Tropical Medicine and International Health doi:10.1111/tmi.12386
volume 19 no 12 pp 1397–1410 december 2014
Retention and risk factors for attrition among adults in
antiretroviral treatment programmes in Tanzania, Uganda
and Zambia
Olivier Koole1,2, Sharon Tsui3,4, Fred Wabwire-Mangen5, Gideon Kwesigabo6, Joris Menten2, Modest Mulenga7,
Andrew Auld8, Simon Agolory8, Ya Diul Mukadi3, Robert Colebunders2,9, David R. Bangsberg10,11, Eric van
Praag3, Kwasi Torpey3, Seymour Williams8, Jonathan Kaplan8, Aaron Zee8 and Julie Denison3,4
1 Department of Infectious Disease Epidemiology, London School of Hygiene and Tropical Medicine, London, UK
2 Clinical Sciences Department, Institute of Tropical Medicine, Antwerp, Belgium
3 FHI 360, Durham, NC, USA
4 Department of International Health, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA
5 Infectious Diseases Institute, Makerere University College of Health Sciences, Kampala, Uganda
6 Muhimbili University of Health and Allied Sciences, Dar es Salaam, United Republic of Tanzania
7 Tropical Diseases Research Centre, Ndola, Zambia
8 Division of Global AIDS, United States Centers for Disease Control and Prevention, Atlanta, GA, USA
9 Epidemiology and Social Medicine, University of Antwerp, Antwerp, Belgium
10 Massachusetts General Hospital, Boston, MA, USA
11 Harvard Medical School, Boston, MA, USA
Abstract objectives We assessed retention and predictors of attrition (recorded death or loss to follow-up) in
antiretroviral treatment (ART) clinics in Tanzania, Uganda and Zambia.
methods We conducted a retrospective cohort study among adults (≥18 years) starting ART during
2003–2010. We purposefully selected six health facilities per country and randomly selected 250
patients from each facility. Patients who visited clinics at least once during the 90 days before data
abstraction were defined as retained. Data on individual and programme level risk factors for
attrition were obtained through chart review and clinic manager interviews. Kaplan–Meier curves for
retention across sites were created. Predictors of attrition were assessed using a multivariable Cox-
proportional hazards model, adjusted for site-level clustering.
results From 17 facilities, 4147 patients were included. Retention ranged from 52.0% to 96.2% at
1 year to 25.8%–90.4% at 4 years. Multivariable analysis of ART initiation characteristics found the
following independent risk factors for attrition: younger age [adjusted hazard ratio (aHR) and 95%
confidence interval (95%CI) = 1.30 (1.14–1.47)], WHO stage 4 ([aHR (95% CI): 1.56 (1.29–1.88)],
>10% bodyweight loss [aHR (95%CI) = 1.17 (1.00–1.38)], poor functional status [ambulatory aHR
(95%CI) = 1.29 (1.09–1.54); bedridden aHR1.54 (1.15–2.07)], and increasing years of clinic
operation prior to ART initiation in government facilities [aHR (95%CI) = 1.17 (1.10–1.23)].
Patients with higher CD4 cell count were less likely to experience attrition [aHR (95%CI) = 0.88
(0.78–1.00)] for every log (tenfold) increase. Sites offering community ART dispensing [aHR (95%
CI) = 0.55 (0.30–1.01) for women; 0.40 (0.21–0.75) for men] had significantly less attrition.
conclusions Patient retention to an individual programme worsened over time especially among
males, younger persons and those with poor clinical indicators. Community ART drug dispensing
programmes could improve retention.
keywords ART, HIV, retention, sub-Saharan Africa
people receiving antiretroviral treatment (ART) reached
Introduction
about 13 million in 2013. Sub-Saharan Africa achieved
At the end of 2013, two-thirds of the estimated 35 the greatest increase in ART coverage by reaching
million people globally living with HIV lived in 9 million people, to about 37% coverage (UNAIDS
sub-Saharan Africa (UNAIDS 2014). The number of 2014).
© 2014 John Wiley & Sons Ltd 1397
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
Corresponding with efforts to expand access to ART, ered to fulfil the sample size requirement of 250 patients
there has been an increasing emphasis on attaining the per study site. The site-selection process was conducted
high levels of retention and adherence necessary to in consultation with country-specific stakeholders includ-
achieve good clinical outcomes (Bangsberg et al. 2001; ing Ministries of Health (MOH) and United States Gov-
Hogg et al. 2002; Paterson et al. 2000; Nachega et al. ernment (USG) partner organisations.
2007). Retention is a critical determinant of adherence as
patients must actively attend and participate in an ART
Inclusion and exclusion criteria for medical chart review
care programme to receive their medication and to have
their HIV clinical indicators monitored. Therefore, reten- Patients who were at least 18 years of age at ART initia-
tion is a key indicator of programme quality (Giordano tion, who initiated ART treatment (a combination of 3
et al. 2007). antiretroviral drugs) at the site when free drugs were
However, retention of patients in ART care remains a available and who had started ART at least 6 months
major challenge in sub-Saharan programmes. Results prior to the data collection were included. Patients
from a meta-analysis of 32 studies from programmes in involved in other ART-related research were excluded.
sub-Saharan Africa showed that only 80% of patients
started on ART were still in care after 1 year, 77% after
Data collection and sampling
2 years and 72% after 3 years (Rosen et al. 2007; Fox &
Rosen 2010). Loss to follow-up (LTFU) and recorded During the period from April to July 2010, a retrospec-
death were the major causes of non-retention or attrition. tive chart review of 4500 medical charts was conducted.
This study is the first component of a study examining During study start-up, the study team noticed that one
retention and adherence to antiretroviral treatment facility in Tanzania consisted of two clinics that served
among adults in three countries in sub-Saharan Africa. In adults on ART: an adult-only clinic and a family clinic
this manuscript, we report the results of a retrospective where adults may seek care and treatment with their fam-
medical chart review of adult ART patients from ART ilies. One hundred and 25 patients from each of the
programme sites in Tanzania, Uganda and Zambia. The adult-only and the family clinics were selected.
objectives of the study were to characterise the level of The study team worked with sites to develop sampling
retention of patients on ART across multiple and differ- frames consisting of patient identification numbers (IDs).
ent programme settings and to examine the relationship The study data analyst used these sampling frames to
between individual and programme level characteristics generate the 250 random numbers using a computerised
and retention proportions. random number generator in Microsoft Excel that indi-
cated which patient charts would be abstracted. Data
abstractors at each site pulled the randomly selected
Methods charts and screened them for eligibility. They recorded on
a screening log if a chart was missing, ineligible, or eligi-
Design and study setting
ble and abstracted. Replacement numbers were generated
A retrospective review of 4500 randomly selected medical for those charts that were either ineligible or were miss-
records of ART patients from Tanzania, Uganda and Zam- ing after a minimum of three attempts to locate the chart
bia was conducted. In each country, six sites were purpose- over three consecutive weeks. Three sources of data were
fully chosen to explore the impact that different programme used: patient medical files, pharmacy logbooks and labo-
characteristics may have on retention and adherence out- ratory registers.
comes. This process resulted in study sites from different The study sites varied greatly in terms of how specific
levels in the health system (ranging from primary/commu- their sampling frames could be. Some sites had sampling
nity-based health centres to national referral hospitals), frames consisting only of adult patients who had started
from different types of health facilities (public sector, non- on ART at that site. Other sites could only produce sam-
governmental organisations (NGOs) or faith-based organi- pling frames with all patients (adults and children) ever
sations), from urban-rural locations, and with different registered in HIV care. The ability to focus the sampling
ART provision experiences and adherence strategies. frame depended on the site’s data management systems.
