Interventions to improve adherence to treatment for paediatric tuberculosis in low- and middle-income countries: a systematic review and meta-analysis

Abstract Objective To assess the design, delivery and outcomes of interventions to improve adherence to treatment for paediatric tuberculosis in low- and middle-income countries and develop a contextual framework for such interventions. Methods We searched PubMed and Cochrane databases for reports published between 1 January 2003 and 1 December 2013 on interventions to improve adherence to treatment for tuberculosis that included patients younger than 20 years who lived in a low- or middle-income country. For potentially relevant articles that lacked paediatric outcomes, we contacted the authors of the studies. We assessed heterogeneity and risk of bias. To evaluate treatment success – i.e. the combination of treatment completion and cure – we performed random-effects meta-analysis. We identified areas of need for improved intervention practices. Findings We included 15 studies in 11 countries for the qualitative analysis and of these studies, 11 qualified for the meta-analysis – representing 1279 children. Of the interventions described in the 15 studies, two focused on education, one on psychosocial support, seven on care delivery, four on health systems and one on financial provisions. The children in intervention arms had higher rates of treatment success, compared with those in control groups (odds ratio: 3.02; 95% confidence interval: 2.19–4.15). Using the results of our analyses, we developed a framework around factors that promoted or threatened treatment completion. Conclusion Various interventions to improve adherence to treatment for paediatric tuberculosis appear both feasible and effective in low- and middle-income countries.


Introduction
Paediatric tuberculosis can be controlled or cured if timely and appropriate treatment is completed. 1,2 More than 75% of affected patients live in low-and middle-income countries in Asia and Africa and have substantial tuberculosis -related morbidity and mortality. 2 Up to 20% of children with tuberculosis in low-and middle-income countries fail to complete treatment. 3 Interrupted tuberculosis treatment poses a public health challenge because it permits the development of drug-resistant disease and allows patients to remain infectious for a relatively long time. Poor adherence results in disease progression, morbidity and death. The most extreme form of incomplete treatment is known as treatment abandonment or treatment default. For tuberculosis, such abandonment is generally represented by a break in treatment of at least two consecutive months. 1 The barriers to treatment completion in low-and middleincome countries include medical expenses, the indirect costs of transportation and time away from work, the stigmas associated with the illness and/or the treatment, communication breakdowns between providers and patients, limited health literacy, the presence of too few health workers and problems in drug procurement. 2 We conducted a systematic review and meta-analysis of interventions designed to reduce such barriers to treatment completion among children with tuberculosis in low-and middle-income countries. Our main aim was to appraise the design, delivery and impact of such interventions in such a vulnerable population.

Search and selection
Using a registered protocol (PROSPERO: CRD42013005800), we searched the PubMed and Cochrane databases for relevant publications that had been published between 1 January 2003 and 1 December 2013. Grey literature was hand-searched. Until 1 May 2014, we attempted to contact the authors of relevant articles and other researchers with experience of tuberculosis in low-and middle-income countries. The search strategy (Box 1; available at: http://www.who.int/bulletin/volumes/93/10/14-147231) was piloted by two researchers and reviewed by two medical librarians.
To be included in our analyses, a study had to have participants with active tuberculosis who were younger than 20 years and lived in a country that, according to the World Bank, was low-income or middle-income in December 2013. Studies with adult participants were included only if the cohort outcomes for participants younger than 20 years were available. We were only interested in studies on interventions targeted at the improvement of treatment initiation or completion, the improvement of adherence to medications or appointments, the prevention of treatment refusal or adherence surrogates such as self-efficacy or enablement.
Objective To assess the design, delivery and outcomes of interventions to improve adherence to treatment for paediatric tuberculosis in low-and middle-income countries and develop a contextual framework for such interventions. Methods We searched PubMed and Cochrane databases for reports published between 1 January 2003 and 1 December 2013 on interventions to improve adherence to treatment for tuberculosis that included patients younger than 20 years who lived in a low-or middleincome country. For potentially relevant articles that lacked paediatric outcomes, we contacted the authors of the studies. We assessed heterogeneity and risk of bias. To evaluate treatment success -i.e. the combination of treatment completion and cure -we performed random-effects meta-analysis. We identified areas of need for improved intervention practices. Findings We included 15 studies in 11 countries for the qualitative analysis and of these studies, 11 qualified for the meta-analysisrepresenting 1279 children. Of the interventions described in the 15 studies, two focused on education, one on psychosocial support, seven on care delivery, four on health systems and one on financial provisions. The children in intervention arms had higher rates of treatment success, compared with those in control groups (odds ratio: 3.02; 95% confidence interval: 2. 19-4.15). Using the results of our analyses, we developed a framework around factors that promoted or threatened treatment completion. Conclusion Various interventions to improve adherence to treatment for paediatric tuberculosis appear both feasible and effective in low-and middle-income countries.
Included studies required a control or comparison population. Retrospective or contemporaneous comparisons from the same region were accepted if the between-population similarities and differences were clearly stated. No language, follow-up or study quality restrictions were imposed.

