770 • CID 2006:43 (15 September) • HIV/AIDS
H I V / A I D S M A J O R A R T I C L E
Determinants of Mortality and Nondeath Losses
from an Antiretroviral Treatment Service in South
Africa: Implications for Program Evaluation
Stephen D. Lawn,1,3Landon Myer,2,4Guy Harling,1Catherine Orrell,1Linda-Gail Bekker,1and Robin Wood1
1The Desmond Tutu HIV Centre, Institute for Infectious Disease and Molecular Medicine, and
of Public Health and Family Medicine, Faculty of Health Sciences, University of Cape Town, Cape Town, South Africa;
Department of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, London, United Kingdom; and
of Epidemiology, Mailman School of Public Health, Columbia University, New York, New York
2Infectious Diseases Epidemiology Unit, School
3Clinical Research Unit,
(See the editorial commentary by Weller on pages 777–8)
The scale-up of antiretroviral treatment (ART) services in resource-limited settings requires a
programmatic model to deliver care to large numbers of people. Understanding the determinants of key outcome
measures—including death and nondeath losses—would assist in program evaluation and development.
Between September 2002 and August 2005, all in-program (pretreatment and on-treatment) deaths
and nondeath losses were prospectively ascertained among treatment-naive adults (
in a community-based ART program in South Africa.
At study censorship, 927 patients had initiated ART after a median of 34 days after enrollment in the
program. One hundred twenty-one (9.8%) patients died. Mortality rates were 33.3 (95% CI, 25.5–43.0), 19.1 (95%
CI, 14.4–25.2), and 2.9 (95% CI, 1.8–4.8) deaths/100 person-years in the pretreatment interval, during the first 4
months of ART (early deaths), and after 4 months of ART (late deaths), respectively. Pretreatment and early
treatment deaths together accounted for 87% of deaths, and were independently associated with advanced im-
munodeficiency at enrollment. Late deaths were comparatively few and were only associated with the response to
ART at 4 months. Nondeath program losses (loss to follow-up, 2.3%; transfer-out, 1.9%; relocation, 0.7%) were
not associated with immune status and were evenly distributed during the study period.
Loss to follow-up and late mortality rates werelow, reflectinggoodcohortretentionandtreatment
response. However, the extremely high pretreatment and early mortality rates indicate that patients are enrolling
in ART programs with far too advanced immunodeficiency. Causes of late access to the ART program, such as
delays in health care access, health system delays, or inappropriate treatment criteria, need to be addressed.
) who were enrolledn p 1235
Although sub-Saharan Africa is home to just 10% of
the world’s population, more than 60% of the world’s
HIV-infected people live there; in 2005 alone, an esti-
mated 2.4 million people in the region died of HIV/
AIDS . As one component of a strategy to address
this devastating epidemic, access to antiretroviral treat-
ment (ART) is now being rapidly expanded within the
Received 20 February 2006; accepted 17 April 2006; electronically published 8
Reprints or correspondence: Dr. Stephen D. Lawn, Desmond Tutu HIV Centre,
Institute of Infectious Disease and Molecular Medicine, Facultyof HealthSciences,
University of Cape Town, Anzio Rd., Observatory 7925, Cape Town, South Africa
Clinical Infectious Diseases2006;43:770–6
? 2006 by the Infectious Diseases Society of America. All rights reserved.
region. In June 2005, it was estimated that 6.5 million
people urgently required treatment in resource-limited
settings. In view of the enormous scale of this inter-
vention, a simplified programmatic approach has been
adopted to facilitate delivery of treatment [2–4]. The
efficacy of ART, as reflected by virological and im-
munological responses, is similar among patients
treated in high-income countries and patients treated
in resource-limited countries [5, 6]. The impact of ART
programs in low-income countries is, therefore, un-
likely to be related to questions of drug efficacy, but
rather to health system issues andprogrameffectiveness
. Parameters with which to evaluate the effectiveness
of programs need to be identified. Tuberculosis treat-
ment programs provide a useful model of how evalu-
ation of carefully defined outcome measures permits
HIV/AIDS • CID 2006:43 (15 September) • 771
longitudinal program assessmentsandcomparisons.Similar
principals might usefully be applied to ART programs.