Study teams underwent 1 week of data abstraction
training and consisted of 2–3 data abstractors (mainly
Inclusion criteria for site selection
data entry staff and nurses from the facility) and one
Only sites with a minimum cohort size of 300 patients at supervisor (typically the sister-in-charge or lead doctor
the time of the protocol development (2006) were consid- managing the clinic). To better understand the
1398 © 2014 John Wiley & Sons Ltd
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
programme factors related to retention and adherence, tors presumed to be missing at random, conditional on
each study-site ART programme manager was inter- site and individual level predictors. Estimates were com-
viewed about their ART clinic model in June 2011. bined across imputed data sets according to Rubin’s rules
(Little 1992).
Data management and analysis
Sample size
Data were double-entered in a study database using Epi-
Data Entry 3.1 (EpiData Association, Odense, Denmark, A random sample of 250 ART-treated patients in each
Europe) at the in-country research organisation and then site was needed to allow the estimation of the retention
transferred to the central data office at FHI 360 for fur- proportion to be measured with a precision of 5% at site
ther cleaning and consistency checks. Consistency checks level, assuming the retention proportion at 6 months
across the different sources of data were performed. Data after ART initiation was at least 80%.
analysis followed a statistical analysis plan, which was fi-
nalised before the completion of the study database. All
Ethics statement
analyses were performed using Stata versions 10 and 12
(Stata Corporation, College Station, TX, USA). The study was approved by the Institutional Review
Attrition from the programme was defined as having Board (IRB) of the U.S. Centers for Disease Control and
no clinic, pharmacy or laboratory visit in the 3 months Prevention (US), FHI 360’s Protection of Human Subjects
prior to data collection among patients whose charts did Committee (US), the Muhimbili University of Health and
not indicate being transferred out or death. The primary Allied Sciences’ Research and Publications Committee
outcome variable was time to discontinuation from the (Tanzania), the National HIV/AIDS Research Committee
programme, defined as the time from start of first antiret- (Uganda), the Uganda National Council for Science &
roviral (ARV) drug treatment prescribed at the health Technology, the Tropical Diseases Research Centre’s Eth-
facility until the last visit. The time to discontinuation ics Review Committee (Zambia), the Ministry of Health
was analysed using survival data methods (Altman & (Zambia), the Massachusetts General Hospital’s IRB (US)
Bland 1998) where lost to follow-up and known deaths and Universitair Ziekenhuis Antwerpen (Belgium).
were considered the event of interest. Patients who were As the study was a retrospective chart review with
retained or transferred out were censored. minimal risks, and requesting consent would mean inclu-
Retention proportions were estimated using the Kap- sion of only retained patients, the IRBs waived informed
lan–Meier method (Pocock et al. 2002). The analysis of consent requirements to abstract participant charts. Writ-
risk factors for attrition was performed using Cox-pro- ten informed consent was required prior to interviewing
portional hazards regression, with a shared frailty term the ART clinic managers.
to account for clustering of outcomes at each site (Houg-
aard 1995). The proportional hazards assumption of pre-
Role of the funding source
dictors was assessed using graphical methods. A multiple
Cox-regression model was constructed using a hierarchi- This research was supported by a contract with FHI 360
cal approach. This approach included first individual by the U.S. Centers for Disease Control and Prevention
level (P-value <0.1) and subsequently programme level (CDC) and the Health Resources & Services Administra-
(P-value <0.2) characteristics. Variables that were missing tion (HRSA) with funds from the President’s Emergency
in more than 30% of subjects or were not available at all Plan for AIDS Relief (PEFPAR). CDC provided technical
sites or were present in a small minority of patients input into the study design, data collection, data analysis,
(<10%) were excluded from the multiple Cox-regression data interpretation and writing of the manuscript.
model. The model was simplified using Akaike’s Informa-
tion Criterion, retaining predictors and clinically plausible
Results
interaction terms that increased model fit, with a penali-
sation for increasing model complexity (Collet 2003). A total of 7755 patient medical files were screened for
Confidence intervals for effects were estimated using the eligibility at the participating sites. Of these, 1951 were
Wald method and P-values using likelihood ratio tests. ineligible, 1310 files were missing, and 4494 files were
The final Cox-regression model was corrected for missing abstracted. During preparation for analysis, 84 duplicate
data in the predictors of attrition using Multivariate files were excluded and another 22 were excluded for not
Imputation by Chained Equations (Little 1992; Royston meeting the eligibility criteria. One site in Zambia used
2004). Ten data sets were imputed with baseline predic- an incorrect sampling frame (excluding patients who
© 2014 John Wiley & Sons Ltd 1399
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
were known dead or lost to follow-up) which led to the per person (IQR: 0.7–3.2). The retention proportion per
exclusion of that site’s 241 patients, leaving 4147 site ranged from 58.7% to 99.2% at 6 months, from
patients for the final analysis. 52.0% to 96.2% at 1 year, from 39.7% to 93.8% at
2 years, from 32.7% to 90.4% at 3 years and from
25.8% to 90.4% at 4 years (Figure 1). Among the 1312
Characteristics of the study population
non-retained patients, 260 (19.8%) were known to have
Patients started ART between 2003 and 2010. The median died. The remaining 1052 were lost to follow-up.
age at ART initiation was 36 years (interquartile range
(IQR): 30–42), and 2670 (64.4%) were female. Three quar-
Predictors for attrition
ters of patients (3141 or 75.7%) had a baseline CD4 cell
count, with the median CD4 cell count of 134 cells/ll During univariate analysis, significant associations for
(IQR: 63–206). About half of patients (2197 or 53.0%) attrition were found for the following baseline character-
were inWHO stage 3 or 4. A vast majority were started on istics: younger age (<30 years), male sex, further distance
one of the four traditional non-nucleoside reverse transcrip- to the clinic, increasing years of clinic operation prior to
tase inhibitor (NNRTI)-based regimens (3598 or 86.8%) ART initiation, a higher WHO stage, weight loss of
and about 11% on a tenofovir-based regimen. Other char- >10% of body mass, wasting syndrome (weight loss of
acteristics, stratified by country, are presented in Table 1. >10%, unexplained chronic diarrhoea >1 month and
Characteristics of the patients whose charts were miss- unexplained fever >1 month), a lower CD4 cell count, a
ing could not be described as the sampling frame con- lower total lymphocyte count, a lower haemoglobin
sisted of patient identifiers only, and although date of count and a poorer functional status (Table 4).
