Implementation and Operational Research: Evaluation of Swaziland's Hub-and-Spoke Model for Decentralizing Access to Antiretroviral Therapy Services : JAIDS Journal of Acquired Immune Deficiency Syndromes

Secondary Logo

Journal Logo

Clinical Science

Implementation and Operational Research

Evaluation of Swaziland's Hub-and-Spoke Model for Decentralizing Access to Antiretroviral Therapy Services

Auld, Andrew F. MBChB, MSc*; Kamiru, Harrison DDS, DrPH; Azih, Charles MD; Baughman, Andrew L. PhD, MPH*; Nuwagaba-Biribonwoha, Harriet MBChB, PhD; Ehrenkranz, Peter MD, MPH§; Agolory, Simon MD*; Sahabo, Ruben MD; Ellerbrock, Tedd V. MD*; Okello, Velephi MD; Bicego, George PhD§

Author Information
JAIDS Journal of Acquired Immune Deficiency Syndromes 69(1):p e1-e12, May 1, 2015. | DOI: 10.1097/QAI.0000000000000547



In 2007, Swaziland initiated a hub-and-spoke model for decentralizing antiretroviral therapy (ART) access. Decentralization was facilitated through (1) down-referral of stable ART patients from overburdened central facilities (hubs) to primary health care clinics (spokes) and (2) ART initiation at spokes (spoke initiation).


We conducted a nationally representative retrospective cohort study among adult ART enrollees during 2004–2010 to assess the effect of down-referral and spoke-initiation on rates of loss to follow-up (LTFU), death, and attrition (death or LTFU). Sixteen of 31 hubs were randomly selected using probability-proportional-to-size sampling. Seven selected facilities had initiated the hub-and-spoke model by study start. At these facilities, 1149 of 24,782 hub-initiated and maintained and 878 of 7722 down-referred or spoke-initiated patient records were randomly selected and analyzed. At the 9 hub-only facilities, 483 of 6638 records were randomly selected and analyzed. Multivariable proportional hazards regression was used to assess effect of down-referral (a time-varying covariate) and spoke-initiation on outcomes.


At ART initiation, median age was 35, 65% were female, and median CD4 count was 147 cells per microliter. Controlling for known confounders, down-referral was strongly protective against LTFU [adjusted hazard ratio (AHR) 0.38; 95% confidence interval (CI): 0.29 to 0.50] and attrition (AHR = 0.50; 95% CI: 0.34 to 0.76) but not mortality. Compared with hub-initiated and maintained patients, spoke-initiated patients had lower LTFU (AHR 0.59; 95% CI: 0.45 to 0.77) and attrition rates (AHR 0.60; 95% CI: 0.47 to 0.77), but not mortality.


Down-referral and spoke-initiation within a hub-and-spoke ART decentralization model were protective against LTFU and overall attrition and could facilitate future ART program expansion.


Initiating and retaining HIV-infected adults on antiretroviral therapy (ART) reduces patient morbidity and mortality and risk of HIV transmission to seronegative partners.1 At the community level, ART coverage scale-up has potential to reduce population-level HIV incidence.2 To scale-up ART coverage, many programs in resource-limited countries have decentralized ART services by moving ART delivery from hospitals to peripheral health facilities, like primary health care clinics (PHCs), or even beyond health facilities.3 World Health Organization (WHO) guidelines categorize decentralization models as (1) partial decentralization, where patients start ART at central hospitals and can be down-referred to PHCs when stable; (2) full decentralization, where patients start and continue ART at PHCs; and (3) decentralization beyond health facilities, where trained volunteers deliver antiretrovirals to people in their homes.4

In Swaziland, the country with the highest adult HIV prevalence globally (31%),5 where 80,103 of 93,000 ART-eligible adults (87%) (CD4 ≤ 350 cells per microliter) were receiving ART by 2013, both partial and full decentralization models have been initiated since 2007 through rollout of a hub-and-spoke system. Swaziland's partial decentralization model allows for patients, who are initiated on ART by doctors at central ART initiation facilities (hubs), and are stable after 3 months, to be down-referred to PHCs (spokes) closer to their homes, where continued clinical monitoring and monthly medication refills are provided by nurses. The full decentralization model allows for doctors to initiate ART for patients at spokes during outreach visits from the hub.

