Outcome assessment of decentralization of antiretroviral therapy provision in a rural district of Malawi using an integrated primary care model

Authors


Corresponding Author Monique van Lettow, Dignitas International, c/o PO Box 1071, Zomba, Malawi. Tel.: +265 888 117736; Fax: +265 1 527 558; E-mail: m.vanlettow@dignitasinternational.org

Summary

Objective  To assess the effect of decentralization (DC) of antiretroviral therapy (ART) provision in a rural district of Malawi using an integrated primary care model.

Methods  Between October 2004 and December 2008, 8093 patients (63% women) were registered for ART. Of these, 3440 (43%) were decentralized to health centres for follow-up ART care. We applied multivariate regression analysis that adjusted for sex, age, clinical stage at initiation, type of regimen, presence of side effects because of ART, and duration of treatment and follow-up at site of analysis.

Results  Patients managed at health centres had lower mortality [adjusted OR 0.19 (95% C.I. 0.15–0.25)] and lower loss to follow-up (defaulted from treatment) [adjusted OR 0.48 (95% C.I. 0.40–0.58)]. During the first 10 months of follow-up, those decentralized to health centres were approximately 60% less likely to default than those not decentralized; and after 10 months of follow-up, 40% less likely to default. DC was significantly associated with a reduced risk of death from 0 to 25 months of follow-up. The lower mortality may be explained by the selection of stable patients for DC, and the mentorship and supportive supervision of lower cadre health workers to identify and refer complicated cases.

Conclusion  Decentralization of follow-up ART care to rural health facilities, using an integrated primary care model, appears a safe and effective way to rapidly scale-up ART and improves both geographical equity in access to HIV-related services and adherence to ART.

Introduction

In 2004, widespread roll-out of antiretroviral therapy (ART) in Malawi’s public sector began. Since that time, Dignitas International (DI), a Canadian humanitarian NGO, has been assisting the Malawi Ministry of Health (MOH) HIV Unit (now Department of HIV and AIDS) in supporting scale-up of HIV services in Zomba District in southern Malawi. Initial public sector ART provision focused on developing HIV care and treatment capabilities at a tertiary hospital clinic (Tisungane Clinic, Zomba Central Hospital (ZCH)), which in a predominantly rural district such as Zomba meant that services were located far away for the majority of the rural poor.

In January 2007, because of human resource constraints within the central hospital clinic, and to improve geographical equity in ART access for the rural poor, DI began to pilot support of the Zomba District Health Office (DHO) in decentralizing ART follow-up to six rural health centres in the district. Implementation was expanded to 16 sites by April 2008. This article describes the impact of rapid roll-out of this decentralized approach on defaulting and death.

Zomba District in southern Malawi comprises 670 000 inhabitants. More than 80% of the population in Zomba is rural (Malawi National Statistics Office 2008). HIV prevalence is high with antenatal care surveillance data at sentinel health centres ranging from 12% in an urban health care centre to >30% at a rural hospital. The estimated prevalence among adults 15–49 in southern Malawi is 16.5% (Malawi Ministry of Health 2008). The majority of the rural population works as subsistence farmers for <$4 US per day.

Public health services in Zomba District are provided at one central hospital and 33 health centres, some of which are managed by the mission sector (Christian Health Association of Malawi). Services and medications are provided by the MOH without user fees. A staff shortage within the public health system remains a major bottleneck to service provision. There are currently two Malawian physicians in MOH health posts in the district, and approximately 40% of health care posts remain vacant.

In the context of existing human resource constraints, the rapid scale-up of ART has been remarkable, with over 60 000 patient visits to Tisungane Clinic in 2008, managed by three clinical officers (DI), five nurses (two DI, three MOH), five patient care attendants (MOH), two ward attendants (MOH) and seven expert patients. ART specialist consultant services were available from two physicians: the Clinic Coordinator (DI) and a Paediatric ART Consultant seconded from the Baylor Pediatric AIDS Corps in Malawi. In addition, the ZCH Chief of Paediatrics followed special paediatric cases with HIV/TB co-infection or with severe malnutrition through the clinic. As of June 2009, 10 000 people (63% are female, 37% male; 10% are children under 14 years of age) have been enrolled on ART through Tisungane Clinic.

The Zomba District ‘integrated primary care model’ for decentralized ART is unique within Malawi. HIV services are integrated into routine outpatient services at rural health centres. All provision of patient care is by MOH employees through the Zomba DHO, and decentralization (DC) is coordinated by one clinical officer (DI), with technical support from the Clinic Coordinator.

