Background
In 2007, 90% of the estimated 2.0 million HIV-infected children worldwide lived in sub-Saharan Africa, a reflection of the HIV epidemic in adults of the same region [
1]. Opportunistic infections, especially tuberculosis (TB), are a major cause of morbidity and mortality amongst these children [
2,
3]. Internationally, significant progress has been made to minimize morbidity and mortality amongst people living with HIV/AIDS. The benefits of highly active antiretroviral therapy (HAART), trimethoprim-sulphamethoxazole (TMP-SMX) prophylaxis and, more recently, isoniazid (INH) prophylaxis for preventing TB, have been proven in adults and children in various settings [
4‐
8]. However, in resource-limited settings these benefits are rarely realized. Obstacles include cost, interrupted drug supplies and suboptimal adherence [
1,
9,
10].
To enable all those infected by HIV to benefit from these life-saving interventions, it is imperative that cost-effectiveness and barriers to adherence be urgently and innovatively addressed. Common barriers to adherence include complex dosing schedules, toxicity, pill burden and, in many resource-limited settings, financial cost to the patient [
11,
12]. Intermittent dosing schedules can potentially address these adherence barriers and may prove cost-saving. Intermittent dosing for the treatment and prevention of TB has been successfully used for adults and efficacy has been proven in children [
8,
13‐
15]. Even if the efficacy of a treatment is similar for different dosing schedules, it is possible that subtle adherence differences may impact on the feasibility and effectiveness when the treatment is rolled out on a large scale. To our knowledge, there have been no randomized controlled trials evaluating the impact of dosing schedule on adherence to INH prophylaxis in HIV-infected children.
In a recent placebo-controlled trial evaluating INH prophylaxis for HIV-infected children in Cape Town, South Africa, INH markedly reduced mortality and TB incidence with no difference in the efficacy between a daily and an intermittent dosing schedule [
8]. Using the same study cohort, we describe the overall adherence rates as measured by pill counts and caregiver self-reports. We investigate the potential differences in adherence between the two dosing schedules and assess the predictors of adherence.
Discussion
Daily INH prophylaxis showed no adherence benefit over the three times per week regimen in our study. As reported previously, the efficacy of the two dosing schedules was equivalent, with both groups experiencing a marked reduction in mortality and TB incidence when on INH as compared to placebo [
8]. Furthermore, no toxicity difference was found between the two groups (unpublished data). In our cohort, three times a week INH prophylaxis was therefore not only highly efficacious in reducing mortality and TB in HIV-infected children, but also as easily adhered to and tolerated as daily INH. The use of isoniazid as TB prophylaxis in HIV-infected adults, after the exclusion of active TB, is recommended as part of the World Health Organization's Stop TB strategy [
23]. INH prophylaxis is also recommended for children younger than 5 years with proven or suspected latent TB, irrespective of HIV status, although experts advise that HIV-infected children of all ages would benefit [
24,
25]. Yet the implementation of these measures has been slow, limited by concerns regarding the emergence of TB resistance as well as by economic and health system constraints [
1]. For HIV-infected children in whom active TB has been excluded, an intermittent dosing schedule of INH prophylaxis would be cheaper and potentially easier to administer than daily INH. This could, in part, help to address the latter concerns.
Overall adherence to INH among our cohort of children was excellent, with the majority achieving a mean adherence of ≥ 90%. This is in keeping with population HAART adherence rates found in HIV-infected adults and children from similar resource-limited settings [
26‐
28]. There is no gold standard for adherence [
29]. As with adherence to HAART, comparing estimates of adherence to TB prophylaxis between studies is complicated by varying definitions of adherence and the use of different adherence tools [
11,
12]. Traditionally, completion rates (based on taking > 80% of the prescribed doses) have been used to describe adherence to TB prophylaxis [
16]. In our study, 88% of children prescribed INH completed a minimum of 6 months of prophylaxis. Only 47.1% of South African HIV-infected adults completed 6 months INH prophylaxis in an operational setting [
30], compared to 76% of adults in an Ugandan study [
31]. In seminal pre-HIV studies of INH prophylaxis, approximately 70% of Alaskan children were considered to be adherent to study medication based on completion rates [
32]. In one prospective and one retrospective study evaluating adherence to INH prophylaxis in operational settings in Cape Town, South Africa, only 15% and 27% respectively of the children completed 5-6 months of INH [
33,
34]. The majority of these children had an unknown HIV status or were HIV negative. Medication completion rates allow for easy documentation of poor adherence to short-term therapy and are important for programmatic evaluation. However, its use in clinical settings is limited as there is no scope for timely adherence interventions. Our cohort of children and caregivers were provided with substantial adherence support. In particular, the adherence measures we used allowed for immediate feedback and counselling. This allowed the care providers to focus attention on those children and caregivers who were struggling and may, at least in part, explain our high completion rates.
