Effects of social protection on tuberculosis treatment outcomes in low or middle-income and in high-burden countries : systematic review and meta-analysis

Tuberculosis (TB) is a poverty infectious disease that affects millions of people worldwide. Evidences suggest that social protection strategies (SPS) can improve TB treatment outcomes. This study aimed to synthesize such evidences through systematic literature review and meta-analysis. We searched for studies conducted in lowor middle-income and in high TB-burden countries, published during 1995-2016. The review was performed by searching PubMed/MEDLINE, Scopus, Web of Science, ScienceDirect and LILACS. We included only studies that investigated the effects of SPS on TB treatment outcomes. We retained 25 studies for qualitative synthesis. Meta-analyses were performed with 9 randomized controlled trials, including a total of 1,687 participants. Pooled results showed that SPS was associated with TB treatment success (RR = 1.09; 95%CI: 1.03-1.14), cure of TB patients (RR = 1.11; 95%CI: 1.01-1.22) and with reduction in risk of TB treatment default (RR = 0.63; 95%CI: 0.45-0.89). We did not detect effects of SPS on the outcomes treatment failure and death. These findings revealed that SPS might improve TB treatment outcomes in lower-middle-income economies or countries with high burden of this disease. However, the overall quality of evidences regarding these effect estimates is low and further well-conducted randomized studies are needed.


Introduction
Tuberculosis (TB) is still considered a major global health problem, mainly in socially vulnerable population groups living in low-and middle-income countries 1 .In 2015, 10.4 million new cases and 1.4 million deaths because of TB were estimated worldwide 2 .According to the United Nations Sustainable Development Goals (2016-2030), the World Health Organization (WHO) set a 2030 target of 90% reduction in deaths, 80% decline in TB incidence and zero TB-affected families facing catastrophic costs because of this disease 3,4,5 .Poverty is one of the most important determinants of TB and accounts for almost one third of the global burden of diseases.Besides being a poverty related illness, TB also worsens this social condition.In low-income countries, approximately 17% of TB-deaths affect the economically productive age group of 15-49 years 6 .In those countries, investments in prevention, diagnostics and treatment should consider social protection and urban planning interventions in order to improve access to TB care and treatment adherence 7,8 .
Social protection strategies (SPS) enable individuals and households to protect and build their capital assets, leading them to move structurally out of poverty 7,8,9 .Their main components are social and income security, cash transfer programs, food provision, transport incentives, unemployment insurance, education, microcredit and income generation policies, as well as psychosocial support, protection against stigma/discrimination and public health systems with universal coverage, which are also included in the field of social protection 1,10,11 .
Universal health coverage and social protection account for reduction of income losses and costs incurred by TB patients, leading to positive impacts on the reduction of this disease burden, measured by incidence, prevalence and mortality rates 1,2 .For these reasons, bold policies and supportive systems are among the most important pillars of the post-2015 global tuberculosis strategy 3,5 .
In this context, the aim of this systematic review was to identify and assess evidences of social protection effects on TB treatment outcomes in low-and middle-income countries or in high TBburden countries.

Methods
Following the reporting guidelines recommended in PRISMA statement 12 , we conducted a systematic review to identify studies that have investigated the effects of SPS on TB treatment outcomes.The review protocol was registered in PROSPERO international database on September 17, 2015 (registration number CRD42015026305).

Eligibility criteria
Eligibility of studies followed predetermined inclusion criteria.We included interventional (randomized controlled trials -RCTs and studies with quasi-experimental design) and observational studies covering the period from 1995 to May 31 st , 2016.All included studies were developed in low-and middle-income countries, according to World Bank income classification 13 or in one of the 30 high TB-burden countries listed by WHO 2 .
We defined individuals under TB treatment as population of reviewed studies.The interventions of interest were SPS.Controls should be individuals under usual TB care.Primary outcome was TB treatment success rate (TSR), expressed by the proportion of all new cases that successfully completed treatment, with or without bacteriological evidences of cure 14 .
Secondary outcomes were: (a) cure rate, i.e. proportion of TB cases with a negative sputum smear result recorded during the last month of treatment and in at least one previous occasion during treatment; (b) default rate, i.e. proportion of TB cases that interrupted treatment for two consecutive months or more; (c) treatment failure rate, i.e. proportion of TB cases with smear positive results five months or later after initiating treatment; (d) death rate, i.e. the proportion of TB cases that died during treatment, irrespective of cause.All these outcomes should meet WHO international definitions 14 .

