The impact of conditional cash transfers on health outcomes and use of health services in low and middle income countries

  • Review
  • Intervention

Authors


Abstract

Background

Conditional cash transfers (CCT) provide monetary transfers to households on the condition that they comply with some pre-defined requirements. CCT programmes have been justified on the grounds that demand-side subsidies are necessary to address inequities in access to health and social services for poor people. In the past decade they have become increasingly popular, particularly in middle income countries in Latin America.

Objectives

To assess the effectiveness of CCT in improving access to care and health outcomes, in particular for poorer populations in low and middle income countries.

Search methods

We searched a wide range of international databases, including the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE and EMBASE, in addition to development studies and economic databases. We also searched the websites and online resources of numerous international agencies, organisations and universities to find relevant grey literature. The original searches were conducted between November 2005 and April 2006. An updated search in MEDLINE was carried out in May 2009.

Selection criteria

CCT were defined as monetary transfers made to households on the condition that they comply with some pre-determined requirements in relation to health care. Studies had to include an objective measure of at least one of the following outcomes: health care utilisation, health expenditure, health outcomes or equity outcomes. Eligible study designs were: randomised controlled trial, interrupted time series analysis, or controlled before-after study of the impact of health financing policies following criteria used by the Cochrane Effective Practice and Organisation of Care Group.

Data collection and analysis

We performed qualitative analysis of the evidence.

Main results

We included ten papers reporting results from six intervention studies. Overall, design quality and analysis limited the risks of bias. Several CCT programmes provided strong evidence of a positive impact on the use of health services, nutritional status and health outcomes, respectively assessed by anthropometric measurements  and self-reported episodes of illness. It is hard to attribute these positive effects to the cash incentives specifically because other components may also contribute. Several studies provide evidence of positive impacts on the uptake of preventive services by children and pregnant women. We found no evidence about effects on health care expenditure.

Authors' conclusions

Conditional cash transfer programmes have been the subject of some well-designed evaluations, which strongly suggest that they could be an effective approach to improving access to preventive services. Their replicability under different conditions - particularly in more deprived settings - is still unclear because they depend on effective primary health care and mechanisms to disburse payments. Further rigorous evaluative research is needed, particularly where CCTs are being introduced in low income countries, for example in Sub-Saharan Africa or South Asia.

摘要

背景

在中低收入國家中有條件的現金救助對於健康結果及健康照護使用的影響

有條件的現金救助(Conditional cash transfers (CCT))是提供金錢救助予家庭,條件是他們符合一些預訂的標準。有條件的現金救助計畫是合理的,因為需要需求面的補貼以解決貧民在接近健康與社會服務的不平等現象。在過去幾十年已變得愈來愈受歡迎,特別是在拉丁美洲的中等收入國家中。

目標

評估有條件的現金救助對於增進照護及健康結果的效果,尤其是中低收入國家中的貧民。

搜尋策略

我們檢索廣泛的全國資料庫,包括the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE與EMBASE,除了進行中的研究及經濟學資料庫。我們也檢索網站及許多全國的機構,組織與大學的線上資源以尋找相關的文獻。開始進行檢索是在2005年11月至2006年4月。MEDLINE的檢索更新是在2009年5月完成。

選擇標準

有條件的現金救助被定義為金錢轉移予家庭,條件是他們符合有關健康照護的預訂標準。研究必須至少包括下列其中之一的測量結果:健康照護利用,健康照護花費,健康結果或公平的結果。合格的研究設計為:隨機對照試驗,間段時間序列分析法,或前後對照研究有關健康金融政策的影響,其遵循Cochrane Effective Practice與Organisation of Care Group的標準。

資料收集與分析

我們採用證據的質性分析。

主要結論

我們蒐集了10篇文章其報告結果來自6項介入措施研究。大致上,設計的品質與分析限制了偏差風險。數篇有條件的現金救助計畫提供有關健康服務,營養狀況及健康結果正向影響的有利證據,其個別評估人體測量結果與自我報告疾病事件。很難將這些正向效果歸因於明確的現金動機,因為其它的因素也可能有關。數篇研究提供有關兒童及懷孕婦女其使用預防性服務正向影響的證據。我們沒有找到關於健康照護花費效果的證據。

作者結論

有條件的現金救助計畫為一些妥善設計評估的主題,其強烈的認為它們可以是一種有效的方式以增進預防性服務的可近。他們的可複製性在不同條件下特別是在較差的環境現在還不清楚,因為他們依靠有效的初級健康照護和機制支付費用。需要更進一步嚴謹的評估研究,特別是在低收入國家中實施有條件的現金救助計畫,如撒哈拉沙漠以南非洲或南亞洲.

翻譯人

本摘要由高雄榮民總醫院金沁琳翻譯。

此翻譯計畫由臺灣國家衛生研究院(National Health Research Institutes, Taiwan)統籌。

總結

在中低收入國家中有條件的現金救助對於健康結果及健康照護使用的影響:我們找到29篇文章關於有條件的現金救助對於接近照護及健康結果的影響。其中10篇文章的報告來自6項研究的結果,其符合納入標準;其中4項研究為隨機實驗。儘管某些研究中有一些方法學上的缺點,整體的研究證據認為有條件的現金救助方式也許可以使貧民獲得一些健康上的益處。有條件的現金救助計畫包括許多的組成部分,包括鼓勵參加健康教育,測量身高體重,免疫接種與營養補充。有條件的現金救助計畫似乎是一種有效的方式以增加預防性服務的使用並鼓勵一些預防性的行為。在某些計畫有顯著的改善健康結果,雖然哪個部分貢獻了正向的影響尚未明確。