Two key goals of ART programs are to prevent mortalityand
to retain people within the program, receiving treatment in the
long term. In this study, we describe mortality occurring (1)
in the short interval betweenprogramenrollmentandinitiation
of ART (the pretreatment interval), (2) during early ART (0–
4 months), and (3) during late ART (4 months to 3 years). We
also characterized nondeath losses. We have accurately quan-
tified these different losses during the first 3 years of a com-
munity-based ART program in South Africa. We identified the
temporal distribution and determinants of these different out-
comes, which have thereby provided important insights into
SUBJECTS AND METHODS
lethu Community Health Centre in Cape Town and has pre-
viously been described in detail [9–11]. This district has a pre-
dominantly African population of 1300,000, the vast majority
of whom live in conditions of low socioeconomic status. In
2003, the antenatal HIV seroprevalence was 28%. Patients are
referred to the ART program from primary care HIV clinics.
Treatment criteria are based on the World Health Organiza-
tion’s (WHO’s) 2002 recommendations , which include a
prior AIDS diagnosis (WHO stage 4 disease) or a blood CD4
cell count !200 cells/mL.
The time between enrollment of a patient in the service and
initiation of ART is ∼1 month, to permit thorough evaluation
of patients and preparation for treatment as described previ-
ously [9–11]. Assignmentof eachpatienttoacommunity-based
therapeutic counselor facilitates this preparation and provides
an efficient system for determining outcomes for all patients,
including tracing those who do not attend follow-up appoint-
ments. A proportion of patients do not initiate treatment in
this service for a variety of reasons other than death; such
individuals are deferred from the program, and follow-up is
censored at this time-point. However, all individuals contrib-
uted to the pretreatment person-time of observation,regardless
of whether they subsequently received ART.
First-line ART consisted of stavudine and lamivudine plus a
nonnucleoside reverse transcriptase inhibitor (efavirenz or ne-
virapine). The second-line regimen for those for whom first-
line treatment failed was composed of lopinavir/ritonavir, zi-
dovudine, and didanosine. All treatment was provided free of
charge. Treatment adherence and viral load suppression at a
level of !400 copies/mL in this cohort both exceeded 90% at
1 year [10–12]. All patients with CD4 counts !200 cells/mL
received prophylaxis withdailytrimethoprim-sulfamethoxazole
or dapsone as an alternative. In addition to scheduled clinic
The ART service described here is based at the Gugu-
appointments at 4, 8, and 16 weeks and once every 16 weeks
thereafter, patients had open access to the clinic for medical
”Permanent deferrals” were patients who did
not receive ART for a reason other than pretreatment death
and who were subsequently excluded from the program. “Pre-
had enrolled into the program but who had not yet initiated
ART. “Early on-treatment deaths” were those that occurred
during the first 4 months of ART, and “late on-treatment
deaths” werethosethat occurredafter4monthsofART.“Trans-
fers-out” were patients whose care was transferred to another
ART program. “Relocations” were patients who moved to an-
other location but who were not referred to an ART service in
the new area. “Losses to follow-up”werepatientsreceivingART
who were 14 weeks late for a scheduled clinic or pharmacy
visit and who were neither transfers-out nor relocations.“Non-
death losses” were the sum total of transfers-out, relocations,
and losses to follow-up.
Structured clinical records were maintained
for all patients screened on entry to the ART program. This
basis. Data were analyzed from the start of the program in
September 2002 until data censorship in August 2005. This
study was approved by the Research Ethics Committee of the
University of Cape Town, and all patients who were enrolled
gave written, informed consent.