There are few studies that have analysed the impact on programme retention associated with DC of ART delivery. High transport costs and geographical distance for patients who do not have access to decentralized care impact both ART uptake and follow-up (Zachariah et al. 2006; Yu et al. 2007; Deribe et al. 2008). Médicins Sans Frontières (MSF) has published literature on their decentralized programmes in South Africa demonstrating faster enrollment into treatment and better retention of patients (Bedelu et al. 2007; Fredlund & Nash 2007). Another recent publication by MSF in a nearby district, Thyolo, Malawi, using a different DC model, demonstrated a decreased loss to follow-up at decentralized sites, but a higher rate of reported deaths (Massaquoi et al. 2009).

The objective of this study was to establish the impact on defaulting and mortality, when DC of ART provision occurs employing a horizontal model of integrated primary care using lower health care worker cadres working within the existing public health system, with NGO support for training, supportive supervision and mentorship.

Methods

Study design, setting and description of decentralization model

A retrospective cohort analysis was performed using routine data from standard MOH ART monitoring tools. Follow-up outcomes were compared between patients with ART managed at the central hospital tertiary clinic against those decentralised and managed at rural health centres.

Antiretroviral therapy provision was implemented per MOH guidelines whereby all patients assessed as being in WHO Clinical Stage 3 or 4, or with a CD4 count <250 cells/mm3 (regardless of stage) are eligible for ART initiation. Once patients are assessed for eligibility, patients and guardians are asked to return for both group and individual counseling and education sessions to prepare for ART. After initiation, patients are followed up after 2 weeks and from then on, provided there are no problems, routine follow-up is arranged every month. After 6 months, patients may be followed less often pending provider assessment of satisfactory adherence. The first line ART regimen in Malawi is stavudine, lamivudine and nevirapine in fixed-dosed combination. Alternative first line and second line regimens are available for patients who experience treatment toxicity or failure because of resistance. (Malawi Ministry of Health 2004)

Decentralization of ART services in Zomba District was accomplished using a four-step model. The first step began in January 2007 with health centre site assessment and selection, by the DC coordinator and the DHO. Sites were chosen based on standardized MOH criteria, which included the following: (i) available medical assistants and/or nurses to staff the clinic; (ii) available HIV Testing And Counseling (HTC) services; (iii) the presence of adequate counseling and consulting rooms; (iv) capability for safe storage of drugs. DI assisted with logistical set up of rooms and storage, and facilitated HTC training and HTC supervision, and was given permission by the MOH to formally open the selected decentralized sites for ART services under the supervision of Tisungane Clinic. The second step involved capacity building and training of health centre staff using a standardized national training course. In total, 41 clinicians and 98 nurses in Zomba District were trained in ART provision by the DC Coordinator, Clinic Coordinator and MOH trainers. Trained ART providers (clinicians and nurses), counselors and clerks (patient care attendants) were attached to Tisungane Clinic to assess the amount of ongoing supportive mentorship that would be required at each site. Not all trainees have remained in district or work at the sites of analysis. The third step involved implementation of ART services through integration of ART follow-up visits with primary outpatient care services at health centres, whereby patients queuing for HIV-related services could access them anytime as a part of regular outpatient services. Patients were decentralized for ART follow-up from the mother clinic (ZCH) to a health centre if they met several criteria: (i) adults or older children on first line ART (or alternative first line) for more than 3 months and stable; (ii) no evidence of active opportunistic infection or drug intolerance; (iii) ART provider confidence in patient adherence; (iv) patient lives in location closer to health centre than hospital. From March 2007, health centres began to provide ART with 16 of 24 planned decentralized sites starting during the study period. The fourth step was ongoing mentorship and supportive supervision provided by the DC Coordinator or Clinic Coordinator on a biweekly or monthly basis depending on the number of patients followed at the site, with priority given to higher volume sites. Initially, complicated cases or patients developing unusual opportunistic infections (OIs) or toxicities would be referred to be seen during these mobile clinics. Following a 1-day intensive Pediatric ART Training Course, mentorship at selected sites (5 of 16 HCs included in cohort) regarding Pediatric HIV/ART was also provided during mobile clinics monthly or every 2 months by the Baylor Pediatric AIDS Consultant from October 2007 until December 2008. Over time as health centre staff became more comfortable with managing complicated follow-up cases independently, mentorship regarding ART initiation at health centres by the DC coordinator also began at selected sites, starting in April 2008. From April 2008, 7 out of 16 decentralized sites began to provide ART initiation services.