Down-referral (partial decentralization) has been associated with reduced loss to follow-up (LTFU) and mortality among patients starting ART at a well-resourced hub in a single hub-spoke pair in South Africa 6,7 and within a single hub-spoke system within a Malawi district.8 However, these findings cannot be extrapolated to larger, nation-wide, down-referral programs involving multiple hubs and spokes.7 PHC-initiation (full decentralization) models have been associated with lower LTFU and attrition rates among adults3; however, some studies have reported increased mortality among PHC-initiated patients,9 and a recent meta-analysis encouraged further research.3

Therefore, to inform future ART scale-up in Swaziland, the Ministry of Health (MOH) and partners initiated a study to evaluate whether down-referral (partial decentralization) and PHC-initiation (full decentralization) were associated with higher or lower rates of LTFU and mortality.


Swaziland's National ART Cohort

During 2004–2010, Swaziland graduated certain central heath facilities into the role of ART initiation facilities or hubs based on certain site-specific criteria (see Table S1, Supplemental Digital Content,; during 2004–2010, the number of hubs increased from 1 to 31. Since 2007, Swaziland has initiated the hub-and-spoke model for some hubs that had a large patient load. Of the 31 hubs, 9 initiated the “hub-and-spoke” model during 2007–2010, linking with 59 PHCs that served as spokes. A PHC could only graduate into the role of a spoke if it met certain human resource, infrastructure, data capture, and operational criteria (see Table S1, Supplemental Digital Content, During 2004–2010, the number of adults ever enrolled on ART increased from <5000 to 78,919, with 60,757 currently on ART by December 31, 2010.

ART Eligibility and Monitoring

During 2004–2009, Swaziland implemented the 2006 WHO ART-eligibility guidelines,10 while in early 2010, Swaziland adopted the 2010 WHO guidelines.11 Prescription of cotrimoxazole was indicated for all ART enrollees. At ART initiation, monthly for the first quarter, and quarterly thereafter, patients attended clinic visits and standardized MOH-recommended records were completed to monitor disease progression or improvement. Patients collected medications monthly from either hubs or spokes. For all patients late for clinic appointments, telephonic tracing, and if necessary, home visits, were recommended.

Study Design and Population

This was a retrospective cohort study among a nationally representative sample of adult ART patients. Patient-level data were abstracted from standardized, MOH-recommended medical records onto study questionnaires by trained abstractors from November 2011 through March 2012. Only medical records of adults, ≥15 years at ART initiation, who started ART during 2004–2010, were eligible. For down-referred and spoke-initiated patients, paper records maintained at the spoke were abstracted.

Treatment Outcomes

The primary ART outcomes of interest were documented mortality and LTFU, whereas the composite outcome of attrition (documented death or LTFU) was a secondary outcome. A patient was considered LTFU if absent from the facility in the 90 days preceding data abstraction and if there was no documentation of death or transfer out. The date of LTFU was the most recent visit date. Mortality ascertainment occurred largely through passive reporting and to a lesser extent through MOH-recommended tracing activities.

Exposure Variables

The key exposure variables were (1) the binary time-varying covariate of whether an adult ART patient was down-referred from a hub to one of its spokes or was hub-maintained and (2) the type of facility that initiated ART, which was coded as a 3-level variable depending on whether the site was a hub only, a hub with spokes, or a spoke that could initiate ART during outreach visits.

In Swaziland, down-referral is different from transfer out. Whereas patients can be transferred out at any time, patients should only be down-referred when they (1) have completed 3 months of ART, (2) have demonstrated good adherence in the opinion of a clinician, (3) appear clinically stable and have increased their CD4+ T-cell (CD4) count, and (4) have agreed to down-referral. In this analysis, transfer outs are censored at the transfer date.

Other exposure variables of interest abstracted from charts included age, sex, marital status, WHO stage, functional status, weight, CD4 count, hemoglobin, and presence or absence of a treatment supporter.

Sample Size

Assuming a design effect of 2 and that 12-month attrition was 25% among down-referred or spoke-initiated patients and 35% among hub-initiated and maintained patients, a sample size of ≥2500 patients was sufficient to detect the estimated attrition difference with >80% power and alpha at 0.05.


Funding restricted the number of study hubs to 16. To sample 16 of 31 hubs, the hubs were stratified into hospitals (6), health centers (8), and clinics (17). Within strata, MOH monitoring data on ART enrollees defined site size. A probability–proportional-to-size sampling program randomly selected 4 from 6 hospitals, 4 from 8 health centers, and 8 from 17 clinics (Fig. 1).

Selection of study patients to evaluate Swaziland's hub-and-spoke model for ART decentralization.