Data were analysed using Stata, version 9.0
(StataCorp). Wilcoxon rank-sum and Fisher’s exact tests were
used to compare medians and proportions, respectively.Insep-
arate analyses, we calculated rates of mortality and other out-
comes from either program enrollment (the date of initial
screening by the service) or from ART initiation. Person-time
was censored at the end of August 2005 for individuals who
were alive and who had been retained by the service. Product-
limit analyses were used to calculate the instantaneous hazard
of death or other losses through time among individuals re-
ceiving ART; we plotted smoothed hazard-function estimators
using weighted kernel-density estimates based on an Epane-
chnikov function . In other product-limit analyses, log-
rank tests were used to examine the effect of baseline WHO
clinical stage and category of CD4 cell count on survival prob-
abilities. All statistical tests are 2-sided at a p .05.
Multivariate analysis employed proportional hazard models
to examine determinants of mortality among individuals re-
ceiving ART. Separate models were developed to examine fac-
tors associated with early mortality, late mortality, and all
deaths. In separate models, baseline CD4 cell count was mod-
eled as both a continuous variable (per 50-cell/mL change in
the CD4 cell count) and a categorical variable, to demonstrate
772 • CID 2006:43 (15 September) • HIV/AIDS
the analysis (
Flow diagram showing the outcomes of patients included in
) at the time that data were censored. ART, an-n p 1235
threshold effects. Covariates were included in the model if they
demonstrated an appreciable association with the relative haz-
ard of mortality, or if their removal affected associations in-
volving other covariates. Model diagnostics and the propor-
tional hazards assumption were examined using Schoenfield
and scaled Schoenfield residuals .
Cohort and follow-up.
through August 2005, 1340 patients enrolled in the program.
Those who were not naive to ART (
years of age at enrollment (
n p 52
1235 patients who remained in the analysis, the median age
was 33 years (interquartile range [IQR], 28–38 years), and 882
patients (71%) were female. Baseline plasma viral load and
blood CD4 cell counts were available for 1086 and 1120 pa-
tients, respectively. The median plasma viral load was 4.81 log10
copies/mL (IQR, 4.42–5.23 log10copies/mL), and the median
blood CD4 cell count was 100 cells/mL (IQR, 47–160 cells/mL).
Most patients (79%) had symptomatic disease, and 644 (52%)
and 332 (27%) had WHO stage 3 and 4 disease, respectively.
Nine hundred twenty-seven patients (75%) received ART
during the study period, and 117 (9.5%) were preparing for
treatment at the time the study was censored (figure 1). The
median time between enrollment in the program and initiation
of treatment was 34 days (IQR, 28–50 days). Over the course
of the study, 170 person-years of observation were accrued in
the pretreatment interval, and 808 person-years were accrued
Numbers and characteristics of deaths and nondeath losses.
One hundred thirty-five (9.5%) patients were deferred from
the service before initiation of ART (figure 1) for a variety of
reasons, including decision to access treatment elsewhere, fail-
ure to attend follow-up clinic appointments, movement out of
the area, or for psychosocial reasons. Among total deaths
( ), 56 (46%) occurred in the pretreatment interval, 49
n p 121
(40%) were early on-treatment deaths, and 16 (13%) were late
on-treatment deaths. The death rate was high in the pretreat-
ment interval (33.3 deaths/100 person-years; 95% CI, 25.5–
43.0 deaths/100 person-years) but decreased during the first 4
months of ART (19.1 deaths/100 person-years; 95% CI, 14.4–
25.2 deaths/100 person-years), and was lower still beyond 4
months ART (2.9 deaths/100 person-years; 95% CI, 1.8–4.8
deaths/100 person-years). After 1 year of ART, the mortality
rate was just 1.3 deaths/100 person-years (95% CI 0.4–3.9
Among those who initiated ART (
(11.9%) were lost from the program (figure 1). Among these,
death was the most common reason, accounting for 65 (7.0%)
of losses, whereas 45 (4.9%) were due to other causes. These
During the period of September 2002
) and those !15
n p 53
) were excluded. Among
), 110 patientsn p 927
nondeath losses were due to transfer-out (18 [1.9%]), reloca-
tion (6 [0.7%]), and loss to follow-up (21 [2.3%]).