摘要

有条件的现金救助对于中低收入国家人群健康结局及医疗服务利用的影响

研究背景

有条件的现金救助(Conditional cash transfers ,CCT)是指对符合预设要求的家庭提供现金的帮助。CCT项目合法化的基础在于: 对需方进行补贴以解决贫民在获得健康与社会服务方面的不公平性问题。在过去几十年里,CCT项目已变得越来越受欢迎,特别是在拉丁美洲的中等收入国家。

研究目的

评价CCT项目对于增加保健机会与改善健康结局的效果,尤其是其针对中低收入国家贫民的效果。

检索策略

我们检索了全球范围大量的数据库,包括Cochrane Central Register of Controlled Trials (CENTRAL),MEDLINE,EMBASE,以及发展研究和经济学数据库。我们也检索了许多国际机构、组织及大学的网站和在线资源,以寻找未公开发表的相关文献。初期的检索于2005年11月至2006年4月进行,并于2009年5月在MEDLINE进行了更新检索。

标准/纳入排除标准

有条件的现金救助(Conditional cash transfers ,CCT)是指对符合预设医疗保健相关要求的家庭提供现金的帮助。研究必须包括至少一项以下客观测量指标: 医疗保健利用,健康消费,健康结局或公平性结果。合格的研究设计方案为: 遵循Cochrane Effective Practice and Organisation of Care Group标准的,有关健康财政政策影响的随机对照试验,间断时间序列分析,或前后对照研究。

数据收集与分析

对证据进行定性分析。

主要结果

纳入了来自6项干预研究的10篇文章。总体而言,设计质量与分析方法降低了偏倚风险。一些CCT项目分别采用人体测量数据和疾病发生的自陈报告进行评估,为CCT项目对人群健康服务、营养状况及健康结局的积极影响方面提供了有力的证据。这些正向效果很难被明确地归因于现金刺激,因为其它因素也可能有关。还有一些研究提供了其对儿童及怀孕妇女使用预防性健康服务产生有利影响的证据。但没有找到对医疗保健消费有何影响的证据。

作者结论

一些精心设计的评估研究充分认可,有条件的现金救助计划是一种有效的促进预防性卫生服务的方式。由于该类计划依靠初级卫生保健和不同的费用支付机制,因此还不明确是否可以在不同条件下进行复制,特别是在某些条件不具备的情况下更是如此。需要开展更严格的CCT项目评估研究,特别是那些正在引入CCT项目的低收入国家,如撒哈拉沙漠以南非洲或南亚。

Résumé scientifique

L'impact de rémunérations conditionnelles sur la santé et l'utilisation des services de santé dans les pays à faibles et moyens revenus

Contexte

Les programmes de rémunérations conditionnelles (RC) consistent à transférer de l'argent aux ménages à condition de remplir certaines conditions prédéfinies. Les programmes RC ont été justifiés en alléguant que des subventions du côté de la demande étaient nécessaires pour remédier aux inégalités d'accès aux services médicaux et sociaux dont souffrent les populations pauvres. Leur popularité n'a cessé de croître ces dix dernières années, en particulier dans les pays à moyens revenus d'Amérique latine.

Objectifs

Évaluer l'efficacité des RC pour améliorer l'accès aux soins et les résultats cliniques, en particulier pour les populations les plus pauvres des pays à faibles et moyens revenus.

Stratégie de recherche documentaire

Nous avons consulté de nombreuses bases de données internationales, y compris le registre Cochrane des essais contrôlés (CENTRAL), MEDLINE et EMBASE, ainsi que des études sur le développement et des bases de données économiques. Nous avons également consulté les sites et ressources Web de nombreuses agences, organisations et universités internationales pour identifier la littérature grise pertinente. Les recherches originales ont été menées entre novembre 2005 et avril 2006. Une recherche mise a jour a été effectuée dans MEDLINE en mai 2009.

Critères de sélection

Les RC étaient considérés comme des transferts d'argent effectués aux ménages sous réserve de respecter certaines conditions prédéfinies en matière de soins de santé. Les études devaient intégrer une mesure objective d'au moins l'un des critères de jugement suivants : utilisation des soins de santé, dépenses de santé, résultats cliniques ou résultats en matière d'équité. Les plans d'étude admissibles étaient les suivants : essais contrôlés randomisés, études de séries chronologiques interrompues ou études contrôlées avant-après portant sur l'impact des politiques de financement des soins de santé et conformes aux critères définis par le groupe de revue Cochrane sur l'efficacité des pratiques et l'organisation des soins.

Recueil et analyse des données

Les données probantes ont fait l'objet d'une analyse qualitative.

Résultats principaux

Dix articles rapportant les résultats de six interventions ont été inclus. Dans l'ensemble, la qualité du plan d'étude et l'analyse limitaient les risques de biais. Plusieurs programmes RC apportaient des preuves solides d'un impact positif sur l'utilisation des services de santé, l'état nutritionnel et les résultats cliniques, évalués respectivement par les mesures anthropométriques et les épisodes de maladie rapportés par les intéressés. Il est difficile d'attribuer spécifiquement ces effets positifs aux incitations financières car d'autres composants peuvent également entrer en ligne de compte. Plusieurs études apportent des preuves des effets positifs sur l'utilisation des services préventifs par les enfants et les femmes enceintes. Aucune preuve n'était rapportée concernant les effets sur les dépenses de santé.

Conclusions des auteurs

Les programmes de RC ont fait l'objet d'évaluations bien planifiées qui suggèrent fortement qu'ils pourraient être efficaces pour améliorer l'accès aux services préventifs. Leur réplication dans d'autres conditions (en particulier dans des environnements plus défavorisés) n'est pas encore établie car ils dépendent de l'efficacité des soins de santé primaires et des mécanismes de transfert. D'autres recherches rigoureuses sont nécessaires, notamment concernant les RC introduits dans des pays à faibles revenus, comme en Afrique subsaharienne ou en Asie du Sud.

