BackgroundIn 2014, the Joint United Nations Programme on HIV/AIDS (UNAIDS) proposed new
targets directed at ending the AIDS epidemic, namely that, by 2020, 90% of all
HIV-infected people will know their HIV status; 90% of those diagnosed with HIV
infection will receive sustained combination antiretroviral therapy (ART); and 90%
of all people receiving ART will have viral suppression.1 The rationale underpinning these targets is related
both to the health benefit of ART to infected individuals,2,3 and to the
potent effect of ART on reducing sexual4-6 and
perinatal7,8 HIV transmission.
However, uncertainty remains as to whether these ambitious UNAIDS targets are
achievable, particularly in high-HIV-burden, resource-constrained settings such as
sub-Saharan Africa. According to UNAIDS, in sub-Saharan Africa in 2013, only 45% of
HIV-infected adults knew their HIV status; however, 86% of diagnosed persons were on
ART, and an estimated 76% of persons on ART achieved virologic
suppression.9 This
translates to 29% of all HIV-positive persons in sub-Saharan Africa having virologic
suppression, compared with the overall UNAIDS target of 73%. Recent estimates of
progress toward reaching this target range from 68% in Switzerland,10 62% in Australia,11 and 30% in the United
States,12 down to 9% in
Russia.13
Botswana is a middle-income country with a stable democracy, high HIV
prevalence (25·2% of persons 15–49 years),14 and a mature public ART program that started in
2002. Per current Botswana guidelines, HIV-infected citizens receive free 3-drug ART
from decentralized health clinics if they have a CD4 count ≤350
cells/mm3, a WHO stage III or IV illness (including recent
tuberculosis diagnosis), a history of cancer, or are pregnant or breastfeeding
(regardless of CD4 count). We have a unique opportunity to assess population-level
coverage of HIV testing, ART, and virologic suppression in the context of a large,
ongoing cluster-randomized combination prevention study that is underway in 30
communities across Botswana.
MethodsOverall BCPP study design and proceduresThe Botswana Combination Prevention Project (BCPP), also known as the
“Ya Tsie” study, is a pair-matched, cluster-randomized trial
funded by PEPFAR that is designed to test whether a package of combination
prevention interventions reduces population-level cumulative 3-year HIV
incidence. Another primary objective of the study includes evaluation of
population-level uptake of HIV testing and ART at baseline and over time; the
baseline cross-sectional results are presented in this manuscript. The trial is
being conducted in 30 communities in Botswana (15 pairs matched according to
size, pre-existing health services, population age structure, and geographic
location, including proximity to urban areas), with a total population of
approximately 180,000 persons (Figure 1),
representing nearly 10% of Botswana’s estimated population. Fifteen
communities were randomized to a combination prevention arm, and 15 to a
non-intervention arm. Interventions in the combination prevention arm include
home-based and mobile HIV testing and counseling; point-of-care CD4 testing;
linkage to care support; expanded ART (for CD4 351–500
cells/mm3 or CD4 >500 with HIV-1 RNA≥10,000
copies/mL, in addition to local criteria); and enhanced male circumcision
services. The protocol has been amended to offer universal treatment (regardless
of disease stage) in the intervention communities, and the Botswana Ministry of
Health (MoH) plans to provide universal ART in 2016 (a change that will also
apply to non-intervention communities). ART and other services are provided at
or above the evolving local standard of care, throughout the study.
Study populationSurvey participants were recruited using a household-based probabilistic
sampling strategy at the community level. Within each of the 30 communities, the
sampling strategy began with identifying and geocoding every plot with a
household-like structure using satellite imagery captured between 2012 and 2015
(Google Earth, Mountain View, California) overlaid with enumeration boundaries
used in the 2011 Botswana Census.15 A simple random 20% sample was then selected from the list
of geocoded plots. For each household on the selected plots, a household
representative (aged 18 years or older) was asked to list each household member
with their age, gender, time spent in the household and relationship to the
household head. Based on this information, research staff identified potentially
eligible household members and invited them to provide written consent to
participate, complete a questionnaire and undergo HIV testing. Eligibility
criteria included age 16–64 years; spending on average at least three
nights per month in the household; documentation of Botswana citizenship or
marriage to a Botswana citizen; and ability to provide informed consent.
Participants received a BWP20 (approximately USD$2) cellular telephone prepaid
voucher as compensation for their time. Plots were visited up to three times to
enumerate household members; up to three additional attempts were made to enroll
each enumerated household member.