To meet the 2500 sample size, simple random sampling was used to select 391 records from each hospital, 150 records from each health center, and 42 records from each clinic.

To maximize power to answer the primary study questions, at those facilities which had initiated the hub-and-spoke system by study start, we aimed to sample equal numbers of adults who were (1) hub-initiated and maintained as were (2) either down-referred or spoke-initiated (Fig. 1).

Analytic Methods

Data were analyzed using STATA 11 (Stata Statistical Software, Release 11, 2009; Stata Corp, College Station, TX). Survey procedures were used to account for the complex survey design. Health facility and facility type were specified as the primary sampling unit and stratification variables, respectively, and sampling weights were used to account for unequal selection probabilities.

Complete data were available for time-to-event analysis and for key exposure variables, including down-referral status, down-referral date, ART initiation facility type, age, and sex. For certain other variables (Table 1), some data were missing.

TABLE 1-a:
Clinical and Demographic Characteristics of Adults at Initiation of ART in Swaziland During 2004–2010
TABLE 1-b:
Clinical and Demographic Characteristics of Adults at Initiation of ART in Swaziland During 2004–2010

Similar to other cohort studies using routine data,12 to best manage missing covariate data,13,14 multiple imputation with chained equations was used to impute the missing data.15 The ice16–18 procedure in STATA was used to create 20 imputed datasets for each of the following outcomes: (1) death, (2) LTFU, and (3) overall attrition.19 The imputation model included the event indicator, all study variables, and the Nelson–Aalen estimate of cumulative hazard.20 For all analyses using imputed data, estimates were combined across the imputed datasets according to Rubin's rules,15 using the mim procedure in STATA.21

To describe the cohort, 4 patient categories were created: (1) patients who were started and maintained on ART at hubs that had not yet implemented the hub-and-spoke model (hub-only patients), (2) patients who were hub-initiated and maintained within facilities with hub-and-spoke systems (hub-maintained patients), (3) patients who were down-referred from hubs to spokes (down-referred patients), and (4) patients who started ART at spokes (spoke-initiated patients). To assess associations between baseline characteristics and patient category, linear, logistic, ordered, or multinomial logistic regression models were used for continuous, binary, ordinal, and nominal categorical baseline covariates, respectively.

A competing risks model was used to estimate 6-month, 1-, 2-, 3-, and 4-year mortality and LTFU for each patient category starting ART during 2004–2010.22

To assess the effect of down-referral and location of ART initiation on outcomes, proportional hazards regression models were used to estimate crude and adjusted hazard ratios (AHRs) and 95% confidence intervals (CIs).23 The multivariable model was adjusted for covariates considered a priori risk factors for death and LTFU. The proportional hazards assumption was assessed using visual methods and the Grambsch and Therneau test.24 Because only 7 of 16 selected hubs were capable of down-referral, a subpopulation analysis, including only adults starting ART at these 7 hubs, was conducted to assess robustness of AHRs estimated for the down-referral covariate.

Ethics Approval

This study was approved by the Swaziland Scientific and Ethics Committee, the Columbia University Medical Center Institutional Review Board (IRB), and the US Centers for Disease Control and Prevention (CDC) IRB.


Characteristics at ART Initiation

Medical records of 2510 study-eligible adults were abstracted and analyzed; 483 (17%) were hub-only patients, 1149 (40%) were hub-maintained patients, 367 (25%) were down-referred patients, and 511 (18%) were spoke-initiated.

Median age at ART initiation was 35 and was similar across patient groups (Table 1). Among hub-only, hub-maintained, and down-referred patients, 62%–63% of patients were female, whereas 74% of spoke-initiated patients were female (P = 0.007). The highest point estimate of pregnancy prevalence at ART initiation was among spoke-initiated patients (19%); however, prevalence differences between groups were not significant (P = 0.941). The proportion married at ART initiation was 47% overall but lowest among spoke-initiated patients (40%, P = 0.002). The proportion employed was 44% overall but lowest among down-referred (28%) and spoke-initiated (35%) patients (P = 0.016). Year of ART initiation varied between groups with 75%–78% of hub-only, hub-maintained, and down-referred patients starting ART during 2008–2010, compared with 96% of spoke-initiated patients (P < 0.001). Overall, 82% (95% CI: 76 to 87) provided a telephone number to allow tracing, and this did not vary significantly across groups.