Plain language summary

The impact of conditional cash transfers on health outcomes and use of health services in low and middle income countries

We found 29 papers on the impact of conditional cash transfers (CCT) on access to care and health outcomes. Of these, ten papers, reporting results from six studies, satisfied the inclusion criteria; four of these studies were randomised experiments. Despite a number of methodological weaknesses in some studies, overall the research evidence suggests that CCT schemes may result in a number of benefits to health for poor populations. Many conditional cash transfer programmes include a number of components, including incentivising attendance for health education, measurements of height and weight, immunisations and nutritional supplementation. Conditional cash transfer programmes appear to be an effective way to increase the uptake of preventive services and encourage some preventive behaviours. In some cases programmes have noted improvement of health outcomes, though it is unclear to which components this positive effect should be attributed.

概要

有条件的现金救助对于中低收入国家人群健康结局及医疗服务利用的影响

在检索到的涉及CCT项目对于卫生服务可及性及健康结局影响的29篇文章中,来自6项研究的10篇文章的报告结果符合纳入标准,其中4项研究为随机试验。尽管某些研究有方法学上的缺陷,但总体而言可以认为有条件的现金救助方式能够使贫民获得一些健康方面的利益。有条件的现金救助计划包括许多内容,如: 鼓励参与健康教育活动、测量身高体重、免疫接种与营养补充。它似乎是一种增进预防性服务利用并鼓励预防性行为的有效方式。虽然在某些情况下,有条件的现金救助计划显著地改善了健康结果,但是尚未明确哪些因素有积极影响。

翻译注解

本摘要由重庆医科大学中国循证卫生保健协作网(China Effective Health Care Network)翻译。

Translated by: China Effective Health Care Network

Résumé simplifié

L'impact des rémunérations conditionnelles sur la santé et l'utilisation des services de santé dans les pays à faibles et moyens revenus

29 articles portant sur l'impact des rémunérations conditionnelles (RC) sur l'accès aux soins et les résultats de santé ont été identifiés. Dix de ces articles (rapportant les résultats de six études) étaient conformes aux critères d'inclusion ; quatre étaient des essais randomisés. Malgré la faiblesse méthodologique de certaines études, les données probantes suggèrent que les programmes RC pourraient avoir des effets bénéfiques sur la santé des populations pauvres. De nombreux programmes de RC incluent plusieurs composants, tels qu'une incitation à recevoir une éducation sanitaire, des mesures de la taille et du poids, des vaccinations et des suppléments nutritifs. Les programmes de RC semblent efficaces pour améliorer le recours aux services préventifs et encourager certains comportements préventifs. Dans certains cas, les programmes ont permis une amélioration des résultats cliniques. On ignore cependant quels composants sont associés à cet effet positif.

Notes de traduction

Traduit par: French Cochrane Centre 1st July, 2012
Traduction financée par: Ministère du Travail, de l'Emploi et de la Santé Français

Laički sažetak

Mogu li uvjetovani financijski poticaji utjecati na zdravlje i uporabu zdravstvenih usluga u zemljama srednjih i niskih prihoda

U ovom Cochrane sustavnom pregledu pronađeno je 29 kliničkih pokusa u kojima je ispitan učinak uvjetovanog plaćanja na korištenje zdravstvene skrbi i zdravlje. U 10 radova opisani su rezultati 6 studija, a 4 od tih studija bili su randomizirani eksperimenti. Usprkos nizu metodoloških nedostataka koji su utvrđeni u nekim studijama, ukupno dokazi ukazuju da uvjetovani financijski poticaji mogu dovesti do niza pozitivnih učinaka na zdravlje siromašnih stanovnika. Brojni uvjetovani financijski poticaji uključuju komponente kao što su financijski poticaj za pohađanje zdravstvene edukacije, mjerenje visine i težine, rimanje cjepiva i dodataka prehrani. Čini se da su uvjetovani financijski poticaji djelotvoran način za poticanje siromašnih osoba na sudjelovanje u preventivnim zdravstvenim uslugama i da potiču određena preventivna ponašanja. U nekim slučajevima programi su doveli do poboljšanja zdravlja, iako je nejasno koji dijelovi programa su zaslužni za pozitivan učinak.

Bilješke prijevoda

Hrvatski Cochrane
Prevela: Livia Puljak
Ovaj sažetak preveden je u okviru volonterskog projekta prevođenja Cochrane sažetaka. Uključite se u projekt i pomozite nam u prevođenju brojnih preostalih Cochrane sažetaka koji su još uvijek dostupni samo na engleskom jeziku. Kontakt: cochrane_croatia@mefst.hr

Summary of findings(Explanation)

Summary of findings for the main comparison. 
OutcomesRelative effectQuality of evidenceComments

Health services
utilisation

 

All studies reported an increase in the use of health services in the intervention groups  (27% increase in individuals returning for
voluntary HIV counselling, 2.1 more visits per day to health facilities, 11-20% more children taken to the health centre in the past month, 23-33% more children < 4 yrs attending preventive healthcare visits)
LowFindings taken from 5 studies (3 C-RCT, 1 RCT, 1 CBA). Two studies report results from facility-based routine data which are not always reliable (in one of these studies, there was a risk of contamination bias).
ImmunisationcoverageMixed results were found  (increased vaccination rates in children for measles and tuberculosis but only in specific groups or temporarily, and no change in one study)Moderate

Findings from 3 C-RCT and 1 CBA.