Data collection and statistical analysis

Data collection was completed using MOH standardized registers and mastercards from ZCH and 16 rural health centres from the period of clinic inception (1 October 2004) until 31 December 2008. Study outcomes were defined as follows: (i) died – for any reason while on ART; (ii) defaulted – a patient on ART who has not been seen at the ART provision facility for >3 months after intended date of follow-up. Outcomes were censored on 31 December 2008; therefore, maximum follow-up time was 50 months.

Differences between groups were compared using the Chi-square test for categorical variables and t-tests for continuous variables. Multivariate logistic and Cox proportional hazard regression models were used to evaluate the relationship between DC and risk of dying or defaulting over the study follow-up period from time of ART initiation. Adjustments in multivariate analysis were made for sex, age, clinical stage at initiation, type of regimen, presence of side effects because of ART, duration of treatment and duration of follow-up at site of care. Differences in survival time between strata were determined using Kaplan–Meier survival plots. Data analysis was performed using SPPS v.17, SAS v.9.2, and plots created using the R statistical computing package v.2.9.

Results

Comparison of baseline characteristics and outcomes of cohorts

There were 8093 patients who started ART during the study period. Of these, 4653 were followed at the mother clinic and 3440 at a decentralized health centre. From April 2008, 1005 of the decentralized patients were initiated on ART at decentralized sites. A description of the differences between the central hospital tertiary clinic and the decentralized sites is presented in Table 1. Table 2 demonstrates the baseline characteristics at initiation and treatment outcomes of patients on ART being followed in either cohort. At health centres, there was a lower proportion of children followed and a higher proportion of women. In addition, there was a higher proportion of patients initiated because of low CD4 (and WHO clinical stage 1 or 2) or WHO clinical stage 3. At the mother clinic, there was a higher proportion of patients starting on ART as a result of being WHO clinical stage 4. Patients were followed up for a total follow-up time of 4243 person-years in the mother clinic and 2309 person-years in the health centre cohort.

Table 1.   Role of health staff in HIV/AIDS care in mother clinic compared with health centres
CategoryHospitalHealth Centers (HC)
  1. ART, antiretroviral therapy; DHO, District Health Office; MOH, Ministry of Health; OIs, Opportunistic Infections.

