This collaborative analysis of ten pediatric treatment programs in four countries in Southern Africa found that one-year mortality in children starting ART varied across treatment programs, with heterogeneity partly explained by differences in prognostic factors at the start of ART, and differences in LTFU. Our study illustrates that ignoring LTFU may lead to substantial underestimation of mortality: mortality in children remaining in care overall was below 5% but was estimated to be about twice as high at the program level, when considering deaths in children LTFU.
We examined mortality and risk factors for death in children remaining in care, and then estimated program-level mortality. Previous analyses of mortality in patients starting ART have generally truncated (censored) follow-up time in patients LTFU, thus assuming that their mortality experience is comparable to similar patients remaining in care.12;24, 15
This assumption is problematic: many patients who are LTFU stop taking ART, and their mortality is high. In addition, patients may not return for a follow-up appointment because they have died. A meta-analysis of studies that traced adult patients LTFU found that in sub-Saharan Africa mortality in patients whose vital status could be ascertained was 46% (95% CI 39%–54%).7
Interestingly, there was an inverse relation between mortality among those LTFU and the rate of LTFU. We took this into account and used the mortality estimate that corresponded to the rate of LTFU observed in a given program. A recent analysis of patients LTFU in the rural Médecins sans Frontières (MSF) program in Chiradzulu, Malawi found that mortality among patients LTFU was similar for adults and children,25
thus supporting our use of estimates from studies in adults. Another study from Lilongwe, Malawi, described a mortality estimate of 33% (CI 20.4–48.4%) among children LTFU.26
The high early mortality in infants and young children means that adult estimates of mortality in those LTFU are not necessarily applicable to cohorts which include a large proportion of very young children.27;28
Our sensitivity analyses showed, however, that combined results were fairly robust to different mortality assumptions for children LTFU.
The context of LTFU in children is different and more complex compared to adults: children are dependent on caregivers, and child outcomes may be related to those of their parents. Even though caregivers are motivated to support children to adhere to ART, the caregiver may change frequently due to ill health, death or adult migration. In addition, unlike adult ART provision, most pediatric ART services remain located in larger hospitals in urban centers and the cost of transport and inconvenience of having to attend separate clinics for caregiver and child may contribute to LTFU.29–31
Interestingly, at one of the few family centered ART clinics, it has been shown that children whose caregiver is in HIV care at the same clinic experience better outcomes.32
The cost of pediatric drugs and difficulty of giving children large amounts of medicine which may be unpalatable are further obstacles to effectively remaining in care. 29;31;33
Furthermore, many caregivers are elderly grandparents for whom these obstacles may become a substantial barrier to continuing on treatment.28;31
Some of the children LTFU may have continued treatment at another clinic, perhaps a primary health care center closer to their home, without notifying the program where ART was started (“silent” transfer out).
Our study included children who were treated in ten ART programs in four countries in Southern Africa, including pediatric programs and programs mainly treating adults, and programs at tertiary and lower levels of care. These sites may not necessarily reflect the situation of this region as a whole: all sites were located in urban areas and cohorts from South Africa with a link to a research program predominated. However, this is one of the largest pooled analyses of children on ART ever done, and from one of the regions of the world most heavily affected by the HIV epidemic. Results should therefore be applicable to many other children on ART.
Few previous studies reported mortality among children starting ART: a pooled analysis from the Kids-ART-LINC collaboration on over 2000 children on ART in sub-Saharan Africa reported a probability of LTFU of 5.0% and a mortality of 6.0% at one year,9
which is slightly higher than the mortality estimate in children remaining in care in the present study. The number of children included was smaller than in the present study and, more importantly, the analysis did not take mortality among children LTFU into account. The estimate from the Kids-ART-LINC analysis is therefore probably an underestimate. A large program from Lusaka, Zambia, with 2398 children on ART showed a mortality rate of 6.6 deaths per 100 child-years, with an estimated cumulative mortality at one year (from the Kaplan-Meier curve) of 7.9%. Children LTFU were excluded from this analysis.10
A pediatric ART treatment program in Haïti indicated that 9% of children had died and 10% were lost after a median follow-up time of 20 months.34
The definition of LTFU was uniform across sites, but the period of six months chosen for this definition may be too short for some programs and too long for others. In the absence of dedicated studies on the appropriate definition of LTFU, this remains speculative. Follow-up of children who were transferred out was censored, thus assuming that their prognosis is the same as that of comparable children remaining in care at the site. Some children who were transferred out might, however, have a better or worse prognosis than the children remaining in a program. The proportion of children transferred out was particularly high in tertiary care, and these children will have been transferred to a lower level of care.
We observed differences between tertiary and other care programs in baseline characteristics as well as a higher mortality in tertiary care sites. The two tertiary programs are located in a region in the Republic of South Africa with a large pediatric ART network and a well functioning referral system; hence patients who are younger and represent the severe end of the pediatric HIV disease spectrum predominate at these sites. Statistical adjustment for prognostic factors showed that if children treated in tertiary centers were similar to the children followed in the other programs, particularly with regard to age and CD4 percentage, their mortality would be similar to that observed in the other programs. Indeed, young age is an important prognostic factor as shown in a recent collaborative analysis of 33 European cohorts, which demonstrated that mortality was almost double in children aged less than 2 years, compared to older children.27
Children treated at two township programs in South Africa3
had lower mortality and this was not explained by more favorable prognostic factors at baseline. These programs mainly treat adult patients and the children followed there may represent a selected group of children with a good prognosis. Other sources of heterogeneity might include different patterns of transfer out, different parental outcomes related to children’s outcome, incorrect assumptions on mortality in children LTFU, and incomplete or absent adjustment for other prognostic variables, for example exposure to antiretroviral drugs for preventing mother-to-child transmission. Furthermore, where programs do not have the capacity to commence ART quickly, mortality on ART may be lower due to survival bias among children who lived long enough to access treatment. The (non-HIV-related) background mortality in the different populations may also have played a role.4
Treatment programs should systematically trace patients LTFU where not yet done and record outcomes in these patients. Where such mechanisms are already in place as in most of the sites included in this study, capacities should be increased to intensify these activities. A modeling study based on a program in Côte d’Ivoire found that interventions that enhance program retention would both improve survival and be cost-effective by international criteria.35
An additional benefit is that program-level mortality can then be estimated by using double-sampling designs to account for mortality among children LTFU.36;37
For example, data from an outreach program have been used to adjust mortality estimates from an adult treatment program in Kenya.38;39
An important assumption made by this method is that a random sample of patients LTFU was selected and that the vital status of everyone from that sample was ascertained. These assumptions are not generally met: patients to be traced tend to be selected according to their assumed place of residence, excluding patients who live far from the treatment site, and typically about a third of patients cannot be located.7
Our approach is useful for sensitivity analysis in programs where no dedicated tracing studies are available, and to test the robustness of estimates based on the results of tracing activities.
In conclusion, our study indicates that mortality of HIV-infected children starting ART in Southern Africa is substantially higher than the mortality directly observed in treatment programs. A better understanding of outcomes in children LTFU, as well as children transferred out of programs is required, and the determinants of LTFU and transfer-out need to be defined. Dedicated field studies tracing children LTFU and studies linking data from treatment programs with routine mortality data from national death registries are of high priority. It is important both from a clinical and programmatic point of view to ensure capacity to trace this vulnerable patient group as soon as possible after a missed clinic visit.