Differences in effects might be due to initial rates of immunisation (effects found in cases where pre-intervention rates were relatively low)

Health outcomes

 

Mixed effects on objectively measured health outcomes (anaemia) and positive effects on mothers reports of childrens health outcomes (22-25% decrease in the probability of children <3 years old being reported ill in the past month)

 

 

ModerateResults from 3 studies (2 C-RCT and 1 CBA)
Childrens anthropometric outcomes

Positive effects found on childrens growth (increase in height by about 1cm amongst children < 4 years old and ); however there were two contradictory findings on the impact on height-for-age Z scores (1 study found a significant increase while another one found a negative impact, equivalent in size)

Decrease in the effects of malnourishment (decrease in the probability of being stunted, underweight or chronically malnourished)

ModeratePositive effects found in three studies (2 C-RCT, 1 CBA) ; the only one negative outcome was found in a quasi-C-RCT (more risks of bias) which might have arisen from misunderstanding on programme conditions

Background

Cash transfers are defined as the provision of assistance in the form of cash, with the objective of increasing the household's real income. They are generally made to the poor or to those who face a probable risk of falling into poverty in the absence of the transfer. Conditional cash transfers (CCT) have recently been introduced in several Latin American countries. Based on a similar principle, they provide monetary transfers to households on the condition that they comply with a set of requirements. Conditional cash transfers are increasingly being promoted over in-kind transfers and unconditional for several reasons. First, unlike in-kind transfers, which pre-determine the provision of a particular commodity, CCT are more flexible safety nets, that allow individuals to buy items according to their needs or preferences. Secondly, in-kind transfers have sometimes been criticised for the important logistical costs they usually entail (e.g. transportation costs to bring bulky products to remote areas, costs associated with loss of food, etc.). In addition, the conditionalities of  CCT programmes have provided useful arguments against the critiques sometimes made to social transfers, which is that they are useless and a waste of resources. Promoters of CCT have emphasised that conditional transfers were a direct investment in human capital, from which there would be some long term benefits. Finally, CCT have been advocated for being more ambitious than unconditional transfers, since they are an incentive for households to adopt a behaviour that would positively impact on their well-being.

CCT programmes were developed in Latin America in the mid-1990s to counteract the devastating social and economic effects of the debt crisis of the 1980s and the financial crises of Mexico (1995) and Asia (1997). Some municipalities in Brazil introduced conditional cash transfers as early as 1995. In 1997, Mexico started a large-scale pilot programme, the Programa de Educación, Salud y Alimentación (Progresa, later called Oportunidades), which was extended at national level two years later. The widespread positive results from Progresa served as an encouragement to extend these programmes in many other countries.

CCT programmes are justified on the grounds that demand-side subsidies are necessary to address particular constraints and bottlenecks of social services provision. Market failures are usually cited as the main economic rationale: the consumption of some goods creates positive externalities that justify their subsidy, in order to maximise their uptake by the population. This is the reason why CCT programmes usually aim to increase demand for preventive health services and education, because such programmes generate positive spillover effects. CCT are also supposed to help overcome different barriers to access to social services. Monetary transfers provide households with money to compensate for indirect costs (e.g. costs of transport, or food during hospitalisation) or opportunity costs (for example the loss of income due to the time not spent on the usual income-generating activity) related to seeking health care or sending children to school.  Finally, these programmes are often justified by social equity concerns. As poor people usually accumulate the detrimental effects of different barriers to access, CCT mechanisms are seen as a single transfer mechanism that can “level the playing field” and redistribute endowments in order to equalise opportunities in a society.

It is important to underline that the overall objective of recent CCT programmes is usually to provide support to families living in extreme poverty, in order to develop the long-term potential of the household members. Therefore, their aims are broader than those of scaling-up effective (preventive) health interventions, and include the larger issue of human capital building. They not only provide a financial incentive for households to comply with beneficial behaviours, but also usually entail free access to basic health services. Consequently, CCT can create a positive effect on the demand for health services by reducing or eliminating financial barriers to access, and potentially have positive effects on incomes for beneficiary households.

CCT have grown very popular in the recent past, and they have started to develop in many developing countries, notably outside of Latin America. Examples include conditional incentive programmes for pregnant women to deliver in health facilities in India and Nepal (Ministry of Health and Family Welfare 2005; Powell-Jackson 2009), but also programmes in Ecuador, Jamaica, Turkey and Kenya. Future impact assessments of their benefits should contribute to the current debate and knowledge on the issue, and will be included in an updated version of this review.

No systematic review has been done on this subject, although a couple of narrative reviews exist (Ensor 2003; Rawlings 2005). This review was published in a past issue of JAMA (Lagarde 2007).

Objectives

This review aims to assess the effectiveness of conditional monetary transfers in low and middle income countries to improve the health outcomes of populations and their access to health care services. Changes in access to health services will be evaluated through changes in the use of health services and changes in health care expenditures.

Methods

Criteria for considering studies for this review

Types of studies

We examined all studies that met the Effective Practice and Organisation of Care Group (EPOC) inclusion criteria for study design and compared the effects (on a determined range of outcomes) of offering conditional cash transfers to the populations with the absence of such incentive.

We included three types of studies:

1. Randomised controlled trials (RCT) or cluster-randomised controlled trials (C-RCT)

2. Controlled before and after studies

For these two types of studies, the comparison intervention was the provision of the same type of health services (by the same providers), but without offering incentives to the populations to come and use health services.

3. Interrupted time-series analyses provided that:

  • the point in time when the intervention/change occurred was clearly defined;

  • there were at least three or more data points before and after the intervention.

Types of participants

The review includes only studies that took place in low and middle income countries as defined by the World Bank (World Bank 2006).

Units of study were the populations who would potentially access health services. Issues of interest were the populations’ access to health services, their utilisation patterns, and possibly their health outcomes. Hence, “participants” included users and non-users of health services, as well as institutions such as health facilities, where utilisation data could have been collected.

We permitted study designs that used facilities or districts as units of allocation and were thus cluster trials.

We included studies on all types of providers (private, governmental, NGOs). We did not limit the scope of our study to a particular level of health care delivery and all types of health services were eligible for inclusion.

Types of interventions

To be included, interventions had to meet the following criteria:

  • consist of direct monetary transfers made to households. We did not include in-kind transfers because the review focused on the effectiveness of financial incentives, which could be easily compared to each other;

  • the transfer had to be conditioned on a particular behaviour or action (e.g. visit to a health facility for regular check ups) – unconditional transfers were not considered.