birth and sex were recorded on the study screening logs, All programme characteristics described in Table 2
the information could not be collected for patients whose were examined for association with attrition. Univariate
charts were missing. analysis found that the level and type of health facility
and the dispensing of ARV drugs at the community level
were significant predictors of attrition. Participants from
Programme characteristics
primary/community-based facilities experienced lower
Eighteen sites were included in the analysis, seven sites in proportions of attrition. The same was true of faith-based
Tanzania (the facility consisting of two different models and NGO-based facility participants. Facilities that had
of care was considered as two separate sites), six sites in community-based ARV drug dispensing also experienced
Uganda and five sites in Zambia (one site was excluded lower proportions of attrition. We confirmed the effects
because of using an incorrect sampling frame). of programme characteristics identified in the analyses
Half of the health facilities were government facilities. across countries by assessing these effects in each country
Non-governmental facilities were either faith-based or run separately. During the stratified analysis by country, attri-
by a non-religious non-governmental organisation (NGO). tion was found to be significantly worse for governmental
However, the level and type of health facilities were not programmes in Tanzania and Zambia, but not in
evenly distributed across the three countries (with more Uganda. Country was also associated with retention, with
primary health facilities and NGO-supported facilities in considerably less attrition found in Uganda compared
Uganda). Two-thirds of the sites were located in an urban with Tanzania and Zambia (Table 5). This was explained
setting. At the time of data extraction, 8 sites had less than by the fact that three of the four sites with community-
2000 ART patients (range: 350–1967), 6 sites had between based ARV distribution were based in Uganda. Country
2000 and 4000 patients (range: 2095–3989) and 4 had was not formally considered as a site-level predictor as
more than 4000 patients on ART (range: 4807–7471). sites were not necessarily representative for the country
Other characteristics, stratified by country, are presented as a whole, and including country as site-level predictor
in Table 2. Four programmes used community-based dis- in the final model was not possible due to computational
tribution of ARV drugs for stable patients. An overview of difficulties.
the modalities of community-based ARV distribution According to the predefined statistical analysis plan,
among the study sites is given in Table 3. distance to the clinic was not included in the multivari-
able Cox-regression model building because this variable
was missing in two of the research sites. Other known
Retention proportions
predictors (Coetzee et al. 2004) excluded due to missing
The total period of follow-up by 4147 patients was data include lymphocyte count, haemoglobin level (more
8378.5 years, with a median follow-up time of 1.7 years than 30% missing data), oral candidiasis and wasting
1400 © 2014 John Wiley & Sons Ltd
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
Table 1 Patient characteristics at ART initiation in multicountry retention study
Total number of
Tanzania Uganda Zambia participants
Characteristic (n = 1458) (n = 1472) (n = 1217) (N = 4147)
Demographics
Age (years): median (IQR) 37 (31–43) 35 (30–41) 35 (30–42) 36 (30–42)
Age 18–29 years: n (%) 266 (18.2) 344 (23.4) 291 (23.9) 901 (21.7)
Age ≥30 years: n (%) 1191 (81.7) 1122 (76.2) 923 (75.8) 3236 (78.0)
Gender: n (%)
Male 484 (33.2) 504 (34.2) 488 (40.1) 1476 (35.6)
Female 974 (66.8) 968 (65.8) 728 (59.8) 2670 (64.4)
Distance from clinic: n (%)
<10 km 631 (43.3) 1073 (72.9) 212 (17.4) 1916 (46.2)
10–25 km 251 (17.2) 185 (12.6) 15 (1.2) 451 (10.9)
>25 km 461 (31.6) 139 (9.4) 13 (1.1) 613 (14.8)
Missing 115 (7.9) 75 (5.1) 977 (80.3) 1167 (28.1)
ART related
Year of ART initiation: n (%)
2003–2004 36 (2.5) 54 (3.7) 51 (4.2) 141 (3.4)
2005 232 (15.9) 226 (15.4) 188 (15.5) 646 (15.6)
2006 278 (19.1) 229 (15.6) 278 (22.8) 785 (18.9)
2007 287 (19.7) 337 (22.9) 336 (27.6) 960 (23.2)
2008 360 (24.7) 296 (20.1) 212 (17.4) 868 (20.9)
2009–2010 265 (18.2) 330 (22.4) 152 (12.5) 747 (18.1)
Years of clinic operation prior to ART initiation: n (%)
<1 year 286 (19.6) 183 (12.4) 72 (5.9) 541 (13.1)
≥1 to <2 years 312 (21.4) 196 (13.3) 196 (16.1) 704 (17.0)
≥2 to <3 years 272 (18.7) 325 (22.1) 305 (25.1) 902 (21.8)
≥3 to <4 years 341 (23.4) 304 (20.7) 277 (22.8) 922 (22.2)
≥4 to <5 years 196 (13.4) 207 (14.1) 211 (17.3) 614 (14.8)
≥5 years 51 (3.5) 257 (17.5) 156 (12.8) 464 (11.2)
Prior ART experience: n (%) 112 (7.7) 85 (5.8) 55 (4.5) 252 (6.1)
Prior exposure to NVP for PMTCT: n (%) 0 (0.0) 86 (8.9) 94 (12.9) 180 (6.7)
Starting ART regimen: n (%)
D4T-3TC-NVP 991 (68.0) 540 (36.7) 404 (33.2) 1935 (46.7)
D4T-3TC-EFV 86 (5.9) 23 (1.6) 79 (6.5) 188 (4.5)
ZDV-3TC-NVP 128 (8.8) 542 (36.8) 240 (19.7) 910 (21.9)
ZDV-3TC-EFV 230 (15.8) 257 (17.5) 96 (7.9) 583 (14.1)
TDF-3TC/FTC-NVP/EFV 7 (0.5) 93 (6.3) 354 (29.1) 454 (11.0)
PI based 2 (0.1) 13 (0.9) 8 (0.7) 23 (0.6)