We compared the characteristics of patients who were lost
from the program due to death with those who were lost due
to other reasons or who continued to receive treatment when
the data was censored (table 1). In univariate analyses, those
who died in the pretreatment interval or during earlyARTwere
more likely to have a prior AIDS diagnosis, a baseline CD4 cell
count !100 cells/mL, and a baseline plasma viral load 1105
copies/mL, compared with thosewho remainedintheprogram.
Those who died after 4 months of ART (late deaths) were also
more likely to have a baseline CD4 cell count !100 cells/mL,
although this association did not persist in the multivariate
analysis (see below). In contrast, nondeath losses were not as-
sociated with baseline immunodeficiency. Kaplan-Meier anal-
yses confirmed that the probability of in-program death was
strongly associated with WHO stage of disease and baseline
CD4 cell count (figure 2).
Temporal distribution of program losses.
from the program changed markedly during follow-up of pa-
tients receiving ART (figure 3A). Risk of death had 3 distinct
phases: an initially high—but steeply decreasing—risk during
the initial months of ART, followed by a moderate risk of death
up to ∼1 year, and a very low risk of death after 1 year (figure
3B). In contrast, the risk of program loss due to other causes
was relatively constant (figure 3C). After ∼1 year of ART, risk
of nondeath losses to the program exceeded losses duetodeath.
Multivariate analysis for risk of death during ART.
In multivariate analysis to predict the relative hazards of death
during ART, death was significantly associated with baseline
CD4 cell count and WHO clinical stage, but not with age, sex,
or baseline viral load. However, risk factors for early deaths
versus late deaths differed markedly (table 2). Early on-treat-
ment deaths were associated withadvancedWHOclinicalstage,
lower baseline blood CD4 cell counts, and male sex (table 2).
The risk of loss
HIV/AIDS • CID 2006:43 (15 September) • 773
into the program stratified by World Health Organization (WHO) clinical
stage (A) and baseline blood CD4 cell count (B). These reveal that risk
of death early in the program was strongly associated with the baseline
degree of immunodeficiency at enrollment.
Kaplan-Meier plots showing survival of patients from entry
lost from the program for other reasons (transfer-out, relocation, or loss to follow-up).
Baseline characteristics of patients who either remained in the program, died, or were
No. of patients
Age, median years (IQR)
Prior AIDS diagnosis
CD4 count !100 cells/mL
Viral load 1105copies/mL
817 5649 16 45
acteristics of those who were retained in the program, compared with the characteristics of those who were lost to
the program. IQR, interquartile range.
Data are no. (%) of patients, unless otherwise indicated. Statistical comparisons were made between char-
In contrast, late on-treatment deaths were only independently
associated with the response to ART at 4 months, as reflected
by blood CD4 cell count and viral load. Although the number
of late deaths wassmall (datawereavailablefor12of16deaths),
this association with CD4 cell count was statistically highly
significant, and the trend towards an association with viral load
approached statistical significance.
CD4 cell count increases and risk of loss to program.
We examined how on-treatment program losses (death and
nondeath) were associated with CD4 cell counts at baselineand
after 4 months of ART. Although the median CD4 cell increases
among patients retained in the cohort (99 cells/mL; IQR, 49–
162 cells/mL) were similar to those among nondeath losses (83
cells/mL; IQR, 49–133 cells/mL), those who subsequently died
had much smaller CD4 cell count increases (44 cells/mL; IQR,
5–83 cells/mL). The vast majority of deaths occurred among
individuals who had a baseline CD4 cell count !100 cells/mL
and a CD4 cell count !200 cells/mL at 4 months (figure 4A).
In contrast, nondeath losses were not associated with the CD4
cell count distribution (figure 4B).
In this study we carefully quantified mortality and nondeath
losses in a community-based ART program in South Africaand
identified the temporal distribution and risk factors associated
with these losses. We defined pretreatment, early and late ART
mortality, and nondeath losses as useful outcome measures of
ART programs. Loss to follow-up and late mortality rates were
low, reflecting excellent cohort retention and treatment re-
sponse in this program. In contrast, however, pretreatmentand
early mortality rates were very high: this finding very strongly
suggests that patients were enrolling with far too advanced
immunodeficiency. To reduce in-program mortality, the causes
of late program entry need to be addressed.