Types of outcome measures

Primary outcomes

Primary outcomes were changes in use of health services and changes in health outcomes.

  • Only objective measures relating to the final consumption of health services were taken into consideration. Access to care can be measured by changes in utilisation patterns of health facilities or services (immunisation coverage, number of visits, rates of hospitalisation, numbers of people having bought an insecticide-treated net, etc.) and/or equivalent information collected directly from the study population through rigorous survey techniques. Information related to distance travelled or travel time was outside the scope of this review.

  • Changes in health outcomes, measured by morbidity and mortality incidence (broken down by age group, sex, etc.) were also considered where available.

Secondary outcomes

Secondary outcomes included health care expenditures and outcomes reflecting changes in equity of access:

  • Health care expenditure was considered when it reflected direct (and indirect) costs borne by the patient and/or her family.

  • Changes in equity of access – increased access for disadvantaged groups or a reduction in gaps in coverage – could also be an important outcome measure. This required a preliminary analysis and categorisation of the population of interest along a socio-economic scale. We accepted any relevant methodology (e.g. wealth/asset index) provided it was rigorous and described in detail.

Objective measures of utilisation, performance or patient outcomes were required. We did not include studies based only on measurements of attitudes, beliefs or perceptions.

Search methods for identification of studies

Electronic searches

The search to identify studies for this review was initially done as part of a much wider review on health financing mechanisms dealing with the effects of several financing  strategies (Lagarde 2006). The broad review has been split into several sub-reviews, including the present one. Therefore the search methodology included terms that encompass a broader scope that the one defined for this review.

The following electronic databases were originally searched without language or date restrictions (the dates indicated refer to the original searches performed):

PubMED, 11/11/2005

EMBASE (Athens), 19/04/2006

Popline, 08/12/2005

African Healthline (bibliographic databases on African health issues), 28/04/2006

IBSS (International Bibliography in Social Sciences, Athens interface), 19/04/2006

The Cochrane Central Register of Controlled Trials (CENTRAL), 20/01/2006

The Database of Abstracts of Reviews of Effectiveness and the EPOC Register (and the database of studies awaiting assessment), 20/01/2006

BLDS, 03/11/2005

ID21, 24/11/2005

ELDIS, 25/11/2005

The Antwerp Institute of Tropical Medicine database, 26/01/2006

Jstor, 26/01/2005

Inter-Science (Wiley), 16/12/2005

ScienceDirect, 16/12/2005

IDEAS(Repec), 20/01/2005

LILACS, 19/04/2006

CAB-Direct (Global Health), 17/04/2006

Healthcare Management Information Consortium (HMIC), 17/04/2006

World Health Organization Library Information System (WHOLIS), 18/04/2006

MEDCARIB, 19/04/2006

ADOLEC, 19/04/2006

FRANCIS, 16/12/2005

BDSP, 16/12/2005

USAID database, 04/11/2005.

An updated search was done in May 2009. The detailed search strategy used for this updated search is indicated in Appendix 1. We have identified a few other studies as potentially relevant for this review and these will be assessed for inclusion in the next version of this review. These studies can be found under Studies awaiting classification.

The PubMED search strategy was mainly developed using reviews cited in the background section of the protocol and their references (Lagarde 2006).

The original search strategy was developed without the usual EPOC methodology filter. However, the updated search strategy included such a methodology filter to limit study designs to randomised trials, controlled trials, time series analyses and controlled before-after studies.

The detail of the search strategy used for PubMed for can be found in Appendix 1. We translated this search strategy into the other databases using the appropriate controlled vocabulary, as applicable. Search strategies for electronic databases used selected index terms and free text terms. In addition, we used a number of free text terms to browse more simple databases or lists of studies: “health financing”, “contracting”, “pay for performance”, “outsourcing”, “supply-side incentive”, “performance payment”, “output-based payment”, “P4P”.

Searching other resources

We also searched the following grey literature resources between December 2005 and February 2006.

  • Websites and online resources of UNICEF, USAID and the World Bank, Partnerships for Health Reforms, Abt Associates, Management Sciences for Health (MSH), Oxford Policy Management, Save the Children, Oxfam, and a number of other networks or organisation websites such as The Private Sector Partnerships-One, the Indian Council for Research on International Economic Relations, Equinet - The Network for Equity in Health in Southern Africa, the Organization for Social Science Research in Eastern and Southern Africa (OSSREA).

  • Websites and online resources (working papers) of numerous university research centres: among others the Institute of Social Studies, The Hague, the University of Southampton, the International Centre for Diarrhoeal Disease Research and  the Centre for Health and Population research, Dhaka, the Boston University  Institute for Economic Development, Harvard Initiative for Global Health, Cornell Food and Nutrition Policy Programme, the Institute of Development Studies (University of Sussex), the London School of Hygiene and Tropical Medicine (HEFP website), the Institute of Policy Analysis and Research (IPAR) in Kenya, the Development Policy Research Unit of the University of Cape Town, the Netherlands Institute for Southern Africa.

We screened the reference lists of all of the relevant references retrieved. We contacted authors of relevant papers or known experts in the fields of interest to identify additional studies, including unpublished and ongoing studies.

Data collection and analysis

Selection of studies

Two authors (ML and AH) independently selected the studies to be included in the review. We resolved any disagreements by discussion.

Data extraction and management

We extracted the following information from the included studies using a standardised data extraction form:

  • type of study (individual or cluster randomised trial, controlled before-after, interrupted time series);

  • duration of the study;

  • study setting (country, key features of the health care system, external support, other health financing options in place, other on-going economic/political/social reforms);

  • characteristics of participants (catchment area size, characteristics of the population, existing health facilities, etc.);

  • characteristics of the intervention (relative and absolute amount of the transfer, conditions to be fulfilled);

  • main outcome measures and results.