Prevalence of WHO stage III/IV disease was 56% overall and similar across groups (P = 0.705). However, prevalence of moderately and severely impaired functional status was highest among hub-maintained patients (35%) compared with the other 3 groups (17%–22%, P = 0.015). Similarly, prevalence of active tuberculosis at ART initiation was highest among hub-maintained patients (13%) compared with the other 3 groups (6%–8%, P = 0.010), and median weight at ART initiation was lowest among hub-maintained patients (58.1 kg) compared with the other 3 groups (59.0–63.0 kg, P = 0.020). Although baseline CD4 count was lowest among hub-maintained patients (131 cells per microliter) compared with the other 3 groups (146–178 cells per microliter), this was not statistically significant (P = 0.684).

Overall, 69% were prescribed cotrimoxazole at ART initiation, and 98% had a documented treatment supporter, and these proportions were similar between groups. However, first-line regimen distribution varied between groups, with tenofovir, lamivudine, and efavirenz prescribed to 27% of spoke-initiated patients compared with 5%–13% of adults in other groups (P < 0.001).

Down-referral Timing

Most down-referred patients were down-referred after 6 months of ART (77%); however, 17% were down-referred during months 3–6 of ART, and 7% were down-referred before 3 months of completed ART.

Treatment Outcomes

Over 5198 person-years of follow-up, 712 adults were lost to attrition; 107 died and 605 were LTFU. Overall attrition was 20% by 1 year; 3% had died and 16% were LTFU.

Six-month, 1-, 2-, 3-, and 4-year LTFU proportions were lowest among down-referred patients compared with hub-only, hub-maintained, and spoke-initiated patients (Table 2). By 2 years of follow-up after ART initiation, LTFU was 3% among down-referred patients compared with 22% among patients in the other 3 groups (Table 2). Similarly, at 4 years, LTFU was 7% in the down-referred group compared with 31%–35% in the other 3 groups.

Treatment Outcomes of Adults Initiating ART in Swaziland's Hub-Only and Hub-and-Spoke Systems During 2004–2010

Effect of Down-referral

After controlling for known risk factors for death and LTFU, the time-varying down-referral covariate was strongly protective against LTFU (AHR = 0.38; 95% CI: 0.29 to 0.50), but not documented death (Table 3). In the subpopulation analysis, restricting the cohort to adults who started ART at the 7 hubs capable of down-referral, down-referral remained protective against LTFU (AHR 0.41; 95% CI: 0.30 to 0.57), but not death (see Table S2, Supplemental Digital Content, Similarly, down-referral was protective against overall attrition in the full model (AHR 0.50; 95% CI: 0.34 to 0.76) and in the subpopulation analysis (AHR 0.56; 95% CI: 0.33 to 0.93) (see Table S3, Supplemental Digital Content,

TABLE 3-a:
Predictors of LTFU and Death Among Adults Starting ART in Swaziland During 2004–2010 (N = 2510)
TABLE 3-b:
Predictors of LTFU and Death Among Adults Starting ART in Swaziland During 2004–2010 (N = 2510)

Effect of Spoke-Initiation

Compared with patients starting ART at the 9 hub-only facilities, patients starting ART at spokes had lower LTFU rates (AHR 0.54; 95% CI: 0.35 to 0.84) (Table 3; Fig. 2) and overall attrition (AHR 0.54; 95% CI: 0.40 to 0.73) (see Table S3, Supplemental Digital Content,, but not death (AHR 0.50; 95% CI: 0.17 to 1.45).

Cumulative incidence of death and LTFU among hub-only and hub-maintained ART enrollees vs. spoke-initiated and down-referred enrollees.

Compared with patients starting ART at the 7 hubs that incorporated the hub-and-spoke model, patients starting ART at spokes had lower LTFU rates (AHR 0.59; 95% CI: 0.45 to 0.77) and attrition (AHR 0.60; 95% CI: 0.47 to 0.77), but not mortality (AHR 0.69; 95% CI: 0.32 to 1.48).

Other Predictors

Being 10-years older at ART initiation was associated with 11% lower LTFU rates but 22% increased mortality (Table 3). Compared with females, males had increased LTFU and mortality rates. Compared with being married, being divorced, single, or widowed was associated with increased mortality, but not LTFU. ART enrollees in later calendar years had higher LTFU rates compared with adults starting ART in earlier calendar years; however, rates of documented mortality remained constant over time. Compared with patients with WHO stage I/II, patients with WHO stage IV had higher rates of both LTFU and death. Having a lower functional status at ART initiation (eg, being bedridden vs. ambulatory, or ambulatory vs. working) was associated with higher LTFU and mortality rates. Being 10 kg heavier at ART initiation was protective against both LTFU and death. Compared with patients with a CD4 ≥ 200 cells per microliter at ART initiation, patients with a CD4 < 50 had higher LTFU and mortality rates. Compared with patients with a hemoglobin ≥8 g/dL, patients with a hemoglobin <8 g/dL had higher LTFU and mortality rates. Patients not prescribed cotrimoxazole at ART initiation had higher LTFU rates, but not mortality. Patients who did not have a documented treatment supporter had higher LTFU rates, but not mortality.