On-treatment mortality rates were similar to or better than
those previously reported from resource-limited settings [15–
19]. Although previous studies have not reported mortality
occurring within the program prior to actual initiation of ART,
we demonstrated that a large proportion of early program
774 • CID 2006:43 (15 September) • HIV/AIDS
death (B), and nondeath losses to program (C; i.e., transfers-out, relo-
cations, and losses to follow-up). Risk of mortality was initially high but
decreased steeply, and there were very few deaths after 400 days of
ART. Risk of nondeath losses was relatively consistent during follow-up.
ART, antiretroviral therapy
Smoothed hazard estimates for total losses to program (A),
hazards of early deaths (
initiation of antiretroviral therapy (ART).
Results of separate Cox’s models predicting relative
) and late deaths (n p 49 ) aftern p 12
1, 2, or 3
Baseline CD4 countc
CD4 count at 4 months
Baseline viral load
Viral load at 4 months
Early deathsLate deathsa
1.01 (0.97–1.05) 1.07 (0.99–1.14)
aComplete data available for 12 of 16 patients.
bAnalyzed as a continuous variable.
cAnalyzed in 50-cell/mL increments.
Data are hazard ratio (95% CI). WHO, World Health Organization.
deaths occur during this pretreatment interval . The pre-
treatment interval in the present analysis (median, 34 days) is
shorter than reported elsewhere butpermittedcarefuleval-
uation, investigation, andtreatmentofopportunisticinfections,
as well as thorough preparation of patients for ART. We believe
this preparation is key to the very high adherence rates and
excellent virological and immunological outcomes observed in
this program [10, 11]. Within this service, clinicians were able
to “fast-track” patients who had the most advanced immu-
nodeficiency; however, shortening the pretreatment intervalfor
all patients may not necessarily reduce early mortality and may
actually compromise long-term outcomes.
The present analysis shows that, after 3 years of this program,
deaths in the pretreatment interval contributed 146% of total
program mortality and therefore represent a very importantout-
come measure. This mortality is likely to reflect a far greater
burden of mortality that is actually occurring prior to program
entry. Those who died in the first 4 months of treatment (early
deaths) shared the same baseline characteristics and risk factors
as those who died in the pretreatment interval (tables 1 and 2);
together, deaths in these 2 intervals constituted 87% of total
program mortality. The strong association of these deaths of
patients who had advanced immunodeficiency at baselineclearly
indicates that many of these patients had advanced disease that
could not be salvaged by ART, despite the active management
in many patients of concurrent infections. One-half of the pa-
tients enrolling into this program had a baseline CD4 cell count
!100 cells/mL. Kaplan-Meier survival analyses revealed that early
mortality was high among patients with WHO stage 3 and stage
4 disease (i.e., symptomatic disease) and patients with baseline
CD4 cell counts !100 cells/mL (figure 2).
The reasons why patients typically enter this and other pro-
grams in resource-limited settings with such advanced disease
HIV/AIDS • CID 2006:43 (15 September) • 775
(dark squares) who either died (A; data available for 12 of 16 patients) or were lost to the program for other reasons (B;
antiretroviral therapy (ART) are indicated. Deaths—but not nondeath losses—occurred among individuals with advanced baseline immunodeficiency
and persisting immunodeficiency after 16 weeks of ART.
