Bulletin of the World Health Organization

Early loss of HIV-infected patients on potent antiretroviral therapy programmes in lower-income countries

Martin WG Brinkhof a, François Dabis b, Landon Myer c, David R Bangsberg d, Andrew Boulle c, Denis Nash e, Mauro Schechter f, Christian Laurent g, Olivia Keiser a, Margaret May h, Eduardo Sprinz i, Matthias Egger a, Xavier Anglaret b, for the ART-LINC of IeDEA collaboration


The increasingly widespread use of potent combination antiretroviral therapy (ART) since 1996 has substantially improved the prognosis of patients infected with HIV in industrialized countries.13 In resource-constrained settings in Africa, Asia and South America, where 90% of people with HIV/AIDS live, access to ART has improved substantially: according to WHO estimates, two million people with HIV/AIDS were receiving treatment in low- and middle-income countries in December 2006, which represents 28% of the estimated 7.1 million people in urgent need of treatment.4

The administration of ART to individual patients and the monitoring and evaluation of HIV/AIDS treatment programmes critically depend on regular and complete patient follow-up. Individual treatment decisions can then be made in the light of clinical and laboratory results, and treatment response, complication and mortality rates can be accurately estimated at the programme level.

Using data from a network of treatment programmes in Africa, Asia and South America, we examined the early loss of patients starting ART programmes in low-income countries; this included no follow-up after the initial visit, and loss to follow-up and death in the first 6 months.


Study population

The Antiretroviral Therapy in Lower-Income Countries (ART-LINC) collaboration5 of the International epidemiological Databases to Evaluate AIDS (IeDEA)6 is a network of HIV/AIDS treatment programmes in Africa, Asia and South America that has been described elsewhere.7,8 Of the 23 treatment programmes in low- and middle-income countries approached, 19 agreed to participate and 15 contributed data to the present analysis. All eligible programmes systematically collected prospective data on adolescents and adults aged 16 years or older starting ART, and institutional review boards or ethics committees approved data collection.

Information obtained on patients included age, sex, date of starting ART, type of treatment initiated, date of last contact with the programme, date of death and, when available, CD4-cell count, WHO HIV clinical stage, total lymphocyte count, haemoglobin level and HIV-1 ribonucleic acid (RNA) plasma level at baseline and during follow-up. The most common ART regimens were: stavudine, lamivudine and nevirapine; zidovudine, lamivudine and efavirenz; zidovudine, lamivudine and nevirapine; and stavudine, lamivudine and efavirenz. These four combinations accounted for 66% of all regimens used.7,8 The type of ART regimen was classified as either protease inhibitor (PI)-based [i.e. two nucleoside reverse transcriptase inhibitors (NRTIs) plus one PI, including ritonavir-boosted PI], non-nucleoside reverse transcriptase inhibitor (NNRTI)-based (i.e. two NRTIs plus one NNRTI), or another combination, which included triple NRTI regimens and any other regimen containing a minimum of three drugs. The characteristics of the treatment programmes were also recorded, including procedures in place for tracing patients lost to follow-up. We included all patients who had not previously received antiretrovirals, except for the prevention of mother-to-child HIV transmission, who were aged 16 years or older and whose date of starting ART was documented.


We considered three endpoints that characterized the loss of patients to a programme: no follow-up, loss to follow-up and death in the first 6 months after starting ART. No follow-up was when a patient did not return to the clinic after the ART initiation visit, although the database remained open for an additional 12 months or more. A patient was considered lost to follow-up if the last follow-up visit occurred during the first 6 months after starting ART and the database remained open for an additional 6 months or more. The 6-month interval was chosen to accommodate the longest interval between visits in participating programmes. The database closing date was defined as the date of the most recent follow-up visit in a given patient cohort.

Statistical analysis

Logistic regression with a random effect for the cohorts was used to examine factors associated with no follow-up. We used competing risk models9 to analyse the time to loss to follow-up and the time to death, as measured from the start of ART (i.e. baseline). Competing risk analysis assumes that each individual is exposed to two risks, namely loss to follow-up and death, and accounts for the fact that these risks may not be independent. The competing risk data set was prepared by stacking the data and generating separate strata for death and loss to follow-up, with each patient appearing in both strata. The effect of prognostic factors on outcome was analysed using a Weibull proportional hazard model, controlling for cohort and stratifying by event, thus allowing the baseline hazard to differ between competing risks. Robust variance adjustment was used to allow for the fact that each individual contributed two data points. P-values for the contribution of prognostic factors to the explained variance were derived using a Wald test. The effect of programme factors was evaluated by controlling for individual patient factors.