Data collectionThe baseline household survey included questions about
sociodemographics, health, and HIV risk behavior. Persons who self-reported a
positive HIV status and who were able to provide corresponding documentation
(e.g. written test result, ART prescription) were not re-tested for HIV. All
other participants were offered counseling and rapid HIV testing according to
the Botswana-government– approved algorithm including KHB (KHB, Shanghai
Kehua Bio-Engineering Co Ltd, China) and Unigold (Trinity Biotech Plc, Ireland)
parallel HIV rapid tests. Documentation of ART receipt was sought for all
persons who reported that they were currently on ART (e.g. prescriptions;
clinical notes indicating ART receipt; or pills). HIV-1 RNA was tested in all
HIV-infected persons, regardless of ART status. HIV-1 RNA was performed using
the Abbott RealTime HIV-1 assay on the automated m2000 system (Abbott
Laboratories, Wiesbaden, Germany) (range 40–10,000,000 copies/mL) in the
Botswana–Harvard HIV Reference Laboratory, which participates in Virology
Quality Assurance and is accredited to ISO 17025. Point-of-care CD4 count (Pima,
Alere Inc., Waltham, Massachusetts) was obtained on HIV-positive persons who
were not currently taking ART. CD4 count and HIV-1 RNA results were shared with
participants. Research assistants participated in the point-of-care CD4 and HIV
testing external quality assurance program and were assessed for competence
using blinded testing.
HIV-infected participants not yet on ART were referred to their local
clinic for prompt ART initiation if their CD4 count was ≤350
cells/mm3 or if pregnant. HIV-infected participants with a CD4
count >350 cells/mm3 were referred to their local clinic for
evaluation for possible ART initiation, and for other services as needed,
including evaluation for tuberculosis.
After completion of the baseline survey, community-wide HIV testing and
counseling, linkage to care, expanded ART, and expanded male circumcision
interventions are started in the combination prevention communities. All
participants in the baseline survey will be re-contacted annually for three
years, for HIV testing (if previously negative) and for re-interview, with
ongoing interventions in the combination prevention arm during this period.
EthicsThe study protocol, informed consent, and other materials are approved
by the Botswana Health Research Development Committee (IRB of the Botswana MoH)
and the United States CDC IRB. All participants provided written informed
consent. Participants aged 16–18 years provided written assent (with
parents/guardians providing written permission). The study is registered at
ClinicalTrials.gov
(NCT01965470).
Statistical analysisWe restricted the present analysis to data collected from the baseline
household survey of all 30 study communities, prior to the roll-out of any
intervention activities.
The 90–90–90 targets were calculated as follows:
%diagnosed=[HIV-positive persons with documented
prior knowledge of HIV status][HIV-positive persons]
%on treatment=[HIV-positive persons with documented
prior knowledge of HIV statuscurrently receiving ART][HIV-positive persons with documented
prior knowledge of HIV status]
%virally suppressed=[HIV-positive persons with documented
prior knowledge of HIV statuscurrently receiving ART with HIV-1
RNA<400copies/mL][HIV-positive persons with documented
prior knowledge of HIV statuscurrently receiving ART]
%overall cascade=[HIV-positive persons with documented
prior knowledge of HIV statuscurrently receiving ART with HIV-1
RNA<400copies/mL][HIV-positive persons]
We used modified Poisson generalized estimating equations to obtain
prevalence ratios (PR), corresponding Huber robust standard errors and 95% Wald
confidence intervals (CI) to examine associations between individual
sociodemographic factors and a binary outcome indicating achievement of the
three individual and combined overall 90–90–90 targets.16 All analyses were conducted
using SAS software version 9·4 (SAS Institute, Cary, NC).
We conducted three additional analyses to assess the extent to which our
results might be sensitive to selection bias due to non-participation. First,
complete-case inverse-probability weighting was used to re-estimate each
90–90–90 target, adjusting for non-participation by accounting for
fully observed potential predictors common to non-participation and the
90–90–90 targets.17,18
Inverse-probability weighting adjusts for non-participation by empirically
breaking the association between observed predictors and non-participation,
allowing for unbiased estimation in the weighted sample provided a regression
for non-participation is correctly specified, and no unobserved correlates of
non-participation and outcomes defining the 90–90–90 targets
exist. Inverse-probability weights for participation were constructed using the
predicted probabilities from a multivariable logistic regression model
containing the following fully observed covariates and their two-way
interactions: age, gender, relationship to household head, community, and
presence of household member during enumeration. We then conducted a weighted,
complete-case crude analysis using modified Poisson generalized estimating
equations to estimate the marginal probability of each 90–90–90
target.