PhysiciansOne Clinic Coordinator (Dignitas International) conducting specialist consultations on patients with complicated ART-related issues or complex diagnostic cases; daily to biweekly clinical mentorship of clinical officers and nursesSame one Clinic Coordinator (Dignitas International) conducting specialist consultations on patients with complicated ART-related issues or complex diagnostic cases; monthly clinical mentorship of clinicians and nurses at some HC sites
One Pediatric AIDS Consultant (Baylor Pediatric AIDS Initiative) conducting specialist consultations on paediatric patients 2 days a weekSame one Pediatric AIDS Consultant (Baylor Pediatric AIDS Initiative) conducting 1 day specialist paediatric ART training for health centre staff at five HCs followed by monthly to q2 monthly mentorship over a 1 year period or until HC staff became comfortable with managing paediatric ART
CliniciansThree Clinical Officers (Dignitas International) conducting patient consultations: opportunistic Infections (OIs), staging, ART initiation, primary HIV care, inpatient consultsOne Decentralization Coordinator – Clinical Officer (Dignitas International) seeing problem cases (OIs, staging, ART initiation), supervising clinics and mentoring clinicians, nurses and counselors; biweekly to monthly mobile support clinics
Five Clinical Officers (Ministry of Health) trained in HIV Management and ARV provision: manage inpatients and outpatients with HIV and refer for ART care to mother clinic, identification of patients for HTC and stagingTwenty Medical Assistants and five Clinical Officers trained in ARV Provision for Zomba DHO (Ministry of Health) at sites of analysis; typical Ministry of Health primary care staffing at most health centres consisted of 0–1 Medical Assistants; larger health centres with catchment areas of >40 000 may have had a Clinical Officer; one large urban health centre had three Clinicians (one Clinical Officer, and two Medical Assistants); clinician staff would conduct patient consultations (OIs, staging, ART follow-up) integrated into regular outpatient primary care clinics
NursesFive Nurses (three Ministry of Health, two Dignitas International) supporting clinicians: prepare and counsel people for ART initiation, monitor ART recipients, manage drug supply and dispense medication, adherence counseling, manage some minor ART toxicities, administer vincristine for Kaposi’s Sarcoma patients; supervise data collection and defaulter tracingSeventy Nurses trained in ARV Provision for Zomba DHO (Ministry of Health) at sites of analysis; Typical Ministry of Health primary care staffing at most health centres consisted of 1–2 nurses in small health centres (catchment area <10 000) to up to seven nurses in one large urban health centre; Nursing staff would conduct patient consultations (OIs, staging, follow-up) integrated into regular outpatient primary care clinics; supervise defaulter tracing and data collection
Patient Care AttendantsOne ART Clerk, one ART Clerk/Counsellor, three Nutrition Counsellors and Adherence Counsellors (Ministry of Health): collect and organize MOH mastercards and register patients on ART, nutrition assessments and counseling and administration of therapeutic feeding, counseling for ART initiation, adherence counsellingSixteen ART clerks and sixteen ART Counsellors trained for Zomba DHO (Ministry of Health) at sites of analysis: collect and organize MOH mastercards and register patients on ART, counseling for ART initiation, adherence counselling
Health Surveillance AssistantsNot utilizedMonthly to quarterly defaulter tracing; HSAs are assigned to each health centre catchment area (supposedly one for every 2000 people although in some sites it is up to one for every 10 000 people) and are organized by the HC In Charge Nurse to trace defaulters
Expert PatientsSeven Expert Patients (Dignitas International): patient escorting, assist ART Clerk, trained in initiation and adherence counseling, trained in light nursing tasks (vital signs, anthropometry – heights, weights, MUACs)Not utilized
Home-Based Care Volunteers and Commu-nity-Based OrganizationsNot utilizedTrained to provide peer support and also advocate for better service delivery
Promote health in the community among PLWHAs
Support adherence to ART
Table 2.   Baseline characteristics at initiation and outcomes of patients followed on antiretroviral therapy at the tertiary clinic (hospital) vs. health centre (decentralized)
VariableHospital n (%)Decentralized n (%)P-value
Total46533440 
Baseline Characteristics
 Sex
  Female2805 (60.3)2291 (66.6)0.01
  Male1848 (39.7)1149 (33.4)
 Age
  Child (0–14)528 (11.3)250 (7.2)<0.01
  Young Adult (15–24)492 (10.6)353 (10.3)0.66
  Adult (25 and over)3633 (78.1)2837 (82.5)<0.01
  Age (years) [mean (std. dev.)]33.2 (±13.6)35.0 (±16.7) 
 WHO clinical stage at initiation
  Stage 1 or 2 with CD4 <250 cells/mm31183 (25.4)962 (28.0)0.011
  Stage 32394 (51.5)1906 (55.4)<0.01
  Stage 41076 (23.1)572 (16.6)<0.01
 Length of time on treatment
  Months [mean (std. dev.)]11.0 (±11.1)13.7 (±10.8)<0.01
  Median (IQR)6 (3–16)11 (5–20) 
  Length of time at site of follow-up (months) [mean (std. dev.)]10.9 (±11.1)8.1 (±6.7)<0.01
  Median (IQR)6 (3–16)6 (2–13) 
 Outcomes
  Defaulted698 (15.0)253 (7.4)<0.01
  Died525 (11.3)99 (2.9)<0.01

The risk of defaulting from care

Decentralization of care was associated with lower odds of defaulting, compared with patients who remained on care at the tertiary clinic. Figure 1 shows the difference in probability of defaulting among patients who were followed on ART at hospital and health centres. In multivariate logistic analysis, lower defaulting [adjusted OR 0.48 (95% C.I. 0.40–0.58)] was independently associated with being decentralized when adjusted for sex, age, clinical stage at initiation, type of regimen, presence of side effects because of ART and duration of treatment and follow-up. Cox regression models showed that during the first 10 months, those decentralized were approximately 60% less likely to default (HR0–10 months = 0.38; 95%CI: 0.32–0.45) than those not decentralized. After 10 months of follow-up time, decentralized study subjects, while still significantly less likely to default than their non-decentralized counterparts, were approximately 40% less likely to default (HR>10 months = 0.60; 95%CI: 0.44–0.82). Multivariate Cox modelling showed that DC was significantly associated with a reduced risk of defaulting from 0 to 8 months (HR = 0.15; 95%CI: 0.11–0.20), as well as from 8 to 25 months (HR = 0.65; 95% CI: 0.44–0.97) adjusted for age, gender, reason for ART treatment initiation, regimen type and side effects. However, DC status was not significantly associated with defaulting after 25 months of follow-up (HR = 1.00 95% CI, 0.28–3.60). Table 3 shows the multivariate adjusted risk of defaulting modelling defaulting status as time spent decentralized. Figure 2 shows data from the entire combined cohort demonstrating that young adults between the ages of 15–24 have a higher risk of defaulting than infants, children and adults. However, in the multivariate model (Table 3), age and gender were not significantly associated with the risk of defaulting over the study follow-up period. Initiation of ART treatment was significantly associated with the risk of default from treatment, with those initiating because of WHO clinical stage 4 AIDS being more likely to default [adjusted HR 1.60 (95% C.I. 1.39–1.86]. Figure 3 shows data from the entire combined cohort demonstrating the impact of clinical stage at initiation on defaulting. The presence of treatment-related side effects was also significantly associated with risk of defaulting but only after 18 months of follow-up where subjects with side effects were more than five times as likely to default as those without treatment-related side effects [adjusted HR 5.30 (95% CI: 2.90–9.69)].