Other 1 (0.1) 1 (0.1) 33 (2.7) 35 (0.8)
Missing/non-sensical 13 (0.9) 3 (0.2) 3 (0.3) 19 (0.5)
CTX prophylaxis: n (%) 981 (67.3) 1186 (80.6) 724 (59.5) 2891 (69.7)
Clinical characteristics
WHO clinical stage at start: n (%)
WHO stage 1–2 301 (20.6) 639 (43.4) 394 (32.8) 1334 (32.2)
WHO stage 3 539 (37.0) 508 (34.5) 553 (45.4) 1600 (38.6)
WHO stage 4 325 (22.3) 160 (10.9) 112 (9.2) 597 (14.4)
Missing 293 (20.1) 165 (11.2) 158 (13.0) 616 (14.9)
Functional status: n (%)
Working/active 748 (51.3) 904 (61.4) 488 (40.1) 2140 (51.6)
Ambulatory 261 (17.9) 73 (5.0) 352 (28.9) 686 (16.5)
Bedridden 31 (2.1) 13 (0.9) 71 (5.8) 115 (2.8)
Missing 418 (28.7) 482 (32.7) 306 (25.1) 1206 (29.1)
Laboratory parameters
CD4 (cells/ll): median (IQR) 133 (59–206) 136 (65–202) 134 (67–217) 134 (63–206)
© 2014 John Wiley & Sons Ltd 1401
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
Table 1 (Continued)
Total number of
Tanzania Uganda Zambia participants
Characteristic (n = 1458) (n = 1472) (n = 1217) (N = 4147)
Missing: n (%) 337 (23.1) 303 (20.6) 366 (30.1) 1006 (24.3)
TLC (cells/ll): median (IQR) 1700 (1100–3000) 1400 (980–1970) 1200 (500–1900) 1400 (900–2230)
Missing: n (%) 903 (61.9) 1151 (78.2) 695 (57.1) 2749 (66.3)
Haemoglobin (g/dl) 10.2 (8.9–11.8) 11.9 (10.5–13.2) 10.6 (9.0–12.1) 11.0 (9.5–12.5)
Missing: n (%) 683 (46.8) 604 (41.0) 512 (42.1) 1799 (43.4)
Opportunistic infections
Weight loss >10%: n (%) 310 (21.3) 96 (6.5) 435 (35.7) 841 (20.3)
Chronic diarrhoea >1 month: n (%) 159 (10.9) 41 (2.8) 282 (23.2) 482 (11.6)
Fever >1 month: n (%) 287 (19.7) 80 (5.4) 246 (20.2) 613 (14.8)
Oral candidiasis: n (%) 103 (7.1) 76 (5.2) 90 (7.4) 269 (6.5)
Wasting syndrome: n (%) 135 (9.3) 34 (2.3) 40 (3.3) 209 (5.0)
Pulmonary TB: n (%) 171 (11.7) 151 (10.3) 169 (13.9) 491 (11.8)
ART, antiretroviral treatment; IQR, interquartile range; PMTCT, prevention mother-to-child transmission; NVP, nevirapine; 3TC, lam-
ivudine; D4T, stavudine; ZDV, zidovudine; EFV, efavirenz; TDF, tenofovir, FTC, emtricitabine; PI, protease inhibitor; CTX, cotrimox-
azole; TLC, total lymphocyte count; TB, tuberculosis.
Working/active: able to perform usual work in or out of the house; ambulatory: able to perform activities of daily living but not able
to work; bedridden: not able to perform activities of daily living.
Wasting syndrome: weight loss of >10%, unexplained chronic diarrhoea >1 month and unexplained fever >1 month.
Data are missing for age and gender when the total number of patient was less than 4147.
(present in <10% of the patients). The final multiple ularly strong among males, with males and females at
Cox-regression model, retaining only significant predic- facilities that offered community-based distribution hav-
tors and interactions, and correcting for missing data ing similar attrition proportions [aHR (95%CI) = 0.95
using multiple imputations, revealed that patients of (0.67–1.33)]. At sites without community distribution
younger age (<30 years) were at higher risk of attrition of ART, however, males had a higher attrition risk
compared with older patients (≥30 years) [adjusted haz- than female [aHR (95% CI) = 1.33 (1.18–1.50)]. In
ard ratio (aHR) for attrition, (95% confidence interval) addition, government run facilities compared with
(95% CI) = 1.30 (1.14–1.47)]. Patients with baseline faith-based or NGO facilities were found to signifi-
WHO clinical stage of 3 or 4 had higher proportions of cantly predict attrition. No significant difference in
attrition compared with patients in stages 1 and 2 [aHR retention was observed between government and
(95%CI) = 1.12 (0.95–1.35) and 1.56 (1.29–1.88), faith-based or NGO facilities during the first year of
respectively]. The same was true for ambulatory (able to operation. Attrition significantly increased over time
perform daily activities but not working) and bedridden in government facilities [aHR/year (95%CI) = 1.17
(not able to perform daily activities) (WHO 2006) (1.10–1.23)], but not in faith-based or NGO facilities
patients compared with working or active patients (able [aHR/year (95% CI) = 1.03 (0.95–1.11)] resulting in an
to perform usual work in or out of the house) [aHR overall lower retention in government hospitals
(95%CI) = 1.29 (1.09–1.54) and 1.54 (1.15–2.07), (Table 6).
respectively]. Patients with a loss of more than 10% of A significant association was found between
body mass were at greater risk of attrition [aHR (95% attrition and the number of randomly selected patients
CI) = 1.17 (1.00–1.38)]. The probability of attrition for whom the patient chart could not be located. Sites
decreased proportionally with an increase in CD4 count that had more missing records had less attrition
[aHR (95%CI) = 0.88 (0.78–1.00) for every log (tenfold) (Table 5). Correcting the multiple Cox-regression model
increase]. for the percentage of selected records which were
At the programme level, community-based dispensing missing, by including this variable as a covariate,
of ARV drugs was significantly related to less attrition did not significantly change the effect estimates for the
[aHR (95%CI) = 0.55 (0.30–1.01) for women and 0.40 predictors included in the final model (data not
(0.21–0.75) for men] (Figure 2). This effect was partic- shown).