Scatter plots of baseline CD4 cell count against CD4 cell count at the 16-week follow-up time-point. On these plots, individual patients
) after 4 months ofn p 22
need to be identified. Possible reasons include (1) barriers to
voluntary counseling and testing and to access to health care,
(2) lack of routine blood CD4 cell count testing for patients
with new HIV diagnoses, (3) health system delays in referral
patients to ART clinics, (4) waiting times to enter programs,
and (5) criteria for initiation of ART that only include patients
with advanced disease. How to best promote early access of
ART-eligible patients to treatment programs is a central chal-
lenge facing the scale-up of these programs in resource-limited
settings . Moreover, the South African national ART pro-
gram uses the WHO 2002 ART guidelines, which restrictstreat-
ment to patients with stage 4 disease or CD4 cell counts !200
cells/mL . The revised WHO 2003 guidelines for resource-
limited settings recommend earlier treatment , and the use
of these may help to lower mortality. Moves to earlierinitiation
of treatment are supported by a collaborative analysis of da-
tasets that reveal that mortality rates in ART programs in re-
source-limited settings are higher than those in high-income
In contrast to early deaths, mortality after 4 months of ART
(late death) was independent of baseline immune status but
was strongly associated with the response to ART, as reflected
As such, the late mortality rate reflected therapeutic success,
including drug regimen efficacy and tolerability as well as pa-
tient adherence to treatment. After the first year of ART, mor-
tality rate was very low—approaching 1% per year—and the
risk of loss to program due to nondeath causes exceeded those
due to death (figure 3). Thus, the effectiveness of programs
beyond 1 year is likely to relate to issues of long-term patient
retention ratherthandeath. Aspatientsremainhealthyonlong-
term medication, their motivation to continue treatment in the
longer term may diminish . It remains unknown whether
an increasing rate of treatment failure and a secondary increase
in the death rate may occur with longer follow-up.
Among ART programs in sub-Saharan Africa, rates of non-
death program losses range from !5% [15, 17] to 150% .
When patients are termed “lost to follow-up” simply on the
basis of persistent failure to attend clinic appointments, it is
possible that some may in fact have died without detection by
the ART program. The true mortality rate among patients re-
ceiving ART may, therefore, often be underestimated.However,
in this analysis, the use of community-based therapeutic coun-
selors that were allocated to each patient enhanced the data
completeness and assessment of outcomes for patients who
failed to attend follow-up appointments.
Nondeath program losses in this setting were heterogeneous
in nature. Patients who moved out of the area accounted for
22% of total on-treatment program losses. As opposed to those
transferred to another ART program, the care of some patients
(here termed “relocations”) was not transferred, often because
of lack of provision of ART services in other areas. Because the
number of patients who are physically well and who are re-
ceiving long-term medication is increasing, the number of pa-
tients moving out of an area for social or economic reasons
may continue to increase. This finding emphasizes the impor-
tance of systems within national ART programs that can ensure
continuity of care for highly mobile populations of young
adults. Losses to follow-up in this treatment service were low,
likely as a result of dedicated community-based counsellors
allocated to each patient and thorough preparation of patients
776 • CID 2006:43 (15 September) • HIV/AIDS Download full-text
In summary, high pretreatment and early on-treatmentmor-
tality rates in this program reflected very advanced immuno-
deficiency. The reasons for patients’ late access to the program
urgently need to be identified. It is likely that these early in-
program deaths reflect a far greater burden of mortality within
the health system and in the community that is occurring “up
stream” of the ART program. However, this study found that,
once patients have initiated ART and survived the initial few
months of treatment, the risk of death or loss to the program
thereafter was very low. Evaluation of these various outcome
measures provides important means of assessment for ART
programs, which thereby facilitates development of optimum
models of care in resource-limited settings.
We are grateful to the staff at the Hanan Crusaid antiretroviral clinic in
Gugulethu and to the staff of the Desmond Tutu HIV Centre (Cape Town,
Wellcome Trust (grant 074641/Z/04/Z; to S.D.L.)
and a National Institutes of Health through a CIPRA grant (1U19AI53217-
01; to L.M., C.O., L.G.B., and R.W.). Provision of ART at the program was
initially by Crusaid, London, UK, and latterly by the Global Fund for
Malaria, Tuberculosis and HIV/AIDS (administered through a provincial
Potential conflicts of interest.
All authors: no conflicts.
1. UNAIDS/WHO. AIDS epidemic update: December 2004 Available at:
http://www.unaids.org/epi/2005/doc/report_pdf.asp. Accessed 5 April
2. WHO/UNAIDS. Progress on global access to HIV antiretroviral ther-
apy: an update on “3 by 5,” June 2005. Available at: http://www.who
Accessed 19 October 2005.