Data extraction
By using standardized forms, two investigators independently screened abstracts and extracted data. Discrepancies between the two investigators were resolved through discussion (16 records) or by the seeking of clarification from an author of an article of potential interest (three records).
We detected 62 studies that met all of our eligibility criteria apart from the provision of explicit outcomes for paediatric patients. Although we attempted to determine such outcomes by contacting the authors of the corresponding study reports, we successfully obtained outcomes for just 10 additional studies. The other 52 reports provided no current contact information for any author (14 studies), had authors who did not reply to our queries (20 studies) or had authors who stated that the data we wanted were not available (18 studies).
From each eligible report, we extracted information on methods, interventions, outcomes, participants, settings and co-infection with human immunodeficiency virus (HIV). Treatment outcomes were extracted according to the World Health Organization's (WHO's) classifications, with treatment success defined as completion or cure 1 -as given in the reports.
Risk of bias in the randomized trials was assessed using the Cochrane Assessment tool 4 and reported according to CONSORT standards. 5 Quality of the non-randomized trials was assessed using the Effective Public Health Practice Project Quality Assessment tool 6 and reported according to TREND standards. 5,7 Funding source was recorded as a possible bias source. Studies that integrated qualitative data were assessed using the relevant tools of the Critical Appraisal Skills programme. 8 Reporting of the systematic review adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement. 9 Interventions to improve treatment adherence among paediatric patients of tuberculosis were summarized through independent iterative re-reading and organization of the identified themes -with discussion to achieve consensus -in alignment with WHO's adherence dimensions for long-term therapies. 2 For the initial data extraction, interventions were divided into five categories: education, psychosocial, care delivery, health systems and social protection or financial (Table 1). We attempted to determine those factors that promoted or threatened treatment completion. These factors might be related to: (i) the patient -e.g. literacy, (ii) the condition, including the presence of comorbidities, (iii) the therapy, including cultural lay beliefs, (iv) the health system, including accessibility, and (v) socioeconomic status, including family income.

Statistical analysis
We did a meta-analysis of the treatment success rates recorded among paediatric patients. We used the Mantel-Haenszel model and the DerSimonian and Laird random-effects method to calculate odds ratios (ORs) and their 95% confidence intervals (CIs) from the unadjusted raw data, with the assumption that intervention effects on treatment success in one setting might differ from those in other settings. We did sensitivity analyses that included only randomized or quasi-randomized studies or excluded studies with comparison population estimates derived from another setting (available from the corresponding author). Heterogeneity across studies was assessed using the I 2 statistic. We summarized the main meta-analysis results as a forest plot but used funnel plots to assess publication bias. Analyses were conducted using Review Manager version 5.2 (Cochrane Collaboration, Copenhagen, Denmark).

Results
We initially identified 413 articles of potential interest. Of these, 164 qualified for full-text review and we included 15 articles in our qualitative synthesis (Fig. 1). [10][11][12][13][14][15][16][17][18][19][20][21][22][23][24] The articles were on 15 separate studies ( Table 2). Three of the studies were published in Portuguese 11,13,14 and the remainder in English. Five studies were based in the upper-middle-income countries of Brazil 11,13,14 and Thailand, 15,16 three in the lower-middle-income countries of India, 12 Lesotho 10 and Pakistan, 17 and seven in the low-income countries of Bangladesh, 23 Ethiopia, 18,19 Kenya, 22 Myanmar, 24 South Sudan 21 and the United Republic of Tanzania. 20 Four settings were urban outpatient, 11,13,16,17 three rural outpatient, 12,18,19 two subur-  21 The remaining studies were done in variable settings. 10,14,15,20,22 The payment system for health services was not described in nine studies 11,12,[14][15][16][17][18][19][20] but the reports on four studies described capped fees 24 or clinic fee coverage. 16,23,24 In seven studies, drug expenses were covered for one intervention group only, 12 for both the intervention and comparison groups, as part of a national scheme, 16,[22][23][24] or for at least the intervention group -with unclear indication if the drug expenses of the comparison group were also covered. 10,21 The included studies were conducted between 1996 and 2011 and reported -including the unpublished data supplied by authors -between 2003 and 2014. The median duration of the investigated interventions was 24 months (range: 9-96). The number of participants younger than 20 yearswhich had to be clarified through author contact for six studies and excluded population-based comparison samples -varied from four to 308 (mean: 106; median: 61) and totalled 1587 across all 15 studies. Such paediatric patients represented between 3% and 100% of the patients investigated (mean: 22%; median: 11%). The prevalence of HIV co-infection, which was only reported for six studies, ranged from less than 5% to 74%. 10,11,13,15,20,21

Systematic reviews
Improving treatment adherence in paediatric tuberculosis Meaghann S Weaver The size of the paediatric sample has not been published previously and had to be obtained by direct contact with an author of the relevant article.