1402 © 2014 John Wiley & Sons Ltd
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
Table 2 Site characteristics in multicountry retention study
Total
number of
Tanzania Uganda Zambia sites
Characteristic (n = 7) (n = 6) (n = 5) (N = 18)
General information
Level of health facility
National referral hospital 2 1 1 4
Provincial/regional Hospital 2 0 2 4
District hospital 3 1 2 6
Primary/community-based health care 0 4 0 4
Type of health facility
Government 4 1 4 9
Mission facility 3 1 1 5
Non-religious NGO 0 4 0 4
Setting
Urban 4 5 3 12
Rural/peri-urban 3 1 2 6
ART-related information
Year ART was started at facility
2003 1 2 2 5
2004 3 2 3 8
2005 2 2 0 4
2006 0 0 0 0
2007 1 0 0 1
Number of adults currently on ARVs
<2000 6 1 1 8
2000–4000 1 4 1 6
>4000 0 1 3 4
Home-based care
No 0 3 4 7
Yes 7 3 1 11
Physician-based care
No 2 1 0 3
Yes 5 5 5 15
ARV-dispensing characteristics
Buddy needed for ART initiation
No 0 0 3 3
Yes 7 6 2 15
Three counselling sessions needed for ART initiation
No 1 2 1 4
Yes 6 4 4 14
Visit frequency after 6 months on ARVs
Monthly 5 1 0 6
Every 2 months 0 4 4 8
Every 3 months 2 1 1 4
Community-based distribution of ARVs
No 7 3 4 14
Yes 0 3 1 4
clinical predictors. This study makes an important contri-
Discussion
bution to our understanding of ART retention by exam-
To date, most studies examining retention to ART care ining not only retention proportions across three
and treatment programmes focus on individual pre-ART countries and 18 study sites, but by going beyond
© 2014 John Wiley & Sons Ltd 1403
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
Table 3 Community-based ARV drug distribution among study clinics in Tanzania, Uganda and Zambia
Community distribution of ARVs: any dispensing of ARVs happening outside the regular clinic, covering models where patients are
only picking up their ARV drugs from a mobile point to models with mobile health posts with clinical check-up and adherence
counselling
Programme and type of facility Activities
Programme 1: Non-governmental Mobile clinic at community drug dispensing points on specific days
organisation ARV drug and non-ARV drug pickup
Clinical investigation (patient monitoring), phlebotomy and adherence counselling
Referral of complicated cases
Programme 2: Faith-based Mobile clinics at peripheral (non-ART) health centres and makeshift community clinics
organisation on specific days
ARV drug and non-ARV drug pickup
Clinical investigation (patient monitoring), phlebotomy and adherence counselling
In addition: ARV drug distribution door to door to stable patients by community ART and
TB treatment supporters for specific patients (patients whose work/study schedule does not
allow them to visit the clinic)
Programme 3: Government Mobile clinics at peripheral (non-ART) health centres on specific days
ARV drug and non-ARV drug pickup
Referral for clinical investigations, phlebotomy and adherence counselling
Programme 4: Faith-based Mobile clinics at peripheral (non-ART) health centres on specific days
organisation ARV drug and non-ARV drug pickup
Clinical investigation (patient monitoring), phlebotomy and adherence counselling
Overall, retention proportions varied widely both
across countries and study sites (25.8% to 90.4% at year
4 for example). These results are comparable to those
from other studies in sub-Saharan Africa settings (Coet-
zee et al. 2004; Ferradini et al. 2006; Calmy et al. 2006;
Weigel et al. 2012). These studies exemplify the chal-
lenges of defining retention in different settings and sys-
tematically accessing information in clinics with different
data collection systems. Retention proportions are also
greatly affected by the choice of LTFU definition (Shep-
herd et al. 2013).
Many of the baseline clinical characteristics predictive
of attrition reinforce findings from other studies in sub-
Saharan Africa, including younger age (<30 years), being
male, having a higher WHO clinical stage, weight loss of
Tanzania >10% of body mass, a lower CD4 cell count and a
Uganda poorer functional status (Coetzee et al. 2004; Ferradini
Zambia et al. 2006; Calmy et al. 2006). These findings reaffirm
the need for early identification of HIV-infected individu-
0 6 12 24 36 als and early initiation of ART. Increasing years of clinic
Months since start ART operation, prior to when a patient initiated ART, was
also an independent risk factor for attrition. This finding
Figure 1 Kaplan-Meier estimates by site in Tanzania, Uganda
and Zambia. has been confirmed by other studies (Braitstein et al.
2006; Cornell et al. 2010). However, this effect was
mainly observed in government facilities and was not sig-
individual baseline clinical predictors of attrition to nificant in facilities run by faith-based organisations or
examine the potential effect different programme charac- NGOs. Rapid scaling-up may have considerably increased
teristics may have on retention. the workload for government health workers. This in
1404 © 2014 John Wiley & Sons Ltd
Percentage retained
0 20 40 60 80 100
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
Table 4 Individual predictors of attrition in multicountry retention study
Retention proportion
Hazard ratio
N 1 year 2 years 3 years (95% CI) P-value†
Total 4147
Demographics
Age at start ART
18–29 years 901 77.4 69.7 62.8 1 0.001
≥30 years 3236 79.1 71.7 67.0 0.81 (0.71, 0.92)
Sex
Female 2670 80.6 73.6 68.5 1 <0.001
Male 1476 75.1 67.1 61.4 1.26 (1.13, 1.41)
Distance to clinic (/10 km) – – – – 1.03 (1.01, 1.05) 0.007
ART-related and other treatment related predictors
Prior ART experience
No 3895 78.8 71.4 66.1 0.86 (0.68, 1.08)
Yes 252 76.9 68.5 63.5 1 0.187
Prior exposure to NVP for PMTCT
No 1753 79.9 73.0 67.3 1 0.326
Yes 180 92.1 84.5 80.0 0.78 (0.53, 1.13)
Missing 738 75.1 67.1 61.4 0.92 (0.77, 1.11)
Years since ART started at programme (/year) – – – – 1.10 (1.05, 1.15) <0.001
TB treatment
No 3327 80.3 73.4 68.9 1 0.131
Yes 386 75.8 67.6 59.0 1.11 (0.92, 1.33)
Missing 434 68.8 58.8 51.1 1.17 (0.99, 1.39)
CTX prophylaxis
No 693 80.4 73.3 68.5 0.96 (0.82, 1.12)
Yes 2891 76.3 69.2 62.7 1 0.810
Missing 563 72.8 63.4 57.7 1.02 (0.86, 1.20)
Clinical Characteristics at ART start
WHO stage at start ART
I & II 1334 86.9 80.0 73.6 1 <0.001
III 1600 78.5 70.8 65.8 1.20 (1.03, 1.39)
IV 597 62.5 55.5 52.4 1.98 (1.66, 2.37)
Missing 616 76.6 68.8 62.9 1.29 (1.07, 1.55)
Functional status
Working/active 2140 84.0 77.8 72.8 1 <0.001
Ambulatory 686 66.2 54.6 48.4 1.69 (1.45, 1.97)
Bedridden 115 51.4 47.4 44.7 2.61 (2.00, 3.41)
Missing 1206 78.9 71.5 66.0 1.28 (1.10, 1.50)
CD4 (log) – – – – 0.64 (0.49, 0.84) <0.001
TLC (cells/ll)
<1200 cells/ll 581 72.7 65.0 58.2 1.21 (1.00, 1.46)
≥1200 cells/ll 817 76.6 68.7 63.7 1 <0.001
Missing 2749 80.6 73.4 68.3 0.86 (0.74, 1.01)
Haemoglobin (g/dl)
<10 g/dl 803 69.3 61.9 57.3 1.43 (1.23, 1.67)
≥10 g/dl 1545 84.1 77.3 72.1 1 <0.001
Missing 1799 78.2 70.3 64.7 0.93 (0.81, 1.08)
Weight loss >10%
No 3306 80.3 73.2 67.8 1 <0.001
Yes 841 72.1 63.8 58.7 1.32 (1.13, 1.54)
Chronic diarrhoea >1 month
No 3665 79.3 71.8 66.2 1 0.651
Yes 482 74.0 67.4 64.1 1.04 (0.87, 1.26)
© 2014 John Wiley & Sons Ltd 1405
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
Table 4 (Continued)
Retention proportion
Hazard ratio
N 1 year 2 years 3 years (95% CI) P-value†
Fever >1 month
No 3534 79.9 72.7 67.0 1 0.062
Yes 613 71.7 63.1 59.8 1.16 (0.99, 1.36)
Oral candidiasis
No 3878 79.1 71.4 66.4 1 0.051
Yes 269 72.2 69.3 60.7 1.24 (1.00, 1.54)
Wasting syndrome
No 3938 79.7 72.2 66.8 1 <0.001
Yes 209 59.1 54.0 50.5 2.0 (1.56, 2.57)
Pulmonary TB
No 3656 79.0 71.4 66.2 1 0.965
Yes 491 76.3 69.9 64.2 1.00 (0.85, 1.19)
ART, antiretroviral treatment; CI, confidence interval; PMTCT, prevention mother-to-child transmission; NVP, nevirapine; CTX, co-
trimoxazole; TLC, total lymphocyte count; TB, tuberculosis.