This is the first evaluation of a nation-wide ART decentralization program in sub-Saharan Africa involving multiple hubs and spokes. The key findings are that down-referral and spoke-initiation were protective against both LTFU and overall attrition. These findings support continued scale-up of the hub-and-spoke decentralization model as Swaziland adopts new WHO guidelines recommending ART initiation at CD4 ≤ 500 cells per microliter, which would nearly double the number of ART-eligible adults from 93,000 to 170,000 in 2014.


At ART initiation, compared with hub-maintained patients, down-referred patients were more likely to be married, pregnant if female, unemployed, and have fewer markers of advanced HIV disease. Demographic characteristics of down-referred patients suggest patient motivation for down-referral might have included (1) lack of employment obligations near hubs, (2) family and child responsibilities at home, and (3) prohibitive transport costs to central hubs given unemployment prevalence. Being healthier at ART initiation is consistent with procedures for down-referralbecause only stable patients should have been down-referred.

Similar to studies in South Africa6,7 and Malawi,8 and findings from a recent meta-analysis,3 down-referral reduced risk of LTFU and overall attrition. Lower LTFU risk may be explained by reduced transport burden and increased convenience.25 In addition, down-referred patients may have been more satisfied with care at spokes6; in South Africa, a patient flow analysis reported >90% of down-referred patients completed clinical appointments, including waiting time, in <1 hour, whereas at central facilities, patients spent a median of 2.5 hours per medical visit and 1.9 hours at the pharmacy.6


Compared with hub-only and hub-maintained patients, at ART initiation, spoke-initiated patients were more likely to be female, unemployed, enrolled in later calendar years, and have less advanced HIV disease. Pregnancy was also more prevalent among spoke-initiated patients, though this was not statistically significant. These features of spoke-initiated patients suggest that HIV testing during antenatal care may have been an important avenue through which spoke-initiated patients were identified as being ART-eligible.

Similar to a recent meta-analysis,3 our study showed reduced LTFU and attrition risk among spoke-initiated adult ART patients compared with hub-initiated patients, but no difference in mortality. This supports findings from South Africa,26 Ethiopia,27 and Malawi's Chiradzulu district,28 and contrasts with findings from Malawi's Thyolo district, where increased mortality among PHC-initiated patients was reported.9 Our findings build on a smaller prospective study in Swaziland that reported better adherence to clinic appointments among PHC-initiated patients compared with hospital-initiated patients.29 Lower LTFU rates among spoke-initiated patients might be explained by reduced transport costs,30 improved patient satisfaction (eg, through reduced wait-times and higher provider-to-patient ratios),6 or increased trust in local care providers.31

Other Outcome Predictors

In this study, as in others,19,32 younger age was predictive of LTFU. This may reflect increased mobility of younger adults. In Swaziland, migration to South Africa for work is common among adults in their twenties and thirties.33 Similar to other studies,12 older age was associated with mortality, possibly reflecting decreased rates of ART-induced CD4 restoration among older patients.34,35 As in most studies from resource-limited settings,36 men had higher LTFU and mortality rates. Increased LTFU may be due to employment-related migration33 or gender differences in health-seeking behavior,37 whereas increased mortality may reflect delays in seeking health care,36 increased risk of opportunistic infections,38 or background sex differences in mortality in the general population.37 Early HIV testing, linkage, and retention programs targeting men could significantly improve the impact of Swaziland's ART program on HIV incidence and HIV-related mortality.39

As in other studies,19 being married was predictive of better outcomes, possibly due to increased likelihood of starting ART earlier,40,41 or better ART adherence though availability of social support.42,43 Alternately, being married may be correlated with increased wealth and thus better outcomes,44,45 since, in Swaziland, marriage is traditionally dependent on the ability of the groom to pay “lobola”.46 Interestingly, having a designated treatment supporter, who could be a relative, friend, or community health worker, was associated with reduced LTFU but not mortality. This might indicate that treatment supporters facilitate adherence to clinic appointments but have less impact on ART pill-taking adherence. Alternately, treatment supporters may facilitate reporting deaths that would otherwise be observed as LTFU; this could mask any survival benefit of having a treatment supporter. Further research is needed to explore these findings.