Tables were prepared for each sub-category of intervention, including the following information: study ID, country and date of the intervention, characteristics of the intervention and the individual (facility/population level) and external/national level, health outcomes.

Assessment of risk of bias in included studies

We adapted slightly the standard criteria recommended by EPOC to match the particularities of the studies found in the field of interest (EPOC 2002). For example, criteria about following-up patients or doctors were not relevant as most of the studies used population survey data. Follow-up surveys, when carried out, would therefore not be done with the same population, but with a new random sample. In addition, we added some specific criteria to account for some of the limitations of studies found (e.g. no statistical analysis performed or failure to account for clustering effects). Appendix 2 presents the detailed list of all quality criteria used, and explains the amendments introduced to the original EPOC criteria for each type of design.

The criteria for RCTs and C-RCTs were:

  1. Concealment of allocation

  2. Protection against exclusion bias

  3. Appropriate sampling strategy

  4. Appropriate analysis

  5. Reliable primary outcomes measures

  6. Protection against detection bias

  7. Baseline measurement of outcomes

  8. Protection against contamination

The criteria for CBA studies were:

  1. Baseline measurement of outcomes

  2. Baseline characteristics of studies using second site as control

  3. Protection against exclusion or selection bias

  4. Protection against contamination

  5. Reliable primary outcomes measures

  6. Appropriate analysis of data

The criteria for ITS studies were:

  1. protection against changes

  2. appropriate analysis of the data (or re-analysis possible)

  3. Protection against selection bias

  4. Reliability of outcome data

  5. Number of points specified

  6. Intervention effect specified

  7. Protection against detection bias

Our assessment of the risk of bias in the included studies is presented in Table 1.

Table 1. Assessment of risk of bias in included studies
Controlled before and after (CBA) studies
Study IDBaseline characteristicsEquivalent control siteProtection against exclusion or selection biasProtection against contaminationReliability of  outcome measuresAppropriate analysisOverall: LimitationsNotes
Attanasio 2005NOT CLEARNOT CLEARDONEDONEDONENOT DONEhigh risk of biasDoes not take cluster correlation into account. Differences at baseline between control and treatment sites are mentioned in the text but no further precision is given.
Morris 2004bNOT DONENOT CLEARDONEDONEDONEDONEhigh risk of biasBaseline measures were reconstructed afterwards. Authors mention potential differences at baseline.
Randomised controlled trials
Study IDConcealment of allocationProtection against exclusion biasSamplingAppropriate Analysis (clustering)Quality/ reliability of the dataProtection against detection biasBaseline MeasurementProtection against contaminationOverall: LimitationsNotes
Maluccio 2004DONEDONENOT CLEARDONEDONENOT CLEARDONEDONEmoderate risk of biasNo reliability data presented on anthropometric measures, and no details on sampling.
Morris 2004aDONEDONEDONEDONEDONENOT CLEARDONEDONEmoderate risk of biasThe only potential bias would be a declaration bias as some outcomes for children are not objective but measured on mothers’ declaration and registries of facilities (the authors mention a problem of over-declaration).
Thornton 2006NOT DONEDONEN/AN/ADONEDONENOT DONENOT DONEhigh risk of biasBiased allocation (non random): higher number of vouchers given out than would be expected by chance even when nurses where threatened with termination of  employment; no baseline (assumption is that randomissation of subjects  is perfect); some contamination was noted.
Gertler 2000NOT DONEDONEDONEDONENOT DONENOT DONEDONEDONEhigh risk of biasClustering effects mentioned on some occasions but not everywhere - health utilisation data from registers not necessarily reliable (+ includes both beneficiaries and non-beneficiaries); reported illness by mothers potentially biased - Berhman and Hoddinot (1999) show that assignment was random at community level but not at individual level.
Barham 2005aNOT DONENOT CLEARDONEDONENOT DONENOT DONENOT CLEARDONEhigh risk of biasThe author had to adjust and modify the collected data that suffered many methodological problems - suffers same problem as other Progresa (randomisation by community but not at individual level where the author considers the results) - problems of data recording (cumulative immunisation collected instead of those in the last 6 months) which can possibly lead to over-estimates of the positive results - differences in measles immunisation rates - problem of attrition of sample (not real cohort or panel data)
Gertler 2004aNOT DONEDONEDONEDONENOT DONEDONEDONENOT DONEhigh risk of bias 
Rivera 2004NOT DONEDONENOT CLEARDONENOT DONEDONENOT DONENOT DONEhigh risk of biasDoubts on the quality of the data confirmed by Berhman 2001/2005 (same set of data used): leakage problems, non-random assignment of papilla, attrition of sample, etc.
Behrman 2005NOT DONENOT DONENOT CLEARDONENOT DONEDONENOT DONENOT DONEhigh risk of biasLeakage problems, non-random assignment of nutrition supplements, attrition of sample between 1998 and 1999 causing bias towards over-representation of poor households in the usable cohort, important differences at baseline (see Table 1 in Behrman 2001) - however strenuous attempts made to reduce bias and overall risk of biased results may be moderate.

Data synthesis

Due to the diversity in the nature of interventions and outcomes reported in the included studies, it was not appropriate to statistically combine the results of the studies.

For all studies, we tried to report the outcome measures before and after the interventions, but these were not systematically available. Ideally, we would have calculated the impact of the studies by comparing the outcome measures in both intervention and control areas. This was not made possible, due to insufficient data reported in the original papers.

All the reported estimates of effects therefore come directly from the original studies. We reported only the estimates of effects that accounted for differences in baseline outcomes. Some studies controlled for other baseline characteristics (e.g. socio-economic individual characteristics of survey participants). This was usually performed in a regression analysis, and therefore the estimated effect represents a change in the outcome of interest (e.g. percentage points if the outcome was a proportion, increase in the probability of the dependent latent variable of the regression is a probability).