Search strategy
We searched the electronic databases PubMed/MEDLINE, Scopus, Web of Science, ScienceDirect and LILACS, without language restriction.The search terms were defined according to Medical Subject Headings (MeSH) and Descriptors in Health Sciences (DeCS).Search strategy applied the terms: social protection, public policy, social welfare, income, food assistance, food supply, transportation, reimbursement, financial support, government financing, social work, social security, public assistance, motivation.We used the operator "OR" to connect synonyms and the operator "AND" to combine them with the term tuberculosis.We applied filters to humans and publication dates.The searches occurred in October 2015 and were rerun in May 2016.
We hand searched in clinical trials registries (Clinical Trials.gov and Brazilian Clinical Trials Registry -ReBEC).Google Scholar was also screened in order to access grey literature (e.g.non-indexed journals, official documents, government reports), abstracts published in annals of scientific meetings, theses and dissertations digital libraries.The full search strategy is available in the supplemental material -cf.Appendix 1: https://www.4shared.com/web/preview/pdf/B-Gv6byoca.
One reviewer (K.V.F.A.) conducted the literature search based on the strategy developed by all the review team.Then, two reviewers (K.V.F.A., R.A.S.) independently examined titles and abstracts.Cohen's Kappa coefficient (k) was used as a measure of inter-rater agreement.Disagreements were resolved by consensus or by consulting a third reviewer (S.M.P.) to adjudicate.
We extracted and summarized data from the fully reviewed studies using a form to list study characteristics, including: author, publication year, study design and location, criteria for inclusion and exclusion of participants, randomization approach (if applicable), description of interventions, and TB treatment outcomes.

Quality assessment
We adapted Downs & Black 15 checklist to assess the methodological quality of non-randomized studies (NRS).The original version of this tool contains 27 items that assess internal (bias and confounding) and external validity (sample representativeness).Two reviewers (K.V.F.A., J.S.N.) performed the quality assessment applying 21 "yes"-or-"no" questions (using the scores 1 for yes and 0 for no) and another one with three answer options in order to check the description of confounding in each study, with these answer options: described (2 points), partially described (1 point) or not described (0 points).The total maximum score was 23.
The reviewers excluded five questions of the original version 15 for considering them inadequate or not applicable for our quality assessment.These questions addressed the following features: reporting of random variability estimates in the data for the main outcome, reporting of adverse events associated with study interventions, presence of unplanned analyses, adequacy of statistical tests and study power.Some of these excluded questions have limitations in their answer options.The scores profile provided a summary of methodological strengths and weakness of each study.However, we did not exclude studies based on it.
We performed a specific quality assessment for RCTs using The Cochrane Risk of Bias Tool 16 .In addition, quality of evidence across studies was evaluated with the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) 17 approach.GRADEpro Guideline Development Tool (GDT) 18 was used to visualize evidences profile and generate the summary of findings (SoF) table.

Data analysis
All studies retained for systematic review were described in qualitative synthesis.Only RCTs were submitted to meta-analysis.TB treatment outcomes were analyzed using risk ratios (RR) and their 95% confidence intervals (95%CI), recalculated from the data provided by RCTs, using the Mantel-Haenszel method.Meta-analyses were performed according to each outcome.Data from cluster-RCTs were adjusted based on the intracluster (or intraclass) correlation coefficient (ICC) provided by these studies 16 .
Cad. Saúde Pública 2018; 34(1):e00153116 To conduct these meta-analyses, a random effects model was fitted.This choice was based on the diversity of interventions and studies characteristics (e.g.participants and methods).The I 2 statistic was used to quantify the heterogeneity, categorized as not important (I 2 value between 0-40%), moderate (I 2 value between 30-60%), substantial (I 2 value between 50-90%) and considerable (I 2 value of 75% or more) 16 .Visual inspection of funnel plots contributed to assess the likely presence of publication bias.Meta-analyses were developed in the software Review Manager (RevMan; The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark) version 5.3.