Risk of bias
The benefits of the investigated interventions may be overestimated because of short follow-up and failure to assess adherence after the interventions were discontinued. Confounders, such as the extra attention given to participants during educational interventions, 16,23 complicate our analyses. Although one study report details how controls -who did not receive the educational intervention -were supervised by health volunteers, 16 it failed to give any idea of the corresponding contact time. The concurrent use of several interventions makes it hard to determine the main reason for successful outcomes. Social feedback loops -in which successful interventions foster a dynamic for more community adherence -were subjectively recognized by several research teams. 16,18,19,21,24 Intervention complexity increased as attention expanded beyond the patient to include the provider, 23 the family, [13][14][15] both the provider and family [10][11][12]16,17,20 or the provider, family and community. 18,19,21,22,24 Complexity was characterized by contextual interactions that were susceptible to policy timing, 13,18,20,21,24 staffing capabilities and attitudes, 12,16,17,19,22,23 relationships 13,16,19,23 and resources. 18,19,23,24 No empiric quality measures of implementation fidelity were described. Two studies incorporated qualitative data from focus groups and in-depth interviews. 19,22 Although context, sampling and data collection were outlined and the findings appeared supported by data, there was no discussion of reflexivity and no detailed description of the analyses. None of the studies we investigated incorporated long-term observational or ethnographic approaches.

Systematic reviews
Improving treatment adherence in paediatric tuberculosis Meaghann S Weaver of the results reported due to limited follow-up data, which had impaired the assessment of cure 21 or treatment outcome beyond referrals. 17 Funding sources included nongovernmental organizations, 10,11,20-24 health departments 18 or international 15,17,19 or local 16 academic institutes or were not specified. 11 Table 4 and Table 5 show the results on study-specific biases (available at: http://www.who.int/bulletin/vol/umes/93/10/14-147231).

Meta-analysis
Treatment success rates for the paediatric participants in both the treatment and comparison groups were reported for 11 studies. [10][11][12][14][15][16][18][19][20]22,23 These studies were included in the meta-analysis and together represented 1279 children -excluding those in any external comparison groups. In three of the four studies excluded from the meta-analysis, the interventions investigated appeared to bring improved rates of treatment success, for all age groups. 13,21,24 The results of the other excluded study 17 indicated that the intervention led to increased referral rates.
Meta-analysis revealed a threefold improvement in odds of treatment success for children receiving the interventions ( Fig. 2; OR: 3.02; 95% CI: 2.19-4.15). There was no evidence of statistical heterogeneity (I 2 : 0%). A funnel plot showed symmetry for the large, high-powered studies but potential publication bias for the smaller studies ( Fig. 3; available at: http://www.who. int/bulletin/volumes/93/09/14-147231). Sensitivity analysis did not modify the overall results (available from the corresponding author). Baseline risk factors reported for poor adherence outcomes are outlined in Box 2.

Discussion
In our review of interventions to promote paediatric tuberculosis treatment adherence in low-and middle-income countries, we found evidence that such interventions can result in clinically important improvements in tuberculosis treatment success. Diverse interventions addressing education, psychosocial support, care delivery, health system strengthening and social protection are reportedly feasible and effective in facilitating treatment completion.
Several studies followed collaborative strategies. For example, there was evidence of social franchise programmes communicating with the media, tuberculosis villages communicating with local leaders, tuberculosis clubs communicating with neighbours, health centres communicating with referral facilities and health providers engaging in motivational communication with patients.
We used systematic methods to identify and analyse a broad range of studies, without language limitations and with solicitation of input from the authors of relevant articles in an attempt to minimize search bias. We provided detailed descriptions and syntheses of interventions -which were often multicomponent and complex -that had been implemented among children in low-and middle-income countries. Our summary findings may help guide future intervention planning and evaluation. Our reviews did, however, have several limitations. For example, few studies included specific details on the nature of their paediatric programme, and no data on individual patients were available. Given the generally small sample sizes, the reported confidence intervals for the effects of individual interventions were often broad. Despite this, all but one of the 11 studies included in the meta-analysis had odds ratios that indicated that the investigated intervention improved the rate of treatment success, and the four largest of these studies provided unequivocal evidence of such benefit.
Heterogeneity in the context and measurement of adherence, outcome definition and reporting limit the value of between-study comparisons. In highincome countries, multi-component interventions are common and often found to be superior to single-component interventions. 26 Several of the relevant studies included in our reviews also attempted to target several adherence