As in other countries,19,32,47 initiating ART in later calendar years was predictive of LTFU. This may reflect increasing undocumented transfer outs,48 or patient frustration with long wait times in crowded clinics, as ART patient-to-provider ratios increased over time.49–52 This may be a symptom of an overburdened health system, making the primary findings of this report (ie, those concerning partial and full decentralization) more important for Swaziland's future plans.

As in many other evaluations,12 markers of advanced HIV disease, including WHO stage III/IV, being bedridden, having a lower body weight, CD4 < 50 cells per microliter, and hemoglobin <8 g/dL, were predictive of mortality and LTFU. Earlier HIV testing, linkage-to-care, pre-ART retention, and ART initiation at earlier disease stages (ie, CD4 > 200 and WHO stage I/II), should be the goal in order for Swaziland's HIV treatment and prevention program to have maximum impact.1

Similar to other evaluations,19 failure to prescribe cotrimoxazole was predictive of LTFU. Whether cotrimoxazole prescription was correlated with better care quality or reduced opportunistic infection risk and mortality53 is unclear from this analysis. Regardless, cotrimoxazole prescription to 100% of ART enrollees should be targeted and could reduce program attrition.


First, these analyses rely on routinely collected and sometimes incomplete data. Missing data on baseline patient characteristics likely introduced nondifferential measurement error. Second, although tracing for patients who missed clinic appointments was recommended, human resource and funding constraints might have resulted in suboptimal tracing efforts for some LTFU patients. In addition, even optimal tracing efforts are not always successful. Therefore, reported LTFU rates might be overestimates due to incomplete death54 or transfer48 documentation and mortality rates might be underestimates.54 There is no evidence that tracing efforts or outcome misclassification was differential across exposure groups, but a tracing study would be needed to exclude this possibility.55 Third, since this was an observational cohort study, effect of down-referral and spoke-initiation may be confounded by unmeasured risk factors. Finally, sample size was limited by funding availability, and there may have been insufficient power to detect true effect of down-referral or spoke-initiation on mortality.


In Swaziland, partial and full decentralization models within a hub-and-spoke system were protective against LTFU and overall attrition and should be continued to facilitate ART program expansion as the 2013 WHO guidelines are adopted. Expansion of partial and full decentralization models may help to limit or reverse concerning trends of increasing LTFU and attrition. Other interventions to reduce ART attrition should include programs targeting earlier ART-initiation, universal cotrimoxazole prescription, compliance with treatment supporter guidelines, and higher male retention.