Results
In total, we identified 8,825 articles.After removal of 1,758 duplicates, two reviewers (K.V.F.A., R.A.S.), working independently, screened 7,067 titles with k = 0.1907.Subsequently, 161 abstracts of the selected titles were independently screened by the same reviewers (with k = 0.4385), retaining 40 articles for full-text review.
In hand searches, the reviewers did not find eligible articles.One additional article was added from the reference lists examined in the full-text review.Finally, 25 articles were retained for qualitative synthesis and nine RCTs for meta-analysis (Figure 1).

Description of studies included in qualitative synthesis (n = 25)
Nineteen studies were conducted in middle-income countries (13 in upper-middle and six in lower-middle).Four studies were developed in three low-income countries: Ethiopia 19,20 , Malawi 21 and Nepal 22 .Two studies 23,24 were developed in Russia, which is not in low-and middle-income categories, but it is listed between the high TB-burden countries 2 .
Seventeen studies were performed in high TB-burden countries.The other studies (n = 8) occurred in countries out of this category, but referred to as low-income: Malawi 21 and Nepal 22 or middle-income countries: Ecuador 25 , Mexico 26 , Republic of Moldova 27 , Peru 28 , Swaziland 29 and Timor-Leste 30 .
Sixteen studies were prospective with primary data collection and nine studies used secondary data sources.Seventeen studies did not include TB vulnerable populations.Eleven studies only included individuals with pulmonary TB and eight did not restrict TB clinical forms.See Table 1 for details of this section.

Description of TB treatment outcomes • Treatment success rates (TSR)
Twenty-two controlled studies showed TSR for TB patients.In quasi-experimental studies (n = 10), TSR ranged from 69.3 to 96.9% in intervention groups and from 31.1 to 96.9% in controls.In RCTs (n = 9), TSR ranged from 70 to 97.7% in intervention groups and from 57.5 to 84.1% in controls.In observational studies (n = 3), TSR ranged from 68.7 to 95.8% in individuals exposed to social protection and from 46.9 to 92.9% in non-exposed individuals.The largest difference in TSR between study groups (47%) was found in a quasi-experimental study conducted in Saint Petersburg, Russian Federation 23 .
Few studies did not show effects of SPS on TB treatment outcomes.In a RCT conducted in Dili, Timor-Leste 30 food incentives did not significantly improve these outcomes.A study 40 that assessed effects of transportation incentives in China did not detect differences in TSR.Pragmatic RCT developed in South Africa 43 showed a small but non-significant improvement in TSR after providing economic support to TB patients.

• Cure rates (CR)
CR of TB patients were presented by 13 studies.In RCTs (n = 8) intervention groups achieved higher CR (ranging from 26.7 to 97.7%) than in controls (ranging from 10 to 81.4%).In quasi-experimental studies (n = 4) intervention groups also achieved higher CR (ranging from 11.8 to 82.4%) than in controls (ranging from 0 to 74.8%).Among observational studies, only one presented CR (equals to 82.1% in exposed and 76.9% in non-exposed individuals).

• Treatment default rates
Twenty studies presented default rates for TB treatment.Quasi-experimental studies (n = 10) showed proportions of default in intervention groups ranging from 0.2-21.8%and in controls from 0.1-68.9%.In RCTs (n = 7) proportions of default in individuals exposed to social protection ranged from

Table 1
Summary of all reviewed studies.indicates bad quality evidences and ≥ 9 indicates good quality evidences).