Social and/or economic related
• Low-socioeconomic level 24 Health system related • Distance from care source 12 Improving treatment adherence in paediatric tuberculosis Meaghann S Weaver factors simultaneously, by using complex interventions. Such complex interventions make it difficult to attribute the results to particular intervention categories or components. One of the studies we reviewed was of an intervention that included education, improved dosing and appointment convenience, patient tracing, reduction of out-of-pocket costs and a deposit that was refunded on treatment completion. 12 It may be that only when implemented together do these elements succeed.
Recognizing the interconnected nature of WHO's five adherence dimensions and intervention categories for long-term therapies, 2 we have summarized contextual factors affecting the adherence interventions we investigated in a framework (Fig. 4). The themes highlighted in this figure are intended to be illustrative across dimensions and intervention categories. For instance, factors that may adversely affect tuberculosis treatment adherence that span psychosocial and educational categories -e.g. low literacy and limited self-efficacy -are shown in the figure alongside adherence-promoting factors such as family education and patient empowerment. The contextual framework may aid further collaborative studies and analyses of adherence-targeted interventions.
Through qualitative analysis, we identified three areas where studies described -or failed to describe -children's unique features that can affect adherence intervention delivery. First, few studies described paediatricspecific disease epidemiology and use of paediatric-inclusive outcomes. Several authors reported an unexpectedly high prevalence of paediatric tuberculosis that warranted management as a public health problem. 14,18,21 However, most of the studies that we screened simply excluded children and 54 studies that would otherwise have been eligible for our analyses had to be excluded because they failed to report paediatric outcomes separately. Even for the eligible studies, adherence outcomes were not explicitly adapted for paediatric patients -although paediatric-specific treatment toxicity was recognized in one study. 10 Second, several reports noted challenges in paediatric tuberculosis diagnosis and care. Children can pose diagnostic dilemmas that complicate epidemiological and outcome estimates. 10,21 One study noted that paediatric lymph-node biopsies could not be safely performed locally. 21 Another considered how children's difficulty with sputum production may contribute to low detection rates 18 while a different study specified distinct sputum collec-tion techniques for younger children. 10 Dosing instructions that were adapted for paediatric treatment were also recommended. 10 Key comorbidities in children -e.g. malnutrition 21 -may benefit from dedicated attention.
Third, several studies acknowledged the need to consider the preferences and social role of children and adolescents, who may need tailored interventions. In one study involving the use of directly observed, short-term treatment, children and women were more likely than men to select community-based over facility-based treatment, when given the option. 20 Another study adapted an intervention, for use among children, according to household and social needs. This intervention included supporting the children in returning to school. 10 As one study commented, tuberculosisand tuberculosis treatment -can cut the economic productivity of adolescents and young adults, who tend to have relatively high burdens of the disease. 12 Based on our review and identified themes, future studies need to: (i) assess interventions in low-and middle-income countries that explicitly analyse paediatric-inclusive and paediatric-distinct needs and outcomes, (ii) use mixed-method approaches that can assess the pathways linking context-dependent factors with outcomes, (iii) use longitudinal evaluations that investigate the sustainability of the effectiveness and benefits of interventions and the potential burdens posed by interventions, and (iv) incorporate and address cost-effectiveness, resource implications and potential scalability.

Resumen
Intervenciones para mejorar el cumplimiento del tratamiento de la tuberculosis pediátrica en los países de ingresos bajos y medios: una revisión sistemática y un metanálisis Objetivo Evaluar el diseño, la prestación y los resultados de las intervenciones para mejorar el cumplimiento del tratamiento de la tuberculosis pediátrica en países de ingresos bajos y medios y desarrollar un marco contextual para tales intervenciones. Métodos Se realizaron búsquedas en las bases de datos PubMed y Cochrane para encontrar informes sobre las intervenciones para mejorar el cumplimiento del tratamiento de la tuberculosis publicados entre el 1 de enero de 2003 y el 1 de diciembre de 2013 que incluyeran pacientes menores de 20 años que vivieran en países de ingresos bajos o medios. Se contactó con los autores de los estudios con artículos relevantes que carecían de resultados pediátricos. Se evaluó la heterogeneidad y el riesgo de sesgo. Se llevaron a cabo metanálisis de efectos aleatorios para evaluar el éxito del tratamiento, es decir, la combinación de finalización del tratamiento y cura. Se identificaron áreas que necesitaban una mejora de las prácticas de intervención.

Box 1. Search strategy to identify studies on interventions to improve adherence to treatment for paediatric tuberculosis in low-and middle-income countries
("low income economies" OR "lower middle income economies" OR "middle income economies" OR "developing countries"