Although inverse-probability weighting adjusts for observed differences
between enrolled and unenrolled enumerated persons found to be eligible, the
enumerated population itself may systematically differ from the general
population. Therefore, in a second sensitivity analysis, we standardized our
observed 90–90–90 estimates to the age and gender distribution of
HIV-positive persons in Botswana using the 2011 Botswana Census15 and the 2013 Botswana AIDS
Indicator Survey.19
Finally, we performed a third sensitivity analysis which entailed
re-estimating the 90–90–90 targets by combining complete-case
outcomes observed among enrolled participants with potential outcomes of
un-enrolled, eligible household members, assuming a hypothetical scenario in
which non-participating residents had substantially lower rates of HIV testing,
ART coverage, and virologic suppression. Programming code is available upon
request by contacting the corresponding author.
Role of the funding sourceThe funders played no role in study design; in the collection, analysis,
or interpretation of data; in the writing of this report; or in the decision to
submit it for publication. The corresponding author, M. Essex, had full access
to all the data in the study and had final responsibility for the decision to
submit for publication.
ResultsThe baseline household survey was conducted in 30 communities from October
2013-November 2015. Research staff visited 13,135 randomly-selected plots identified
on satellite imagery and categorized 9,940 (76%) as residential and habitable. On
these plots, 11,582 households were identified, 10,306 (89%) were found to be
regularly occupied, and 7,696 (75%) were enumerated (Figure 2). Reasons for non-enumeration included absence of an eligible
household informant (10%), absence of any persons (7%), and refusal (6%). A total of
28,174 residents were enumerated; 15,475 (55%) were eligible for participation and
of these, 12,610 (81%) completed the baseline survey. Among the 15,475 enumerated,
eligible household members, 9,286 (60%) were women and 6,189 (40%) were men. Reasons
for non-participation among eligible persons included refusal (10%) and absence
(9%).
Of the 12,610 participants, 2,995 (24%) had documentation of prior positive
HIV test, 9,575 (76%) underwent HIV testing, and 40 (0·3%) refused testing.
Overall, 3,596 participants (including individuals known to be HIV-infected and
those newly testing positive for HIV) were HIV infected for an overall HIV
prevalence of 29%. HIV-1 RNA testing was completed in 99·7% of HIV-infected
participants.
Table 1 summarizes the baseline
characteristics of HIV-infected enrollees. Seventy-three percent were female (median
age 40 years, IQR 33–48 years). Eighty-seven percent (8,050 of 9,268) of
women residing in participating households enrolled, compared to 74% (4,560 of
6,189) of men residing in these households. The majority of participants reported
being single and nearly one-half had a primary school education or less.
Approximately one-third were employed and 56% reported no monthly income. More than
two-thirds of participants spent <1 week outside the community during the
previous 12 months.
Coverage of 90–90–90 targetsAmong the 3,596 HIV-infected persons identified, 2,995 (83·3%;
95% CI: 81·4%, 85·2%) already knew their positive status and
provided supporting documentation. A total of 466 participants newly tested
HIV-positive and 135 self-reported being HIV-positive but could not provide
documentation. Although these individuals were retested and found to be
positive, to be conservative, they were not included in the numerator of the
first 90–90–90 target.
Among the 2,995 HIV-infected persons who knew their positive status,
2,617 (87·4%; 95% CI: 85·8%, 89·0%) were receiving ART; an
additional 26 participants reported taking ART previously, but were not
currently on ART. The 2,617 participants currently receiving ART represented 95%
of the 2,754 persons eligible for ART per Botswana guidelines.
Finally, 2,517 (96·5%; 95% CI: 96·0%, 97·0%) of the
2,609 participants currently on ART with a viral load measurement had HIV-1 RNA
≤400 copies/mL; 2,428 (93·1%; 95% CI: 92·1%, 94·0%)
had HIV-1 RNA ≤40 copies/mL. Among the 2,635 persons who reported ever
starting ART for their health (including defaulters) with a viral load
measurement, 95·6% (95% CI: 95·0%, 96·3%) had HIV-1 RNA
≤400 copies/mL.