For some patients, data on CD4-cell count or clinical stage at baseline were missing. We, therefore, created dummy variables that indicated whether or not CD4 cells and clinical stage had been assessed. In sensitivity analyses, we used multiple imputations of CD4-cell counts and clinical stage, as described previously.8 The following variables were considered for inclusion in logistic and competing risk models: sex, age, CD4-cell count (< 25 cells/µl, 25–50 cells/µl, > 50 cells/µl, and not measured), clinical stage, initial ART regimen, and calendar period of ART initiation (< 2001, 2001–2002 and 2003–2004). The choice of CD4 categories reflects the fact that there was little variation in the rate of loss to follow-up for different cell-count subdivisions above 50 cells/µl. The clinical disease stage was categorized as less advanced [i.e. United States Centers for Disease Control (CDC) stage A/B, WHO stage I/II], more advanced (i.e. CDC stage C, WHO stage III/IV) and not assessed. Two programme variables, namely free access to treatment with no cost to patients and type of follow-up (i.e. active tracing versus passive follow-up), were also included in the models. All analyses were carried out using Stata version 9.2 (Stata Corp. LP, College Station, TX, United States of America). The results are presented as odds ratios (ORs) or hazard ratios (HRs) with a 95% confidence interval (CI).


The IeDEA ART-LINC database included a total of 7651 patients who started ART in 15 treatment programmes in Africa, Asia and South America. Of these, 5491 were eligible for the present analysis. The characteristics of programmes contributing data are shown in Table 1. The number of patients on ART increased rapidly between 2001 and 2005, particularly in some African programmes and at the site in India. Eleven sites actively followed patients using telephone calls (often to mobile phones), letters or home visits and 11 provided free access to treatment. The number of patients included in the analysis ranged from 36 in Thailand to 1219 in Malawi and the median baseline CD4-cell count ranged from 45 cells/µl in the township of Khayelitsha to 241 cells/µl in the Cape Town AIDS Cohort (CTAC), both in South Africa. Patient characteristics at baseline are summarized in Table 2. The patients’ median age was 35 years, 2519 (46%) were women and the median CD4-cell count was 105 cells/µl in the 4087 patients for whom data were available. Most patients (1727 or 69%) were at an advanced clinical disease stage when starting ART. A total of 2498 patients (45%) were treated in programmes with active follow-up and 3298 (60%) had free access to ART.

Overall 211 patients (3.8%) were not seen after the ART initiation visit, 880 (16.0%) were lost to follow-up later on and 141 (2.6%) were known to have died in the first 6 months. There was considerable variation across treatment sites (Table 1). Sites with larger numbers of patients were less likely to actively trace patients who did not return to the clinic, and these sites also had higher percentages of patients lost to follow-up (Fig. 1).

Fig. 1. Effect of ART programme size and type of follow-up on the percentage of patients lost to follow-upa
ART, antiretroviral therapy
a The number of patients in each programme includes all those starting ART during the study period.

Individual patient factors associated with the three outcomes are shown in Table 3. The probability that a patient would be lost to the programme, due to either no follow-up after the initial visit or subsequent loss to follow-up, was greater for those starting ART in more recent calendar years. There was a striking increase after the year 2000: the OR for no follow-up in 2003–2004 compared with 2000 or earlier was 5.06 (95% CI: 1.28–20.0) and the corresponding HR for a loss to follow-up in the first 6 months was 7.62 (95% CI: 4.55–12.8). A low baseline CD4-cell count and the absence of a CD4-cell measurement were also associated with a loss to the programme, with particularly strong associations with no follow-up and death. More advanced HIV disease and the absence of clinical stage assessment were strongly associated with the risk of death, but not with no follow-up or a loss to follow-up in the first 6 months. Patients aged 50 years or over were more likely to have no follow-up after the initial visit. Finally, there was no association between the type of ART regimen and any of the three outcomes.

As shown in Table 4, the programme factor of receiving treatment in a fee-for-service programme was associated with an increased probability of no follow-up or death. Furthermore, programmes in which there was no active tracing of patients were more likely to experience losses after the initial visit and less likely to record deaths. In the programmes with active follow-up, the death rates were higher because deaths were more reliably identified.