Pill counts may overestimate adherence [
17,
19]. Other studies of INH have reported pill count-based adherence estimates similar to ours. An Australian study of INH prophylaxis in adolescents recorded a pill count-based adherence estimate of 91% compared to 79% by urine testing, 83% by clinic attendance and 66% by medication event monitoring system [
35]. A mean adherence of 85% was reported in a placebo-controlled trial of INH prophylaxis for HIV-infected adults in a setting similar to ours, based on pill counts. These adherence measures correlated well with the reports of patient nominated supervisors [
36]. Overestimation of adherence by pill counts is based on falsely assuming that unreturned pills were ingested by the patient. Pill dumping, where containers are returned empty without any medication being taken, is an extreme example, but patients may also discard only a few of the remaining tablets prior to returning the container. In young children, medication doses often have to be repeated by the caregivers due to vomiting, spitting out of medicine or spillage. When more medication is provided than needs to be taken, this can potentially lead to adherence measures of > 100%, even if pill dumping did not occur [
37]. In a Ugandan study of paediatric and adult adherence to HAART, unannounced home-based pill counts provided more conservative estimates of adherence than in-clinic announced pill counts (72% of home-based counts achieved adherence > 95% compared to 94% of in-clinic counts) [
18]. As our adherence estimates were based on announced pill counts and include some measures above 100%, there was a possible overestimation of adherence. However, the comparison between the prophylaxis groups remains valid. Furthermore, there was a high correlation of pill counts and caregiver self-report, indicating good adherence by multiple measures. A feasibility study of INH prophylaxis in HIV-infected Ugandan adults similarly found high adherence rates by both pill counts (82%) and self-report (85%). In the same cohort, high adherence estimates were also reported by clinic attendance (81%) and urine testing (79%) [
31].
In a recent systematic review of paediatric HAART adherence in low- and middle-income countries, self- or caregiver (proxy)-reports were the most frequently used measures of adherence, providing high estimates of adherence (79.5-100%) [
11]. Although recall and social desirability bias can result in inflated estimates of adherence, self-report has shown moderate correlation with virological outcomes of patients on HAART [
19]. Self-report adherence measures were also valid and reliable in a study on treatment of latent TB among North-American Latino adolescents [
38]. Self-report measures are easy to obtain, allow for discussion of obstacles to adherence, are inexpensive and not as labour intensive as some other adherence measures. In our study, unlike most clinical settings, dedicated study pharmacists were responsible for assessing adherence by pill count, whilst study doctors/nurses were responsible for assessing self-report. Therefore, contrary to what would be expected in an operational setting, our clinical staff preferred pill counts to self-reports. Furthermore, caregiver self-reports were often difficult to obtain as the children arrived for study visits with a variety of caregivers, many of whom could not provide information on how medication had been taken in the previous week. Another practical obstacle to measuring adherence encountered by our staff, which may be unique to our setting, was the loss of medicine and diaries in shack fires (fires in informal settlements, usually originating in dwellings where paraffin stoves or open fires are the only sources of heat and energy).
There are three major limitations to our study. First, our main analysis is based on an adherence measure which was only initiated after some children had already been lost to follow up. Attrition is not unexpected in clinical trials and yet it represents an extreme form of non-adherence. Thus, it is possible that we overestimated the overall adherence, as initial defaulters were not accounted for in the analysis. However, as attrition rates were similar in the two prophylaxis groups, this should not influence the comparison between daily and intermittent dosing. Secondly, our estimates are based on measures known to overestimate adherence. However, the measures had a relatively high degree of concordance and our findings are in keeping with the reports on adherence to HAART in HIV-infected children in similar settings. Lastly, we report adherence as measured in a clinical trial setting. Adherence in clinical trial settings is necessarily higher than in operational settings. As clinical trials require informed consent and often preliminary visits before study enrolment, individuals most likely to be non-adherent may well choose to not participate in the trial. Hence, adherence among trial participants is not necessarily predictive of the likely adherence to a similar intervention in the general population. Furthermore, overburdened and unsupported clinic staff members are less likely, and in many cases simply unable, to provide vigilant adherence support such as written or telephonic reminders, meals and payment of transport costs. Such support, however, is integral to a prospective study protocol. Patients are also more likely to be subjected to long waiting times, interrupted drug supplies and worse interpersonal experiences with care providers in resource poor health care centres outside of a study setting [
39]. Yet many of these adherence barriers are being addressed and successfully overcome in the context of HIV and HAART [
1,
26]. Our study suggests that, with adequate adherence support, high levels of adherence to INH prophylaxis can be achieved among selected HIV-infected children despite poor socio-economic circumstances and a prolonged period of prophylaxis.