Overall, among all HIV-infected persons, 83·3% (95% CI:
81·4%, 85·2%) knew their status, 72·8% (95% CI:
70·1%, 75·5%) were currently receiving ART, and 70·2% (95%
CI: 67·5%, 73·0%) were currently on ART and had virologic
suppression (HIV-1 RNA ≤400 copies/mL) (Figure 3). Interestingly, 136 participants reported that they were
not currently receiving ART but had an HIV-1 RNA ≤400 copies/mL. To
confirm the ART status, we queried the Botswana MoH electronic medical records
systems for evidence of recent receipt of ART services. We were able to retrieve
records for 96 (71%) of these individuals and 40 (42%) were found to have
initiated ART prior to the survey. If we include these 40 additional
participants with HIV-1 RNA ≤400 copies/mL, the coverage estimates for
diagnosis, treatment and virologic suppression would increase to 83·3%,
88·7% and 96·5%, respectively (and 71·3% would have
achieved the overall 90–90–90 target).
Sensitivity analysesTable 2 presents the results of
three sensitivity analyses undertaken to examine the potential impact of
selection bias due to either non-participation among enumerated eligible
residents and/or differences between our enumerated sample and Botswana’s
general population. Adjustment by inverse probability weighting and
standardization resulted in an overall target estimate of 69·8% (95% CI:
67·1%, 72·6%) and 63·4% (95% CI: 61·6%,
65·1%), respectively (Table 2).
Finally, when we assumed that HIV prevalence was 50% higher and knowledge of HIV
status, ART coverage and virologic suppression were each 25% lower in
non-participants than what was observed among enrolled participants, the overall
target estimate was 63·8%.
Predictors of achieving 90–90–90 targetsTable 3 presents the univariable
associations between baseline characteristics and achievement of each individual
and overall 90–90–90 target. Male gender, younger age, being
single or never married, increased time spent outside the community, and higher
levels of education were significantly associated with lower levels of coverage
for the overall target. For example, HIV-infected participants aged <30
years were nearly one-half as likely to know their status, be on ART and achieve
viral suppression compared to those aged ≥60 years. Similarly, spending
>12 weeks outside the community and having higher than senior secondary
education were each associated with an approximate 20% decreased likelihood of
meeting the overall target (Table 3).
DiscussionWe found very high levels of diagnosis, treatment and viral suppression
among HIV-infected individuals in our study population: 83·3% of HIV-infected
persons knew their positive status, 87·4% of these individuals were receiving
ART, and 96·5% of persons receiving ART had virologic suppression. Overall,
70·2% of all HIV-infected persons had virologic suppression, compared with
the UNAIDS goal of 73%. Younger age was the strongest predictor of being
undiagnosed, not on ART and not virologically suppressed in our population, followed
by more time spent outside the community (>12 weeks/year). Women were more
likely to know their positive HIV status than men, potentially due to near universal
testing of pregnant women in Botswana.20
Very few countries, including those in North America and Europe, have
achieved similarly high coverage levels. In recent national-level analysis of HIV
treatment cascades by Levi et al, no country met the overall UNAIDS
goal of 73% virologic suppression among persons living with HIV.21 It may be informative for
countries and programs to evaluate their success in reaching the
90–90–90 targets, even if they have not yet moved to universal testing
and treatment.
Several factors likely contribute to Botswana's successful HIV
treatment program. Botswana was among the first high-HIV-burden countries to make
prevention of mother-to-child transmission services (starting in 2000) and ART
(starting in 2002 with expansion nationwide by 2006) available to citizens for free.
There has been a strong political will to speak openly about HIV in public forums
and destigmatize HIV testing. In 2004, former Botswana President Festus Mogae
introduced routine “opt-out” HIV testing and counseling in healthcare
settings. According to UNAIDS, the lack of widely available testing may constitute
the major impediment to achieving high ART coverage, particularly in sub-Saharan
Africa.1 With a widely
dispersed population, Botswana prioritized early decentralization of HIV treatment:
ART is available from more than 600 clinics nationwide alongside other health
services. Distance to services is correlated with retention in care22 and an estimated 95% of the
Botswana population resides within 10km of an ART-dispensing facility. Furthermore,
the program has included virologic monitoring for patients on ART since inception,
enabling adherence assessment and interventions. Many of these approaches could be
adopted by other countries/programs.