The results of sensitivity analyses based on multiply imputed data for baseline CD4-cell count and clinical stage were similar to those presented here. Additional tables with these results are available at: http://www.art-linc.org/fileadmin/IeDEA/Publications/ARTLINC_BullWHO2008_WebTables.pdf. Of note, 81% of cases in which disease stage was unknown were classified as having advanced disease following multiple imputations and, in these analyses, advanced stage was associated with no follow-up after the initial visit.


Main findings

Using data from a large collaborative network of ART treatment programmes in resource-limited settings, we investigated three important early outcomes: failure to return to the clinic after the first ART prescription, and loss to follow-up and death in the first 6 months after starting ART. We found that only 3% of patients were known to have died by 6 months, but on average 21% of patients had been lost to programmes by that time, including about 4% who had not been seen since receiving their first ART prescription. The percentage of patients lost to programmes was substantially greater in more recent calendar periods than in the period before the year 2000. This suggests that many sites find it increasingly difficult to follow-up the growing population of patients and to trace those not returning to the clinic. Treating the maximum number of new patients possible has been the top priority for many public sector programmes, with the possible consequence that documenting and tracing patients lost to follow-up has become increasingly inadequate.10

Strengths and limitations

An important strength of our study is the inclusion of a diverse group of clinics and programmes. This made it possible to examine factors associated with both individual patients and programmes that could influence the probability of patient retention. However, we stress that the reasons patients were lost to follow-up were not collected systematically for individuals. Our study has several other limitations. While the ART-LINC cohorts are broadly representative of the types of ART services operating across resource-limited settings, the generalizability of these results requires careful consideration. One limitation of large collaborative databases is that a relatively small number of variables is available for analysis: the standardization and harmonization of data collection across many different sites is challenging. The result is that factors that are more difficult to assess, for example adherence, cannot be examined.

Another limitation is that we were unable to explore fully the relationships between patients lost to programmes and patient numbers, the approach to tracing patients, staffing and infrastructure. Detailed standardized up-to-date information was not available and the size of a treatment programme was closely related to whether or not patients were actively traced. It was therefore not possible to link unequivocally time trends in the loss to follow-up to changes in specific programme characteristics over time. Programme size and other features of ART services that affect patient outcomes require further investigation to inform decisions on how best to deliver ART to large numbers.

In the context of other studies

We focused on the first 6 months of treatment, which is a crucial period for the long-term success of ART. The initial response to ART has long-term prognostic significance, and optimizing adherence in early months is important for ensuring long-term immunological and virological success.11,12 Data from the ART-LINC collaboration8 and other treatment programmes, for example the Médecins Sans Frontières (MSF) programme in Malawi, show that loss to follow-up and death mostly occur in the first 6 months after ART initiation and that community support improves outcomes.13,14 Not all studies confirm this, however. Data from a South African ART programme demonstrate that while mortality decreased rapidly after ART initiation, the rate of loss to follow-up remained fairly constant during the first 2 years.15

Other studies of treatment programmes in resource-limited settings also found high rates of loss to follow-up. For example, in the urban primary health care setting of Lusaka, Uganda, 3406 (21%) of 16 199 patients starting ART in 2004–2005 were more than 30 days late for a scheduled pharmacy appointment.16 An evaluation of the HIV Drug Access Initiative of the Ministry of Health of Uganda and the Joint United Nations Programme on HIV/AIDS (UNAIDS), in which patients paid reduced prices for their medications, found that 114 (24%) of 476 patients were lost to follow-up in the first year.17 Other studies found lower rates of loss to follow-up: in Port-au-Prince, Haiti, for example, only 71 (7%) of 910 patients were lost over a median follow-up period of 13 months.18 In a large observational cohort of MSF HIV/AIDS programmes, 4.8% of patients were lost to follow-up over a median period of 4.1 months.19 Unfortunately, often study results are not directly comparable because of differences in the definition of a loss to follow-up, or because no clear definition is reported. Some cohorts were established in a clinical research context, and included relatively small numbers of patients who were closely monitored.2023 These studies reported low rates of loss to follow-up that will not reflect the realities of scaling up ART programmes.

