as could have
been anticipated, if those delays do not greatly exceed those we have explored here.
Importantly, 10-year model projections suggested that in the long term, prolonged
interventions similar to those proposed in arms A and B would allow to maintain HIV
incidence at lower levels, but not achieve elimination. This result is different to that of
Granich et al.  who found that annual incidence could be reduced below 1 per thousand
per year within only a few years. However, that model made much more optimistic
assumptions about the reduction in infectiousness for persons on ART (99% versus 90% in
our central target scenario) as well as uptake of universal testing and treatment (92% of
untreated persons per year versus 70% in our central target scenario), which we showed were
important determinants of the reduction in HIV incidence. Our model predictions had
previously been compared to predictions of eleven other models, in a study designed to
assess the influence of assumptions regarding HIV epidemiology on the predicted impact of
ART on HIV incidence . Although long-term predictions varied substantially across
models, our model was generally consistent with others, and rather conservative with regards
to the long-term reductions in incidence due to ART. Epidemiologic and service uptake data
collected during the trial, in conjunction with mathematical modelling, should improve the
accuracy and precision of future model projections and allow re-evaluation of the effort
required to achieve HIV elimination.
In the uncertainty analysis, we found that the relative reduction in incidence over the three
years of the trial was largely insensitive to input parameters (such as structure of the sexual
mixing matrix), despite great uncertainty on some of these parameters. However, such
uncertainty can affect the long term projections of interventions, as illustrated by the
variability in the predicted reduction in cumulative 10-year incidence in Arm A in South
Africa (see Figure S6 in File S1) when assuming that the intervention were extended to a 10-
year horizon. Similarly, improving the mechanistic representation of viral suppression, CD4
cell dynamics, and survival on ART, would probably affect the long-term projections,
although unlikely to change the short-term ones. Data collected during the trial, in particular
through the questionnaires administered in the population cohort and in planned case-control
studies, will help quantify some of those parameters directly, notably the parameters
describing the structure of contacts between and within risk groups. The prospect of
combining these granular survey and biomarker data with novel phylogenetic methods for
associating clusters of transmissions provides a unique opportunity to answer long-standing
epidemiological questions, such as the amount of transmission occurring during primary
HIV infection , patterns of sexual mixing, the geographic patterns of infection [46,47] or
the importance of core groups of highly transmissible or particularly at-risk individuals. This
will be crucial when trying to generalize the trial results to wider spatial and temporal scales.
Indeed, the HPTN 071 (PopART) trial will be performed at an unusually large scale, with
approximately 1.2 million individuals across all clusters and trial arms [11,14]. Therefore
this intervention could serve as a paradigm for routine implementation of universal testing
and treatment on a provincial or even national scale. The economic analysis of HPTN 071
will provide guidance on whether routine implementation is a valuable investment in the
health of populations, and is therefore an integral part of the trial. It will help policymakers
assess the costs and benefits of universal testing and treatment, in comparison to alternative
strategies. In a collaborative effort to estimate the cost-effectiveness of earlier ART
eligibility and expanded access to ART in low- and middle-income settings, both the
population-level health-benefits and particularly the implementation costs of earlier ART
eligibility and achieving high levels of access to early ART were identified as key
uncertainties and sources for caution in policy setting; the HPTN 071 trial is ideally placed
15?? Download full-text
to answer these questions. Many expected benefits of HPTN 071, including saved future
healthcare costs due to secondary infections averted, will occur after the trial. It is therefore
crucial to integrate the projected outcomes from the epidemiological model into an economic
model, with the objective of calculating the cost-effectiveness of HPTN 071 over different
The model used in this analysis relies on simplified representations of the complex dynamics
of HIV infection and the determinants of the spread of HIV. Data collected during the trial
will allow assessing the extent to which simplifications we have made, such as omitting age
structure, considering only heterosexual sex, disregarding the nature of sex, assuming similar
distribution of men and women amongst risk groups, or assuming independence between
risk group and propensity to have sexual contacts outside of the community, are reasonable.
The model also does not include selection and transmission of drug resistant strains of virus.
A more detailed model, informed by those data, will be developed during the trial and used
to help interpreting the trial results. This future model will also be able to account for how
potential changes in the national treatment guidelines in Zambia and South Africa and other
secular changes in the epidemic and the response to the epidemic affect the outcomes of the
trial and the course of these severe HIV epidemics.
Our analysis highlights the role that mathematical modeling can play in trial development
and monitoring, and more widely in evaluating the impact of treatment as prevention. In the
case of the HPTN 071 (PopART) trial, we showed that a reduction in 3-year cumulative
incidence by over 60% could be expected, but would require careful real-time monitoring of
the intervention uptake to ensure adequate program coverage.