High rates of treatment coverage and virologic suppression should help
reduce HIV incidence. However, the most recent estimate of annual adjusted HIV
incidence in Botswana (1·35% in 2013) indicates substantial ongoing
transmission.19 This is
presumably due to ongoing transmission from the 30% of HIV-infected persons who
remain untreated in this setting of high HIV prevalence; furthermore, these data
(and the 90–90–90 targets) do not capture the complexities of sexual
networks, risk behavior patterns, and biologic factors that also contribute to
ongoing HIV transmission. Nevertheless, in totality, evidence from this and other
studies suggests that universal ART (regardless of CD4 count), as recently
recommended by the World Health Organization,23 will be important in decreasing new infections.
The main strength of this analysis is that it draws upon a large
population-based random sample of household residents from 30 geographically and
ethnically diverse communities across Botswana. Routinely collected programmatic
data cannot provide the actual number of persons living with HIV. Rather, this
figure is usually estimated indirectly from population distributions and HIV
prevalence figures. Similarly, programmatic data generally cannot directly estimate
the numbers of HIV-infected individuals who have been diagnosed with HIV. We were
able to collect data on both of these indicators, and to test for virologic
suppression in >99% of HIV-infected participants; viral load information is
infrequently available in resource-constrained settings.
Our study is subject to several limitations. First, approximately 25% of
potentially inhabited households were not enumerated, and nearly 20% of age- and
residency-eligible members of enumerated households were not enrolled. Botswana has
a highly mobile population, and the most common reasons for non-participation were
absenteeism and refusal. It is possible that persons not found in the household may
differ from enrolled participants with respect to diagnosis, treatment and virologic
suppression as defined by the 90–90–90 targets. For example, nearly
three quarters of our participants were female compared to 60% of household
residents. A weighted sensitivity analysis accounting for differences between
enumerated, unenrolled eligible household members and enrolled participants did not
substantially change our findings. Additionally, our estimates were derived from
smaller rural/peri-urban communities. It is not possible to know whether individuals
residing in other areas (including urban settings) would have similar HIV testing,
treatment and adherence rates. However, given earlier and more widespread
availability of HIV testing and treatment facilities, we anticipate coverage to be
higher in urban areas. Furthermore, according to Botswana national treatment and
UNAIDS14 statistics,
approximately 270,000 (~70%) of people believed to be living with HIV in
Botswana are on ART, which is similar to the 72·8% estimate in our study
population. According to the 2013 Botswana AIDS Indicator Survey, the estimated HIV
prevalence was slightly lower in rural/peri-urban villages such as those taking part
in BCPP (17·4%-18·7%) compared with prevalence in cities or towns
(19·5%-21·6%); HIV prevalence varies more dramatically by region
(11·1%-27·5%).19 A sensitivity analysis that standardized estimates of
90–90–90 targets to Botswana’s national HIV-infected population
did not substantially alter our findings. The overall target estimate after
standardization was 63·4% (95% CI: 61·6%, 65·1%), compared with
our observed estimate of 70·2% (95% CI: 67·5%, 73·0%). Lastly,
our eligibility criteria excluded several key populations, including adolescents and
children less than 16 years old, adults 65 years and older, and non-citizens. Given
our home-based recruitment approach (which may reach somewhat different populations
compared with those reached by other HIV testing strategies),24 our study also did not attempt to
enroll other key populations such as commercial sex workers. We estimate that
approximately 2% of enumerated residents were not Botswana citizens. This may be
important, because non-citizens are ineligible to receive ART free of charge in the
country.
Despite these limitations, our data represent among the most complete,
population-level estimates available from the region, and indicate very high rates
of coverage related to HIV testing, ART initiation, and virologic suppression. It is
particularly remarkable that Botswana has achieved such high rates of ART coverage
and virologic suppression given that the national CD4 threshold for ART eligibility
has been ≤350 cells/mm3; this could indicate a more
“mature” epidemic, and may differ in other settings. We found that 95%
of HIV-infected persons eligible for ART by current guidelines are already on ART.
These findings suggest that Botswana should reach and exceed the UNAIDS
90–90–90 targets sooner than 2020, if treatment eligibility changes to
include ART to all persons living with HIV irrespective of their CD4 count, assuming
the health system is able to test and treat those not yet on ART, and that persons
starting ART with higher CD4 count will exhibit reasonably high levels of
adherence.
Although Botswana is a middle-income country with a total population of
approximately 2 million, it has suffered from a very significant burden of HIV
disease. The high rates of HIV testing, ART, and virologic suppression in Botswana
provide good evidence that the UNAIDS targets are achievable elsewhere.