• Quality assessment results
The quality assessment with Downs & Black checklist 15 showed a median score equal to 9 (with interquartile range from 7-12).The lowest score was four and the highest was 14.Studies with the best quality evidences (n = 8) obtained scores greater than 9 (Table 1).The most important limitations of these studies were: lack of blinding of participants and study team, lack of information about follow-up lengths, and regarding the participants' compliance with study interventions.Lack of sample representativeness was also a common limitation in the studies.

• Risk of bias assessment for RCTs
Regarding randomization approach, all RCTs presented low risk of bias for random sequence generation and most of them (five studies) described an adequate allocation concealment.However, all RCTs were associated with high or unclear risk of bias due to lack of blinding of participants and study team.Outcome assessors were blinded in only one study.Incomplete outcome data and selective reporting did not account for relevant bias.Other sources of bias were associated with baseline imbalances, possibility of contamination across groups, low protocol fidelity and inadequate sample sizes (Figure 2).According to GRADE approach 17 , evidences from most TB treatment outcomes presented serious limitations because of study design and execution (risk of bias).Only the outcome cure was downgraded once by inconsistency, which was classified as "serious" because of a high unexplained heterogeneity observation across studies (Table 2).Indirectness of evidences was observed for all outcomes.We downgraded evidences in one level by indirectness for both primary and secondary outcomes because of the diversity of interventions comprehended in the field of social protection and specific characteristics of participants in some studies (restricted to individuals with multidrug-resistant TB, participants with TB and HIV-TB coinfection, and homeless patients).Finally, imprecision did not account for serious limitations in the reviewed RCTs (Table 2).
Risk of bias assessment and visual inspection of funnel plots did not suggest the presence of selective reporting (publication bias).In addition, under the GRADE approach 17 , the overall quality of evidences for effects of SPS on TB treatment outcomes was rated as low, except the evidences for cure (rated as very low quality).

Meta-analysis
The total number of participants in the RCTs included in meta-analysis (n = 9) was 1,687.These studies were conducted in seven countries, among them: South Africa 29,30,31 , Ethiopia, India, Malawi, Mexico, Nepal and Timor-Leste.We separately performed meta-analyses according to each TB treatment outcome.

• Secondary outcomes
In total, eight RCTs provided evidences for the cure of TB patients.A significant association between SPS and cure was found (RR = 1.11; 95%CI: 1.01-1.22).Additionally, substantial heterogeneity studies were not important (I 2 = 23%; p = 0.25) as observed in Figure 3b.