Results

Description of studies

See: Characteristics of included studies; Characteristics of excluded studies.

This review summarises the results from ten papers, reporting the results from six studies.

Study designs

We included ten papers reporting results from four randomised trials, and two controlled before and after studies in the review.

Characteristics of settings and patients

Thornton 2006 reported the results of a small-scale experiment in Malawi. All other included studies reported results from large-scale experiments or projects in Latin American middle-income countries: Mexico (the 5 papers that reported results from Progresa: Barham 2005a; Behrman 2005; Gertler 2000; Gertler 2004a; Rivera 2004), Brazil (Morris 2004b) Nicaragua (Maluccio 2004), Colombia (Attanasio 2005) and Honduras (Morris 2004a). We reported differences in beneficiaries in the following section, as targeting strategies were one of the core features of these interventions.

Characteristics of interventions

In all studies, the intervention was targeted at individuals, but was sometimes provided at the community level to all individuals.  

In CCT programmes, even though all household members are likely to benefit from the monetary transfers, target populations are those who have to abide by some conditionalities. Thornton 2006 focuses on people who have been tested for HIV. All Latin American CCT programmes, which are very similar in their conception, target poor and disadvantaged groups, mostly infants and children, and pregnant and lactating women.

The benefit packages of CCT programmes vary not only across programmes, but also with the characteristics of the beneficiaries within a programme. We chose to report the main differences, but various operational dimensions, like targeting or frequency of transfers, are also noted.

All Latin American studies included are concerned with programmes that aim to strengthen the human capital of beneficiaries (in general children); therefore they provide cash, free access to health services (preventive health check-ups for infants and pregnant women) and sometimes also nutritional supplements. Further, there is an education component in all schemes (see Table 2 for a description of the benefits of each package).

Table 2. Context and intervention description
Study IDNature of intervention and control sitesIndividual contextual factorsBroader contextual factor

Attanasio 2005

Colombia (Familias en Accion’)

 

2 types of monetary transfers to the mothers: 1/ conditional on their children under 7 attending preventive health care visits (where children are weighed), they receive US$15 per month. Mothers are also encouraged to attend courses on hygiene, nutrition and family planning.

2/ conditional on their children aged 7-17 attending at least 80% of school classes, they receive a monthly grant per child (approx. US$8 for primary school and US$16 for secondary school).

 

Eligibility: being Colombian citizen, living in a community where the programme is offered, having children under 18 and belonging to the lowest level of the official socio-economic classification.

 

Eligible municipalities had to have less than 100,000 inhabitants and have enough health and education facilities to guarantee the absence of supply bottlenecks. Of the 1024 municipalities in the country, 691 qualified.

“Familias en Accion” programmeis funded by a IADB and WB loan approved in 2000.

 

An important nutritional programmewhere children receive nutrition supplements, Hogares Communotarios (HC) had been implemented for 16 years in Colombia. Mothers had to choose between enrolling in FA or in HC, so the effects of FA are in comparison with HC.

 

 

 

 

Barham 2005a

Mexico (‘Progresa’)

 

 Same as Gertler 2000Same as Gertler 2000Same as Gertler 2000

Behrman 2005

Mexico (‘Progresa’)

 Same as Gertler 2000Same as Gertler 2000Same as Gertler 2000

Gertler 2000

Mexico (‘Progresa’)

Intervention: 2 cash transfers every two months; one general and one depending on school attendance

- nutrition component: food supplements for children aged 4-23 months, under-weight children aged 2-4 years, and pregnant and lactating women in beneficiary households

- health component: regular health care appointments in health centres for the whole family

- education component:

 

506 out of 50,000 eligible villages were randomly chosen.

 

Intervention groups: households selected from 320 communities.

 

Control group: 186 communities.

 

Value of the transfers: US$25, adding 20-30% to the household income

The controls should originally have acted as controls for 2 years, but for political reasons intervention in control communities occurred in late 1999 so only 1 year ½ of comparison was possible and the control communities were therefore considered as crossover intervention communities after 1 year of observation.

Gertler 2004a

Mexico (‘Progresa’)

Same as Gertler 2000Same as Gertler 2000Same as Gertler 2000

Maluccio 2004

Nicaragua (Red de proteccion social’)

The programme has 2 components :

-    a monthly “food security” cash transfer (bono alimentario= US$224 per year=13% of total amount of household expenditures in beneficiary households before the program) conditional on attendance at monthly health educational workshops, on bringing their children under age 5 for free scheduled preventive child-care appointments (which include the provision of antiparasites, vitamins and iron supplement), on having up-to-date vaccination, and on adequate weight gain.

-    A “school attendance” cash transfer every two months (= US$112 per year=8% of total amount of household expenditures in beneficiary households), contingent on enrolment and regular school attendance of children aged 7-13. Additionally the household receives an annual cash transfer per eligible child for school supplies.

Beneficiaries did not receive the food or education cash transfers if they failed to comply with any of the conditions.

The programme is ultimately targeted at poor households living in rural areas, but the pilot phase analysed in this study occurred in 2 departments (Madriz and Matagalpa) in the Northern part of the Central Region. This region is the only one in the country where poverty worsened during 1998 and 2001.

 

These pilot sites are not representative of the country situation:

-    within the 2 chosen departments, 6 municipalities were chosen (out of 20) because they had benefited from a previous programme that developed the capacity of the governing bodies to implement and monitor social projects: “it is possible that the selected municipalities had atypical capacities to run RPS”.

-    in the chosen municipalities, 78-90% of the population is extremely poor/poor, compared to 21-45% at national level.

 

42 eligible areas (the neediest) were chosen for the pilot programme based on wealth index.

Private providers were specifically trained to deliver the specific health-care services required by the programme.