Wasting syndrome: weight loss of >10%, unexplained chronic diarrhoea >1 month and unexplained fever >1 month.
Data are missing for age and gender when the total number of patient was less than 4147.
†P-value from univariate Cox regression models describing the effects of each individual predictor without correction for other factors,
adjusting for site using shared frailty methods.
turn may have compromised the organisation of services Cornell et al. 2009), being male is an independent risk
and quality of care provided. Faith-based and NGO facil- factor for attrition. However, in this study, this difference
ities might have had better coping mechanisms and fund- between the sexes was only observed in sites without
ing to increase staff levels and to adapt their services and community distribution of ARV drugs, where males were
monitoring/tracking systems to the increasing numbers of 30% more likely than females to experience attrition. In
ART patients. The association between attrition and sites with community ARV drug distribution, attrition
older governmental programmes could also be partly proportions among both men and women were about
explained by misclassification of LTFU which, in reality, 50% smaller compared with women in sites without
consists of unreported (silent) transfer to care elsewhere community distribution (the group with the lowest attri-
(Geng et al. 2010). Initially, only hospital-based referral tion).
centres provided ART treatment, but in the setting of Greater retention in community-based ART pro-
rapid scale-up, patients often transferred to closer lower- grammes may be due to fewer patients transferring out
level facilities (Bedelu et al. 2007; Chan et al. 2010). because many of the transfers seen to date are from ini-
Retention in an ART programme reflects a number of tial centres to community programmes (Bedelu et al.
heterogeneous outcomes including mortality, LTFU and 2007; Chan et al. 2010). As noted above, these urban
transfer of care (both silent and recorded). Geng et al. centres probably had substantial unrecorded (silent)
(2010) found that among 14 studies where outcomes in transfers with the rapid scale-up and decentralisation of
some patients LTFU were reported, about 50% were in ART services (Geng et al. 2010).
care elsewhere. This finding highlights the importance of What are the implications of these data regarding com-
examining not only programme retention to specific ART munity-based distribution of ARV drugs? Only four sites
clinics, but also retention to care regardless of where the in this study had a system of community-based ARV drug
services are rendered. distribution. These systems varied from models where
The most important result of this study is that sites patients only pick up their ARV drugs from a mobile
offering ARV drug dispensing in the community had sig- point to models with mobile health posts with clinical
nificantly greater programme retention. Of particular check-up and adherence counselling. However, most of
importance was the effect of community-based ARV drug these programmes still depended heavily on the support
distribution on retention of men. As well established in of clinic-based staff for community-based ARV drug dis-
the literature (Geng et al. 2010; Ferradini et al. 2006; tribution.
1406 © 2014 John Wiley & Sons Ltd
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
Table 5 Programme level predictors of attrition in multicountry retention study
Retention proportion
Nr Nr
Sites Patients 1 year 2 years 3 years Hazard ratio (95% CI) P-value*
Country
Tanzania 7 1458 71.0 62.7 58.3 1 0.005
Uganda 6 1472 90.5 85.3 81.5 0.35 (0.19, 0.62)
Zambia 5 1217 73.2 64.3 56.8 1.05 (0.57, 1.92)
General health facility characteristics
Level of health facility
National referral hospital 4 749 76.9 67.3 61.8 1 0.043
Provincial/regional hospital 4 980 71.7 63.5 55.6 1.25 (0.57, 2.74)
District hospital 6 1432 75.2 67.1 63.0 1.17 (0.57, 2.39)
Primary health centre/community based 4 986 91.8 87.8 83.4 0.37 (0.17, 0.82)
Type of health facility
Government 9 2188 71.0 61.6 55.1 1 0.007
Mission (faith based) 5 973 85.5 80.5 75.9 0.44 (0.24, 0.79)
Non-religious NGO 4 986 88.8 83.0 79.1 0.35 (0.19, 0.67)
Setting
Rural/peri-urban 6 1441 74.2 66.7 62.7 1 0.383
Urban 12 2706 81.0 73.7 67.8 0.74 (0.38, 1.45)
Number of adults on ARVs
<2000 8 1703 74.4 66.7 62.7 1 0.726
2000–4000 6 1474 83.7 77.3 72.1 0.74 (0.36, 1.53)
>4000 4 970 78.4 70.1 61.1 0.93 (0.41, 2.11)
Home-based care
No 7 1732 77.2 68.7 62.0 1 0.625
Yes 11 2415 79.7 73.1 69.0 0.85 (0.44, 1.63)
Physician-based care
No 3 711 72.5 63.3 58.4 1 0.308
Yes 15 3436 79.9 72.8 67.5 0.66 (0.28, 1.51)
ARV-dispensing Characteristics
Buddy needed for ART initiation
No 3 744 67.0 56.6 46.6 1 0.078
Yes 15 3403 81.2 74.6 70.5 0.50 (0.23, 1.11)
Three counselling sessions needed for ART initiation
No 4 871 81.6 74.1 69.0 1 0.614
Yes 14 3276 77.9 70.5 65.1 1.22 (0.57, 2.64)
Refill frequency (after 6 months)
Monthly 6 1223 74.0 65.8 61.6 1 0.856
Every 2 months 8 1978 80.9 74.0 67.6 0.83 (0.40, 1.72)
Every 3 months 4 946 80.0 72.5 68.6 0.82 (0.34, 1.96)
ARV drug dispensing in community
No 14 3190 74.4 66.0 60.2 1 0.004
Yes 4 957 92.6 88.2 84.1 0.32 (0.17, 0.61)
Sampling
Percentage of selected patient charts not found
<10% 6 1465 63.6 53.5 45.6 1 <0.001
≥10% to <20% 6 1475 88.1 82.4 78.6 0.31 (0.21, 0.54)
≥20% 6 1207 85.1 78.3 73.7 0.37 (0.22, 0.62)
ART, antiretroviral treatment; ARV, antiretroviral; CI, confidence interval.