1. Cohen MS, Chen YQ, McCauley M, et al.. Prevention of HIV-1 infection with early antiretroviral therapy. N Engl J Med. 2011;365:493–505.
2. Tanser F, Bärnighausen T, Grapsa E, et al.. High coverage of ART associated with decline in risk of HIV acquisition in rural KwaZulu-Natal, South Africa. Science. 2013;339:966–971.
3. Kredo T, Ford N, Adeniyi FB, et al.. Decentralising HIV treatment in lower- and middle-income countries. Cochrane Database Syst Rev. 2013;6:CD009987.
4. World Health Organization. Consolidated guidelines on the use of antiretroviral drugs for treating and preventing HIV infection: recommendations for a public health approach. Available at: Accessed February 17, 2014.
5. Bicego GT, Nkambule R, Peterson I, et al.. Recent patterns in population-based HIV prevalence in Swaziland. PLoS One. 2013;8:e77101.
6. Brennan AT, Long L, Maskew M, et al.. Outcomes of stable HIV-positive patients down-referred from a doctor-managed antiretroviral therapy clinic to a nurse-managed primary health clinic for monitoring and treatment. AIDS. 2011;25:2027–2036.
7. Long L, Brennan A, Fox MP, et al.. Treatment outcomes and cost-effectiveness of shifting management of stable ART patients to nurses in South Africa: an observational cohort. PLoS Med. 2011;8:e1001055.
8. Chan AK, Mateyu G, Jahn A, et al.. Outcome assessment of decentralization of antiretroviral therapy provision in a rural district of Malawi using an integrated primary care model. Trop Med Int Health. 2010;15(suppl 1):90–97.
9. Massaquoi M, Zachariah R, Manzi M, et al.. Patient retention and attrition on antiretroviral treatment at district level in rural Malawi. Trans R Soc Trop Med Hyg. 2009;103:594–600.
10. World Health Organization. Antiretroviral therapy for HIV infection in adults and adolescents in resource-limited settings: towards universal access. Recommendations for a public health approach—2006 revision. Available at: Accessed June 6, 2014.
11. World Health Organization. Antiretroviral therapy for HIV infection in adults and adolescents: recommendations for a public health approach—2010 revision. Available at: Accessed June 6, 2014.
12. May M, Boulle A, Phiri S, et al.. Prognosis of patients with HIV-1 infection starting antiretroviral therapy in sub-Saharan Africa: a collaborative analysis of scale-up programmes. Lancet. 2010;376:449–457.
13. Schafer JL, Graham JW. Missing data: our view of the state of the art. Psychol Methods. 2002;7:147–177.
14. Graham JW. Missing data analysis: making it work in the real world. Annu Rev Psychol. 2009;60:549–576.
15. Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York, NY: J. Wiley & Sons; 1987.
16. Royston P. Multiple imputation of missing values. Stata J. 2004;4:227–241.
17. Royston P. Multiple imputation of missing values: update.Stata J. 2005;5:188–201.
18. Royston P. Multiple imputation of missing values: update of ice. Stata J. 2005;5:527–536.
19. Auld AF, Mbofana F, Shiraishi RW, et al.. Four-year treatment outcomes of adult patients enrolled in Mozambique's rapidly expanding antiretroviral therapy program. PLoS One. 2011;6:e18453.
20. White IR, Royston P. Imputing missing covariate values for the cox model. Stat Med. 2009;28:1982–1998.
21. Royston P, Carlin JB, White IR. Multiple imputation of missing values: new features for mim. Stata J. 2009;9:252–264.
22. Wandeler G, Keiser O, Pfeiffer K, et al.. Outcomes of antiretroviral treatment programs in rural Southern Africa. J Acquir Immune Defic Syndr. 2012;59:e9–16.
23. Putter H, Fiocco M, Geskus RB. Tutorial in biostatistics: competing risks and multi-state models. Stat Med. 2007;26:2389–2430.
24. Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika. 1994;81:515–526.
25. Geng EH, Nash D, Kambugu A, et al.. Retention in care among HIV-infected patients in resource-limited settings: emerging insights and new directions. Curr HIV/AIDS Rep. 2010;7:234–244.
26. Fatti G, Grimwood A, Bock P. Better antiretroviral therapy outcomes at primary healthcare facilities: an evaluation of three tiers of ART services in four South African provinces. PLoS One. 2010;5:e12888.
27. Assefa Y, Kiflie A, Tekle B, et al.. Effectiveness and acceptability of delivery of antiretroviral treatment in health centres by health officers and nurses in Ethiopia. J Health Serv Res Policy. 2012;17:24–29.
28. McGuire M, Pinoges L, Kanapathipillai R, et al.. Treatment initiation, program attrition and patient treatment outcomes associated with scale-up and decentralization of HIV care in rural Malawi. PLoS One. 2012;7:e38044.
29. Humphreys CP, Wright J, Walley J, et al.. Nurse led, primary care based antiretroviral treatment versus hospital care: a controlled prospective study in Swaziland. BMC Health Serv Res. 2010;10:229.
30. Geng EH, Bangsberg DR, Musinguzi N, et al.. Understanding reasons for and outcomes of patients lost to follow-up in antiretroviral therapy programs in Africa through a sampling-based approach. J Acquir Immune Defic Syndr. 2010;53:405–411.