Medication adherence among HIV-infected children has been extensively studied in the context of HAART. Variable predictors of adherence have been reported in the literature, some seemingly inconsistent [
40]. Although our study was not primarily designed to evaluate adherence predictors, we found two associations that might serve to assist health-care workers in identifying children at higher risk for non-adherence. Although adherence in young children depends on the care-giver, age has often been cited as an adherence predictor, with some studies finding younger children to be more adherent and others reporting better adherence in older children [
15,
41‐
43]. Many of these studies evaluate age as a continuous measure, without differentiating between the different stages of childhood development. In particular, most studies include adolescents, who frequently struggle with adherence in ways that are different to younger children [
40]. Our study included mostly young children, with very few adolescents. Differentiating between infants, toddlers and pre-school children (> 4 years), we found a significantly better adherence among the pre-school children. This may partly be due to the fact that our study drug (INH/placebo) was administered in tablet form. Toddlers, and especially infants, might easily reject not only the taste but also the texture of crushed tablets. Furthermore, it is perhaps easier to negotiate medication adherence in a verbal child than in an infant or toddler. Unfortunately, child-friendly medication in palatable syrup form remains unavailable in many resource-limited settings. The provision of guidance to parents on sources of palatable, safe substances to disguise texture and flavour of life-saving medication should be particularly focused on those with infants and toddlers.
In contrast to a recent report on adherence to HAART among children from a similar South African setting, we did not find any association between adherence and socio-economic factors such as access to water, electricity or a flush toilet [
28]. We did, however, find a strong association with our indicator for crowding (that is, the number of people per house). A South African study evaluating intermittent versus daily chemotherapy for the treatment of childhood TB similarly found a significant association between household crowding and adherence, with children from crowded homes achieving a poorer adherence [
15]. As children depend largely on caregivers for the administration of treatment, caregiver characteristics are important determinants of adherence [
40,
42]. Studies from North America have described improved adherence where primary caregivers were not the biological parents, possibly relating to parents' emotions regarding their own HIV status [
40,
44]. However, caregiver characteristics and their impact on adherence are complex and it is not possible to generalise between different cultural settings. In contrast to reports from industrialized settings, adherence was better among Togolese children whose primary caregivers were their biological parents [
45]. A study from South Africa noted misunderstandings between multiple caregivers to be a commonly cited reason for non-adherence to HAART [
27]. The association we found between crowding and non-adherence might relate to unmeasured caregiver characteristics, such as having multiple caregivers, rather than to general socio-economic circumstances. Certainly, the complex role of caregiver characteristics in HIV-related paediatric adherence must be explored and refined in the African context. Household size is easily determined and could be a useful tool in assisting health care workers in similar settings to identify children who may be at a higher risk for non-adherence.
Acknowledgements
We thank the children and their caregivers for participating. We are grateful to others who worked on the study: P Apolles, H Bezuidenhout, N Dlaku, E Dobbels, T Fakir, C Ford, D Gray, M George, L Holt, G Hussey, T Jennings, A Joachim, J Karpakis, B Leibbrandt, A Loggie, CJ Lombard, G Lottering, M Louw, I Mong, P Mtiya, D Nchuna, F Ngcokovana, V Nkondlala, K Orpen, H Rabie, HS Schaaf, H Smit, R Streicher, E Swanepoel, E Walters and M A Wolff. The data and safety monitoring board comprised J Kaplan (chair), W El Sadr, P Donald and N Beyers.
Sources of support
The study was funded by Rockefeller Foundation, USA and by MRC, South Africa.
SM le Roux and DM le Roux were supported in part by a fellowship from Fogarty International Center/USNIH (2D 43 TW000010-20-AITRP). JE Golub is supported by National Institutes of Health grant K01-AI066994.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
SLR was the trial physician for the study, participated in data collection, was responsible for statistical analysis and drafted the manuscript. MFC and HJZ conceived the study, wrote the protocol and grant application and supervised the study. JG supervised the statistical analysis. DLR participated in the statistical analysis and the literature review. LW was responsible for the management of the database. All authors read, approved and contributed to the final manuscript.