*P-value from Cox regression models describing the effects of each programme level characteristic, correcting for imbalances in patient
characteristics between sites and adjusting for site using shared frailty methods.
Other models of community-based ART distribution, using a model with intensive community-based treatment
however, are emerging. In Rwanda, for example, Rich support that included ART distribution and directly
et al. reported a retention rate of 92.3% after 2 years observed ART by community health workers (Rich et al.
© 2014 John Wiley & Sons Ltd 1407
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10 Men no CBD Women no CBD
Men CBD Women CBD
0.00
Figure 2 Kaplan-Meier estimates by
0 .5 1 2 3 4 5
Community-Based Distribution (CBD) of
Years on ART ARVs in Tanzania, Uganda and Zambia.
2012). Other programmes use trained HIV-infected peers cific stakeholders and aimed to have a good balance of
(Community Care Coordinators) to visit ART patients site characteristics that might influence retention and
monthly and perform a systematic symptom review and adherence. The intrinsic differences among countries (for
dispense ARV drugs (Wools-Kaloustian et al. 2009). In example, in this study the majority of programmes with
Mozambique, M
edecins Sans Frontieres uses a Commu- community-based distribution of ARV drugs and pro-
nity ART model with stable ART patients who dispense grammes supported by faith-based or non-governmental
monthly ART and provide adherence and social support organisations were found in Uganda) could result in fur-
to other ART clients in the community (Decroo et al. ther confounding.
2011). They reported retention rates of 97.5% after a The strength of the current study is the use of consis-
median follow-up time of 13 months. The effectiveness tent data collecting tools across diverse sites and the pos-
of community pharmacies where ART patients are sibility of controlling for individual patient
trained to distribute ART at community distribution characteristics. By design, our research allowed studying
points needs to be confirmed by further research (MSF & interactions between programme level and individual
UNAIDS 2012). Although these models are showing characteristics, as illustrated by the differential effects of
promising impact on retention, their feasibility and scala- community-based distribution of ARV drugs between
bility still need to be evaluated. men and women. The design is less suited to study inter-
The implementation of the current WHO guidelines actions between programme level characteristics or differ-
(WHO 2013) aims to increase ART coverage and reten- ences between countries.
tion to save lives and to decrease HIV transmission. To Other limitations relate mainly to the constraints of
achieve the ambitious goal of universal coverage in rural retrospective chart review and the challenges of incom-
Africa, treatment will need to expand to serve communi- plete data (for example, WHO clinical stage and CD4
ties beyond the reach of current clinics. The potential of cell count) and the absence of certain variables at some
decentralisation of ART delivery (through mobile clinics of the sites (for example, distance to the clinic). The
and community pharmacies) and community participa- numbers of missing values for these variables are similar
tion (through community health workers and the patients to numbers reported elsewhere (May et al. 2010). They
and their families) need to be explored further. highlight the importance of strengthening data collection
Besides misclassification of transfer to care to LTFU, systems to better respond and assess retention to care
there are other limitations to this study. Study sites were and treatment.
not randomly selected, and this could have introduced Because this was a retrospective chart review, other
some selection bias. However, the selection (performed in important structural predictors of retention, such as mode
2006) was conducted in consultation with country-spe- of transport, educational level and income, could not be
1408 © 2014 John Wiley & Sons Ltd
Overall retention
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
Table 6 Final multivariate model for predictors of attrition in community programmes for ART drug dispensing could
multicountry retention study* be considered for broader implementation.
Adjusted hazard
ratio (95% CI)
Acknowledgements
Individual characteristics We wish to acknowledge the study participants and par-
Age at start ART: <30 years (vs. ≥30 years) 1.30 (1.14, 1.47)
ticipating clinics for their critical role in this study. This
WHO stage at start ART
III vs. I & II 1.12 (0.95, 1.31) research was been supported by PEPFAR through CDC
IV vs. I & II 1.56 (1.29, 1.88) and HRSA. The views, opinions and content of this pub-
Weight loss >10% 1.17 (1.00, 1.38) lication are those of the authors and do not necessarily
CD4 count (/log [tenfold] increase) 0.88 (0.78, 1.00) reflect the views, opinions or policies of the CDC, HRSA
Functional status or any other federal agency or office. This paper was pre-
Ambulatory vs. working/active 1.29 (1.09, 1.54)
sented in part at the XIX International AIDS Conference
Bedridden vs. working/active 1.54 (1.15, 2.07)
Sex (men vs. women) 2012, 22-27 July, Washington DC, USA.
In sites without community 1.33 (1.18, 1.50)
ARV drug dispensing
References
In sites with community 0.95 (0.67, 1.33)
ARV drug dispensing Altman DG & Bland JM (1998) Time to event (survival) data.
Years of clinic operation by Type of health facility (/year) BMJ 317, 468–469.
Government facility 1.17 (1.10, 1.23) Bangsberg DR, Perry S, Charlebois ED et al. (2001) Non-adher-
Mission facility/non-religious NGO 1.03 (0.95, 1.11) ence to highly active antiretroviral therapy predicts progression
Programme characteristics to AIDS. AIDS 15, 1181–1183.
ARV drug dispensing in community by Sex
Bedelu M, Ford N, Hilderbrand K & Reuter H (2007) Imple-
For women 0.55 (0.30, 1.01)
menting antiretroviral therapy in rural communities: the Lus-
For men 0.40 (0.21, 0.75)
ikisiki model of decentralized HIV/AIDS care. Journal of
Type of health facility: Faith-based 0.71 (0.42, 1.21)
Infectious Diseases 196 (Suppl. 3), S464–S468.
facility or NGO vs. Government
(at start of the programme) Braitstein P, Brinkhof MW, Dabis F et al. (2006) Mortality of
HIV-1-infected patients in the first year of antiretroviral ther-
ART, antiretroviral treatment; ARV, antiretroviral; CI, confi- apy: comparison between low-income and high-income coun-
dence interval; NGO: non-governmental organisation. tries. Lancet 367, 817–824.