Reasons for losses to follow-up

Few programmes in resource-limited settings systematically assess the reasons patients are lost to follow-up, but surveys found that a substantial proportion had died. For example, among 727 patients lost to follow-up at the Lighthouse Clinic in Lilongwe, Malawi, 30% had died, 19% had transferred to another facility, 22% were alive and on treatment, 3% had stopped treatment and 26% could not be found. The last group had a low median CD4-cell count, suggesting that some may also have died.24 A smaller study from northern Malawi investigated the fate of 253 patients lost to follow-up and reported that 50% had died, 23% were alive and 27% could not be traced.25 Similarly, among 801 patients traced after they missed scheduled visits in Lusaka, Zambia, 46% had died.26 Taken together, these data suggest that about 50% of patients lost to follow-up in lower-income settings may have died.

The present analysis extends our previous study of mortality in high- and low-income countries8 and shows that free access to treatment is a critical determinant of programme retention and mortality. This association was particularly strong for patients who had no follow-up after ART was started, but less strong for those lost to follow-up later on. Interestingly, in the UNAIDS–Ugandan Ministry of Health HIV Drug Access Initiative, a fee-for-service programme, the majority of patients (65%) who were lost did not return after their initial visit.17 As discussed previously,8 paying for initial antiretroviral treatment does not mean that households can pay later on. Associations were also found between providing ART free of charge and a higher probability of sustained suppression of viral replication and better adherence to therapy.27

In individual patients, older age and profound immune suppression were associated with early losses to treatment. Some older people may fail to return because they do not want to burden their families, who may contribute to costs. Patients with advanced HIV infections may not return because they are too weak. Access to transport is also important. For example, ownership of a bicycle was associated with a reduced loss to follow-up in one programme, but patients’ access to transport was not consistently assessed. In the Lusaka programme, some patients stopped treatment because of high transport costs.25 In the large Academic Model for the Prevention and Treatment of HIV/AIDS (AMPATH) programme in Eldoret, Kenya, men were more likely to be lost to follow-up than women.28 In our study, there was some evidence that men were less likely to return to the clinic than women.


Our results support the notion that evaluations of the scale-up of ART in resource-limited settings should consider not only the number of new patients starting ART but also the number remaining in long-term care: the number lost to follow-up is an important indicator of programme effectiveness.15 Our results indicate that better ART outcomes, including higher programme retention rates, may be obtained in services that have smaller numbers of patients and, therefore, that population coverage should be achieved with smaller decentralized facilities rather than a few large programmes. In general, given the large numbers of patients and the limited resources facing health services, developing strategies that prevent patients from missing appointments may be more cost-effective than tracking those who do not return.10 Future studies should address the causes of the late initiation of ART15 as well as the barriers preventing patients from returning to clinics, record transfers to other programmes and assess mortality in patients lost to programmes. In conclusion, early patient losses to ART programmes are increasingly common in resource-limited countries. This should not detract from the fact that many patients benefit from the introduction and scaling-up of ART or from the need to continue efforts to improve access to therapy. ■


We thank Jack Whitescarver and Paolo Miotti (NIH/OAR), Carolyn Williams (NIH/NIAID), Michel Kazatchkine, Jean-François Delfraissy, Brigitte Bazin and Séverine Blesson (ANRS) for encouraging and supporting this collaborative work.

The organization of the Antiretroviral Therapy in Lower-Income Countries (ART-LINC) Collaboration of the International epidemiological Databases to Evaluate AIDS (IeDEA) is given at: http://www.art-linc.org

Funding: The ART-LINC Collaboration is funded by the United States National Institute of Health (Office of AIDS Research) and the French Agence Nationale de Recherches sur le Sida et les hépatites virales (ANRS).

Competing interests: None declared.



  • Institute of Social and Preventive Medicine, Bern, Switzerland.
  • Unité INSERM 593, Institut de Santé Publique, Epidémiologie et Développement (ISPED), Université Victor Segalen, Bordeaux, France.
  • School of Public Health and Family Medicine, University of Cape Town, Cape Town, South Africa.
  • University of California, San Francisco, CA, United States of America.
  • Mailman School of Public Health, Columbia University, New York, NY, United States of America.
  • Universidade Federal do Rio de Janeiro, Rio de Janeiro, RJ, Brazil.
  • Institut de Recherche pour le Développement (IRD-UR36), Montpellier, France.
  • University of Bristol, Bristol, England.
  • Hospital de Clínicas, Porto Alegre, RS, Brazil.