Discussion
This is the first systematic review focused on studies conducted in low and middle-income countries and/or with high TB-burden.We found an association between SPS and TB treatment success and a reduction in the risks of treatment default and therapeutic failure in patients under SPS.These evidences support the implementation of the social support along with a universal health coverage, especially in high-endemic TB or poor populations 1,8 .However, our findings should be interpreted with caution, considering the low quality of evidences provided by current studies.
Among the reviewed studies, only one RCT developed in South Africa 43 , showed that social protection did not improve the cure of TB patients.However, low fidelity to the trial protocol (leading to a third of eligible patients without intervention) and omission of data about HIV status of participants were important limitations to this study.Despite treatment success being considered a primary outcome, SPS showed a higher effect size against treatment default.This finding can be explained by the capacity that social protection has to increase health care access, leading to better treatment adherence through mechanisms to cope with financial hardship due to TB, alleviate poverty and reduce social vulnerability 1,44 .
Because of the disease, individuals and families with TB face direct and indirect costs that can be reduced by SPS.Direct costs include expenses with transport to and from the health facilities, medication, or consultations incurred by individuals.Indirect costs are associated with income losses that can account for almost 50% of total family expenses (e.g.illness-related work absences) 44,45 .
Social protection covering TB patients can provide means for these individuals to compensate catastrophic expenditures and reduce treatment default, especially for the poorest.They can be implemented in different ways, such as material incentives, cash transfers or food security programs 1 .Evidences from RCTs conducted in the United States show that incentives can contribute to patient attendance in TB health care facilities, improving treatment adherence 46 .
We noticed that SPS are not limited to cash transfers.In this meta-analysis, financial and non-financial interventions also were effective in protecting against default, leading to treatment success.The scope of social protection includes economic support, food security and nutrition 23,27,28,30,36,37 , psychological support and health education approaches 22,26 , social mobilization 35 and training of volunteers to act as patient's supporters 19,28,31,32,33,34 .
Despite these relevant results, there were some limitations to our meta-analyses.First, the limited number of available studies on social protection effects on TB treatment outcomes.Second, possible uncontrolled biases in the individual studies.Third, methodological limitations of reviewed studies and diversity of their interventions also might have affected our results.Future research based on well-designed randomized studies covering low and middle-income populations might provide better quality evidences.
In order to reduce study selection bias in our meta-analyses, we followed the eligibility criteria previously established in the review protocol.The main differences between this review and the protocol were: inclusion of observational studies in qualitative synthesis and an updated list of countries with high burden of TB.The first change aimed to strength our body of evidences about SPS and TB treatment outcomes.The second change was made to meet WHO current recommendations.In addition, we used Downs & Black 15 instrument to perform a quality assessment of all reviewed studies, since it can be applied not only to RCTs.Finally, we did not present all planned subgroup analyses, since heterogeneity did not substantially affect our results.Concerning publication bias, we agree that funnel plots (available in the supplemental material -cf.Appendix 2: https://www.4shared.com/web/preview/pdf/BVpsBkhoca?) do not ensure its absence but suggest its presence.Furthermore, we could not find unpublished studies and relevant materials in grey literature, strengthening the argument of absence of publication bias in the analyzed studies.

Conclusions
Our findings endorse the premises that social protection can contribute to TB treatment success, especially improving its adherence.The reach of goals for TB elimination after 2015 depends on the strengthening of social protection among the National Tuberculosis Programs (NTPs) priorities as a main action.
Evidences from our review suggest that low and middle-income countries should extend health coverage to reach the poorest individuals, associated with effective social support, through income transfer mechanisms and comprehensive interventions that may have beneficial impacts on TB outcomes.
Because of the broad scope of social protection, no conclusions can be drawn on the effect of specific interventions.We highlight the uncertainty in the effect estimates provided by the reviewed studies, given the overall low quality of the data.Further well-conducted randomized studies targeting low-and middle-income populations are needed.

Contributors
K. V. F. Andrade designed the study, searched on databases, selected the articles, performed the statistical analyses and wrote the manuscript.J. S. Nery contributed to quality assessment evidence and to the review of the manuscript.R. A. Souza searched on databases, selected the articles and reviewed the article.S. M. Pereira contributed to all stages of this study, helped in the interpretation of the study findings and reviewed the article.

Figure 1 PRISMA
Figure 1PRISMA flow diagram for systematic review and meta-analysis.

Figure 2 risk
Figure 2risk of bias in randomized controlled trials summary.

Figure 3 Forest
Figure 3Forest plots for comparison of social protection interventions versus tuberculosis usual care.

Table 1 (continued)
C: control group; CadÚnico: Unified Registry for Social Programmes; I: intervention group; NTP: National Tuberculosis Programme; SINAN: Notifiable Diseases information System; TCr: treatment completion rates; TSr: treatment success rates.* Quality scores with Downs & Black 15 tool, except for randomized controlled trials -RCTs (in this review, we arbitrarily established that scores < 9

Table 2
Summary of findings table.: a. most of information was provided by studies with some limitations (no information about allocation concealment, blinding, and with biases from other sources); b. differences in study populations and interventions might influence directness of evidences; c. high unexplained heterogeneity Note esta enfermedad.No obstante, la calidad general de las evidencias, en relación con estos resultados, es baja e indica que son necesarios más estudios controlados aleatorios bien realizados.