31. Arora S, Thornton K, Murata G, et al.. Outcomes of treatment for hepatitis C virus infection by primary care providers. N Engl J Med. 2011;364:2199–2207.
32. Cornell M, Grimsrud A, Fairall L, et al.. Temporal changes in programme outcomes among adult patients initiating antiretroviral therapy across South Africa, 2002-2007. AIDS. 2010;24:2263–2270.
33. African Development Bank. Migration patterns, trends and policy issues in Africa. Available at: Accessed June 6, 2014.
34. Viard JP, Mocroft A, Chiesi A, et al.. Influence of age on CD4 cell recovery in human immunodeficiency virus-infected patients receiving highly active antiretroviral therapy: evidence from the EuroSIDA study. J Infect Dis. 2001;183:1290–1294.
35. Yamashita TE, Phair JP, Munoz A, et al.. Immunologic and virologic response to highly active antiretroviral therapy in the multicenter AIDS Cohort Study. AIDS. 2001;15:735–746.
36. Centers for Disease Control and Prevention. Differences between HIV-infected men and women in antiretroviral therapy outcomes—six African countries, 2004-2012. MMWR Morb Mortal Wkly Rep. 2013;62:946–952.
37. Cornell M, Myer L, Kaplan R, et al.. The impact of gender and income on survival and retention in a South African antiretroviral therapy programme. Trop Med Int Health. 2009;14:722–731.
38. Owens IPF. Sex differences in mortality rate. Science. 2002;297:2008.
39. Suthar AB, Ford N, Bachanas PJ, et al.. Towards universal voluntary HIV testing and counselling: a systematic review and meta-analysis of community-based approaches. PLoS Med. 2013;10:e1001496.
40. Lahuerta M, Lima J, Nuwagaba-Biribonwoha H, et al.. Factors associated with late antiretroviral therapy initiation among adults in Mozambique. PLoS One. 2012;7:e37125.
41. Parrott FR, Mwafulirwa C, Ngwira B, et al.. Combining qualitative and quantitative evidence to determine factors leading to late presentation for antiretroviral therapy in Malawi. PLoS One. 2011;6:e27917.
42. Tiyou A, Belachew T, Alemseged F, et al.. Predictors of adherence to antiretroviral therapy among people living with HIV/AIDS in resource-limited setting of southwest ethiopia. AIDS Res Ther. 2010;7:39.
43. Amberbir A, Woldemichael K, Getachew S, et al.. Predictors of adherence to antiretroviral therapy among HIV-infected persons: a prospective study in Southwest Ethiopia. BMC Public Health. 2008;8:265.
44. Cunningham WE, Hays RD, Duan N, et al.. The effect of socioeconomic status on the survival of people receiving care for HIV infection in the United States. J Health Care Poor Underserved. 2005;16:655–676.
45. McFarland W, Chen S, Hsu L, et al.. Low socioeconomic status is associated with a higher rate of death in the era of highly active antiretroviral therapy, San Francisco. J Acquir Immune Defic Syndr. 2003;33:96–103.
46. African Books Collective. Lobola: It's Implications for Women's Reproductive Rights. Available at: Accessed June 6, 2014.
47. Leroy V, Malateste K, Rabie H, et al.. Outcomes of antiretroviral therapy in children in Asia and Africa: a comparative analysis of the IeDEA pediatric multiregional collaboration. J Acquir Immune Defic Syndr. 2013;62:208–219.
48. Forster M, Bailey C, Brinkhof MW, et al.. Electronic medical record systems, data quality and loss to follow-up: survey of antiretroviral therapy programmes in resource-limited settings. Bull World Health Organ. 2008;86:939–947.
49. Nglazi MD, Lawn SD, Kaplan R, et al.. Changes in programmatic outcomes during 7 years of scale-up at a community-based antiretroviral treatment service in South Africa. J Acquir Immune Defic Syndr. 2011;56:e1–8.
50. Musheke M, Bond V, Merten S. Deterrents to HIV-patient initiation of antiretroviral therapy in urban Lusaka, Zambia: a qualitative study. AIDS Patient Care STDS. 2013;27:231–241.
51. Musheke M, Bond V, Merten S. Individual and contextual factors influencing patient attrition from antiretroviral therapy care in an urban community of Lusaka, Zambia. J Int AIDS Soc. 2012;15(suppl 1):1–9.
52. Duff P, Kipp W, Wild TC, et al.. Barriers to accessing highly active antiretroviral therapy by HIV-positive women attending an antenatal clinic in a regional hospital in western Uganda. J Int AIDS Soc. 2010;13:37.
53. Walker AS, Ford D, Gilks CF, et al.. Daily co-trimoxazole prophylaxis in severely immunosuppressed HIV-infected adults in Africa started on combination antiretroviral therapy: an observational analysis of the DART cohort. Lancet. 2010;375:1278–1286.
54. Bisson GP, Gaolathe T, Gross R, et al.. Overestimates of survival after HAART: implications for global scale-up efforts. PLoS One. 2008;3:e1725.
55. Geng EH, Glidden DV, Bangsberg DR, et al.. A causal framework for understanding the effect of losses to follow-up on epidemiologic analyses in clinic-based cohorts: the case of HIV-infected patients on antiretroviral therapy in Africa. Am J Epidemiol. 2012;175:1080–1087.

antiretroviral therapy; down-referral; outcomes; Swaziland; decentralization

Supplemental Digital Content

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.