Figure 1.

 Probability of defaulting among patients followed on antiretroviral therapy at the mother clinic (ZCH) and decentralized to health centres.

Table 3.   Multivariate risk of defaulting adjusted for time spent decentralized
VariableHazard Ratio95% C.I.
  1. *Reference groups for analysis: Sex, Female; WHO clinical stage at initiation, WHO 3.

Time spent decentralized (months)
 0–100.730.70–0.77
 >10–200.920.89–0.95
 >20–350.950.91–0.99
 >350.930.85–1.01
 Sex*0.960.87–1.14
 Age1.010.99–1.04
WHO clinical stage at initiation*
 Stage 1 or 2 with CD4 <250 cells/mm30.790.66–0.94
 Stage 41.581.36–1.82
Side Effects (Period of follow-up time in months)
 0–80.910.70–1.18
 >184.682.55–8.57
Figure 2.

 Probability of defaulting in relation to age class at initiation.

Figure 3.

 Probability of defaulting in relation to WHO clinical stage at initiation.

Risk of mortality

Figure 4 shows the difference in probability of survival among patients who were followed on ART at hospital and health centres. In multivariate logistic analysis, lower mortality [adjusted OR 0.19 (95% C.I. 0.15–0.25)] was independently associated with being decentralized. Like defaulting, multivariate Cox modelling showed that DC was significantly associated with a reduced risk of death from 0 to 8 months (HR = 0.15; 95%CI: 0.11–0.20), as well as from 8 to 25 months (HR = 0.65; 95% CI: 0.44–0.97) adjusted for age, gender, reason for ART treatment initiation, regimen type and side effects. However, DC status was not significantly associated with death after 25 months of follow-up (HR = 1.00 95% CI, 0.28–3.60). Table 4 shows the multivariate adjusted risk of defaulting modelling defaulting status as time spent decentralized. Multivariate modelling demonstrated that the increased risk of deaths among males was significant during the first 8 months of study follow-up [adjusted HR 1.40 (95% C.I. 1.17–1.67)] but this increased risk was not observed after the first 8 months [adjusted HR 0.94 (95% C.I. 0.63–1.40)].

Figure 4.

 Probability of survival among patients followed on antiretroviral therapy at the mother clinic (ZCH) and decentralized to health centres.

Table 4.   Multivariate risk of death adjusted for time spent decentralized
VariableHazard Ratio95% C.I.
  1. *Reference groups for analysis: Sex, Female; WHO clinical stage at initiation, WHO 3.

Time spent decentralized (months)
 0–100.680.63–0.75
 >10–200.910.87–0.95
 >200.950.89–1.01
Sex (Period of follow-up time in months)*
 0–81.371.14–1.63
 >8–220.870.58–1.30
 >220.610.17–2.28
 Age1.051.02–1.08
WHO clinical stage at initiation*
 Stage 1 or 2 with CD4  < 250 cells/mm30.450.34–0.59
 Stage 41.921.63–2.28
Side effects (Period of follow-up time in months)
 0–81.140.84–1.53
 >8–105.583.49–8.93
 >195.001.59–15.8

Discussion

Decentralization of ART provision employing a sustainable model of HIV service integration into primary care at health centres in the context of a resource-limited setting appears to be both safe and efficacious. Patients followed at rural health centres had a lower defaulter rate compared with those followed at the main hospital. The mortality rate at rural health centres also was lower than those followed at the main hospital.