3. World Health Organization. Scaling up antiretroviral therapy in re-
source-limited settings: guidelines for a public health approach; ex-
ecutive summary. Geneva: World Health Organization, 2002.
4. World Health Organization. Scaling up antiretroviral therapy in re-
source-limited settings: guidelines for a public health approach; 2003
revision. Geneva: World Health Organization, 2002. Available at: http:
.html. Accessed 15 February 2006.
5. Braitstein P, Brinkhof MW, Dabis F, et al. Mortality of HIV-1-infected
patients in the first year of antiretroviral therapy: comparisonbetween
low-income and high-income countries. Lancet 2006;367:817–24.
6. Ivers LC, Kendrick D, Doucette K. Efficacy of antiretroviral therapy
programs in resource-poor settings: a meta-analysis of the published
literature. Clin Infect Dis 2005;41:217–24.
7. Lawn SD, Myer L, Wood R. Efficacy of antiretroviral therapy in re-
source-poor settings: are outcomes comparable to those in the devel-
oped world? Clin Infect Dis 2005;41:1683–4.
8. Enarson DA, Rieder HL, Arnadottir T, Tre ´bucq A. A guide for low
income countries. 5th ed. Paris: International Union Against Tuber-
culosis and Lung Disease, 2000.
9. Lawn SD, Myer L, Orrell C, BekkerLG, WoodR.Earlymortalityamong
adults accessing a community-based antiretroviral service in South
Africa: implications for programme design. AIDS 2005;19:2141–8.
10. Lawn SD, Myer L, Bekker LG, et al. CD4 cell count recovery among
HIV-infected patients with very advanced immunodeficiency com-
mencing antiretroviral treatment in sub-Saharan Africa. BMC Infect
11. Bekker LG, Myer L, Orrell C, Lawn SD, Wood R. Rapid scale up of a
community-based HIV treatment service: programme performance
over three consecutive years in Guguletu, South Africa. S Afr Med J
12. Orrell C, Badri M, Wood R. Measuring adherence in a community
setting: which measure most valuable [abstract WePEB5787]? In: Pro-
gram and abstracts of the XVI International AIDS Conference (Bang-
13. Wand MP, Jones MC. Kernel smoothing. In: Monographs in statistics
and applied probability. New York: Chapman & Hall, 1995.
14. Ng’andu NH. An empirical comparison of statistical tests for assessing
the proportional hazards assumption of Cox’s model. Stat Med
15. Coetzee D, Hildebrand K, Boulle A, et al. Outcomes after two years
16. Laurent C, Ngom Gueye NF, Ndour CT, et al. Long-term benefits of
highly active antiretroviral therapy in SenegaleseHIV-1-infectedadults.
J Acquir Immune Defic Syndr 2005;38:14–7.
17. Tassie JM, Szumilin E, Calmy A, Goemaere E. Highly active antiret-
roviral therapy in resource-poor settings: the experience of Medecins
Sans Frontieres. AIDS 2003;17:1995–7.
18. Djomand G, Roels T, Ellerbrock T, et al. Virologic and immunologic
outcomes and programmatic challenges of an antiretroviral treatment
pilot project in Abidjan, Cote d’Ivoire. AIDS 2003;17(Suppl 3):S5–15.
19. Weidle PJ, Malamba S, Mwebaze R, et al. Assessment of a pilot an-
tiretroviral drug therapy programme in Uganda: patients’ response,
survival, and drug resistance. Lancet 2002;360:34–40.
20. Lawn SD, Wood R. How can earlier entry of patients intoantiretroviral
programs in low-income countries be promoted? Clin Infect Dis
21. Myer L, El Sadr W. Expanding access to antiretroviral therapy through
the public sector—the challenge of retaining patients in long-term
primary care. S Afr Med J 2004;94:273–4.
22. van Oosterhout JJ, Bodasing N, Kumwenda JJ, et al. Evaluation of
antiretroviral therapy results in a resource-poor setting in Blantyre,
Malawi. Trop Med Int Health 2005;10:464–70.