Working/active: able to perform usual work in or out of the Calmy A, Pinoges L, Szumilin E et al. (2006) Generic fixed-dose
house; ambulatory: able to perform activities of daily living but combination antiretroviral treatment in resource-poor settings:
not able to work; bedridden: not able to perform activities of multicentric observational cohort. AIDS 20, 1163–1169.
daily living. Chan AK, Mateyu G, Jahn A et al. (2010) Outcome assessment
*The model was simplified using Akaike’s Information Criterion, of decentralization of antiretroviral therapy provision in a
retaining predictors and clinically plausible interaction terms that rural district of Malawi using an integrated primary care
increased model fit, with a penalisation for increasing model model. Tropical Medicine and International Health 15 (Suppl.
complexity. 1), 90–97.
Coetzee D, Hildebrand K, Boulle A et al. (2004) Outcomes after
assessed. The quality and completeness of the data varied two years of providing antiretroviral treatment in Khayelitsha,
among the study sites. Forster et al. (2008) found that South Africa. AIDS 18, 887–895.
poor quality data were associated with poor retention. In Collet D (2003).Modelling Surival Data in Medical Research, 2nd
this study, the sites with a large proportion of missing edn. Chapmann & Hall/CRC Press, Boca Raton, FL, USA.
records reported better retention proportions. While the Cornell M, Myer L, Kaplan R, Bekker LG & Wood R (2009)
significant association between the number of missing The impact of gender and income on survival and retention in
records and retention may indicate the presence of bias in a South African antiretroviral therapy programme. Tropical
this study, correcting for missing data did not change the Medicine and International Health 14, 722–731.
conclusions. This indicates the robustness of the findings. Cornell M, Grimsrud A, Fairall L et al. (2010) Temporal
changes in programme outcomes among adult patients initiat-
ing antiretroviral therapy across South Africa, 2002-2007.
Conclusion AIDS 24, 2263–2270.
Decroo T, Telfer B, Biot M et al. (2011) Distribution of antiret-
Patient retention to an individual programme worsened roviral treatment through self-forming groups of patients in
over time especially among males, younger persons and tete province, mozambique. Journal of Acquired Immune Defi-
those with poor clinical indicators. Increased use of ciency Syndromes 56, e39–e44.
© 2014 John Wiley & Sons Ltd 1409
Tropical Medicine and International Health volume 19 no 12 pp 1397–1410 december 2014
O. Koole et al. Retention in ART programs in sub-Saharan Africa
Ferradini L, Jeannin A, Pinoges L et al. (2006) Scaling up of apy and virologic outcomes. Annals of Internal Medicine 146,
highly active antiretroviral therapy in a rural district of 564–573.
Malawi: an effectiveness assessment. Lancet 367, 1335–1342. Paterson DL, Swindells S, Mohr J et al. (2000) Adherence to
Forster M, Bailey C, Brinkhof MW et al. (2008) Electronic protease inhibitor therapy and outcomes in patients with HIV
medical record systems, data quality and loss to follow-up: infection. Annals of Internal Medicine 133, 21–30.
survey of antiretroviral therapy programmes in resource-lim- Pocock SJ, Clayton TC & Altman DG (2002) Survival plots of
ited settings. Bulletin of the World Health Organization 86, time-to-event outcomes in clinical trials: good practice and pit-
939–947. falls. Lancet 359, 1686–1689.
Fox MP & Rosen S (2010) Patient retention in antiretroviral Rich ML, Miller AC, Niyigena P et al. (2012) Excellent clinical
therapy programs up to three years on treatment in sub-Saha- outcomes and high retention in care among adults in a com-
ran Africa, 2007-2009: systematic review. Tropical Medicine munity-based HIV treatment program in rural Rwanda. Jour-
and International Health 15 (Suppl. 1), 1–15. nal of Acquired Immune Deficiency Syndromes 59, e35–e42.
Geng EH, Nash D, Kambugu A et al. (2010) Retention in care Rosen S, Fox MP & Gill CJ (2007) Patient retention in antiret-
among HIV-infected patients in resource-limited settings: roviral therapy programs in sub-Saharan Africa: a systematic
emerging insights and new directions. Current HIV/AIDS review. PLoS Medicine 4, e298.
Reports 7, 234–244. Royston P (2004) Multiple imputation of missing values.
Giordano TP, Gifford AL, White AC Jr et al. (2007) Retention Stata Journal 4, 227–241.
in care: a challenge to survival with HIV infection. Clinical Shepherd BE, Blevins M, Vaz LM et al. (2013) Impact of
Infectious Diseases 44, 1493–1499. definitions of loss to follow-up on estimates of retention,
Hogg RS, Heath K, Bangsberg D et al. (2002) Intermittent use disease progression, and mortality: application to an HIV
of triple-combination therapy is predictive of mortality at program in Mozambique. American Journal of Epidemiology
baseline and after 1 year of follow-up. AIDS 16, 1051–1058. 178, 819–828.
Hougaard P (1995) Frailty models for survival data. Lifetime UNAIDS (2014). The Gap Report. UNAIDS, Geneva,
Data Analysis 1, 255–273. Switzerland.
Little RJA (1992) Regression with missing X’s: a review. Journal Weigel R, Estill J, Egger M et al. (2012) Mortality and loss to
of the American Statistical Association 87, 1227–1237. follow-up in the first year of ART: Malawi national ART
May M, Boulle A, Phiri S et al. (2010) Prognosis of patients programme. AIDS 26, 365–373.
with HIV-1 infection starting antiretroviral therapy in sub-Sah- WHO (2006). Patient Monitoring Guidelines for HIV Care and
aran Africa: a collaborative analysis of scale-up programmes. ART. WHO, Geneva, Switzerland.
Lancet 376, 449–457. WHO (2013). Consolidated Guidelines on the Use Of Antiretro-
MSF and UNAIDS (2012). Closer to Home: Delivering Antiret- viral Drugs For Treating and Preventing HIV Infection. Rec-
roviral Therapy in the Community: Experience From Four ommendations for a Public Health Approach 2013. WHO,
Countries in Southern Africa. MSF and UNAIDS, Geneva, Geneva, Switzerland.
Switzerland. Wools-Kaloustian KK, Sidle JE, Selke HM et al. (2009) A model
Nachega JB, Hislop M, Dowdy DW et al. (2007) Adherence to for extending antiretroviral care beyond the rural health cen-
nonnucleoside reverse transcriptase inhibitor-based HIV ther- tre. Journal of the International AIDS Society 12, 12–22.
Corresponding Author Olivier Koole, Department of Infectious Disease Epidemiology, London School of Hygiene and Tropical
Medicine, Keppel Street, London WC1E 7HT, UK. Tel.: +265 997 680 108; E-mail: olivier.koole@lshtm.ac.uk
1410 © 2014 John Wiley & Sons Ltd