In Zomba District, DC of ART services has diminished the burden on the strained human resource capacity at the hospital level, by spreading the increasing patient volume associated with successful ART roll-out from the tertiary care centre to be absorbed into the primary care system in the district (see Table 1). Despite the fact that the total patient cohort during the time period of the study went from 3000 patients enrolled on ART to over 8000 patients, the staffing requirements at Tisungane Clinic and monthly patient volumes (5000 patient visits per month) have not increased.

The mortality difference between health centres and hospitals found in our model contrasts with results found in a similar study performed in Malawi (Massaquoi et al. 2009) where mortality was higher in health centres than in hospitals. The difference in outcomes between the two studies may be related to aspects of the Zomba District DC model, which promotes improved utilization of primary vs. tertiary services within the existing health system. More stable patients with no toxicities or OIs are selected for DC by the mother clinic staff, whereas complicated patients with severe OIs, treatment failure or adverse drug reactions are kept at the tertiary care hospital. Lower cadre health workers at health centres are also mentored by the DC Coordinator to triage and refer complicated cases to the tertiary care centre because the primary care staff need to be able to integrate HIV services into the outpatient care they can provide without additional NGO support for staffing. Selection of patients who are appropriate for primary care accounts for the fact that mortality is higher at the mother clinic, but should not impair quality of services overall as it decreases the primary care burden on strained tertiary care clinics and moves it to primary care sites.

Our experience with this model of DC is that the provision of ART has had a positive effect on the general quality of primary care with improvement in drug supply, diagnostic services, monitoring, staff training, supervision, support and on-site mentorship and infrastructural improvements. The main challenges in delivery of ART services at decentralized sites included retention of health centre staff and rapid turn over of senior level management at the DHO. In addition, access to certain OI drugs and laboratory services that are necessary for DC of primary care ART services into decentralized health centres is not currently available through the Malawi Essential Health Package for Health Centers. Adoption of a decentralized model of care must be accompanied by policies and planning to ensure adequate human resource capacity and expanded access to essential drugs and relevant laboratory services are included as a part of primary health care strengthening.

There were several associations with ART programme attrition found in our analysis that deserve further comment. First, the association of initiation of ART at late clinical stage (WHO clinical stage 4) with increased defaulting compared with those initiated because of immunological staging stresses the importance of initiatives encouraging earlier uptake of testing and referral to treatment and supporting scale-up of decentralized CD4-based initiation (e.g. increasing laboratory and sample-transport capacity, development of point-of-care testing) in resource-limited settings. Second, in the comparison of outcomes between age classes, the association of the lowest survival rate with infants, and the highest default rate in young adults reinforces the need to support programming in paediatric HIV that improves detection of HIV in infants and adherence support measures targeting young adults. Finally, the fivefold increase in defaulting in those who experienced treatment-related side effects after 18 months of follow-up is almost certainly because of the use of stavudine in this setting, implying an urgent need to reassess the continued use of stavudine-based first-line regimens in scale-up in high prevalence settings.

The strength of the study is that the data originates from operational programme information collected as part of routine monitoring and evaluation from a national public system. However, study limitations include concerns regarding accuracy and consistency that are associated with use of operational data. In addition, because data were extracted from routine monitoring and evaluation indicators, information that might be included in a prospective study (e.g. viral load, CD4, follow-up data on new OIs) is not available, as it is not a part of routine clinical care for ART provision under the Malawi MOH public health model. Importantly, there is a significant selection bias programmatically in that more stable patients are selected for DC and initiation of ART at decentralized sites, however, attempts were made to adjust for this by adjusting for reason for initiation as well as length of time on treatment.

In the future, there is need for ongoing study as retention or mortality patterns will change as the cohort matures. Cost effectiveness analysis of interventions comparing various models of DC would also provide useful information to inform policy-makers.

Acknowledgements

We are grateful to the ART providers at Zomba Central Hospital and in Zomba District for their dedication to patient care. We are grateful to the Zomba Central Hospital and Zomba DHO Management and the MOH of Malawi for the collaboration and encouragement in trying to implement HIV-related activities. We are particularly grateful to the DI and Tisungane Clinic Data Management Team for their meticulous work with data collection and support. We thank Darren Brenner for his contributions to the analysis; Drs. Richard Bedell, Jan Hajek, Andrew Pinto and Michael Schull for their useful comments on this paper; Dr. Kevin Clarke from Baylor Pediatric AIDS Corps for both his collaboration in ART scale-up in Zomba and his comments on this paper.

Conflicts of interest

The authors have declared that they have no conflicts of interest.

Ancillary