Incentives were also given to teachers to compensate for the larger classes they had after the implementation of the programme.

 

10% of beneficiaries were penalised at least once during the first two years of the programme; 5% were expelled or left the programme.

Some conditions (adequate weight gain) were dropped at the end of the pilot phase and others were not properly enforced (up-to-date vaccination while there were delays in the delivery of vaccines).

Delays occurred in the implementation of the health component which finally started in June 2001. Therefore when the first follow-up survey was realised in Oct. 2001 the beneficiaries had been receiving the transfers for the education component for 13 months and those for the health and nutrition component for 5 months only.

 

The “Red de Proteccion Social” (RPS) project is financed by a loan from the IADB.

 

The impact analysis of the pilot phase was done by the International Food Policy Research Institute (IFPRI).

 

Possible detection of the “Hawthorne effect” since performance of the programme was slightly lower the second year.

 

Over the 2 years the actual average monetary transfer to households represented 18% of total household expenditure (similar to PROGRESA but 5 times larger than PRAF). The nominal transfers remained constant during the 2 years of the programme, thus the real value of the transfer declined by 8% due to inflation.

 

 

 

 

 

Morris 2004a

Honduras (‘PRAF’)

Either or both of :

1) 2 types of monetary incentives: an education one conditional on school attendance of children aged 6-12; a health transfer conditional on monthly visits to health centres for children and pre-natal check-ups for pregnant women.

2) Resources to local health teams plus community-based nutrition intervention compared with standard services.

Value of the transfer:

- Monthly health bonus=£2.50 (conversion rate late 2001) per pregnant women or child under 3, up to a maximum of two

- Monthly education bonus=£3.70 per child between 6 and 12 enrolled at school, up to a maximum of 3.

Annual entitlement averaged £60 per household.

It is reported that approx 75% of the population live on less than £1 a day.

Municipalities were those that had highest prevalence of malnutrition in country.

Transfer of resources to local health teams could not be properly implemented for legal reasons.

First phase of PRAF funded by the government of Honduras since 1990. Objective of PRAF= increase demand for preventive health care in pregnant women, new mothers and children aged 0-3.

The second phase was funded by a loan from the Inter-American Development Bank (IADB) in 1998. The second phase increased the value of the vouchers, removed subjective elements in beneficiary selection.

 

 

 

Morris 2004b (Bolsa Alimentacao)

Households received a monetary transfer whose size depended on the number of eligible members in the household.

The transfers were conditional on attendance to nutrition workshops by mothers, , regular attendance at antenatal care (if pregnant) and growth monitoring visits for children.

Beneficiaries were selected in a two-stage process: in the first stage municipalities with high rates of malnutrition were chosen ; then selected municipalities identified beneficiaries

Value of the transfers: from US$6.25 to US$18.7

 

Beneficiaries are compared with individuals who were deemed not eligible due to quasi-random administrative errors in the programme management (problems with data transfer from one body to another, problems with some characters in the names, problems of non-concordance of administrative records)

Rivera 2004

Mexico (‘Progresa’)

Same as Gertler 2000

This nutritional impact sub-study was conducted in a randomly selected 205/320 intervention and 142/186 control communities.

Same as Gertler 2000

Same as Gertler 2000

Thornton 2006

Malawi

Vouchers given at time of taking test

sample. Cash payment received on returning  voucher when attending for either HIV or STD tests results.

 

 

Test results became available 2-4 months after blood was taken.

 

Intervention group: monetary incentive ranging from $1-$3.

 

Control group (20% of total participants): no payment.

 

HIV context in Malawi: availability of VCT.

In absolute terms, transfer sizes of the packages by households are quite variable, making it difficult to compare the effects of the packages due to the differences in economic environment across countries. Comparing the relative share of beneficiaries’ income could have been useful, but unfortunately these data were not available in most studies.

Conditionality

All studies from Latin America described interventions combining nutrition, education and health conditionalities (see Table 2 and Table 3), as their objective is to improve the human capital of beneficiaries.  These programmes therefore have several requirements. Monetary transfers are conditional on health check-ups and school attendance at primary level for young children and some programmes add a health education component for the parents, secondary education for older children and nutrition supplements. Unlike the Latin American programmes, Thornton 2006 tested the effectiveness of an incentive to be HIV tested and to collect the result. 

Table 3. Details of requirements of included programmes
 Cash transfers conditional upon:
 Primary EducationSecondary Education

Health visits

(pregnant women)

Health visits

(children)

Nutrition supplementsHealth education workshopsOthers

Progresa

Mexico

 √ 

PRAF

Honduras

√   √   

RPS

Nicaragua

√    √ 

Bolsa Alimentação

Brazil

   √  √ 

FA

Colombia

 √  √  √ 

HIV testing in

Malawi

 

      HIV tested people go back to get their results

Characteristics of outcomes

Health care utilisation is reported as visits to health facilities, which usually constitute one of the conditionalities of the programmes (Attanasio 2005; Gertler 2000; Maluccio 2004; Morris 2004a; Thornton 2006).

Other related outcomes include immunisation coverage, which is reported in four studies (Attanasio 2005; Barham 2005a; Maluccio 2004; Morris 2004a.

Two categories of health outcomes were found among the studies. A first group consisted of objective measures: height, weight, and their corollary measures of height-for-age Z-score, weight-for-age Z-score, haemoglobin value, prevalence of anaemia stunting or wasting. These were reported by Attanasio 2005; Behrman 2005; Gertler 2004a; Maluccio 2004; and Rivera 2004. The second set of health outcomes involved the probability of having reported illness symptoms or having fallen ill in a recall period (Attanasio 2005; Gertler 2000; Gertler 2004a).

None of the studies reported effects on patient health expenditures (although some reported details of other types of household expenditures). No equity outcome was included even though some studies included results broken down by groups which are sometimes used as proxies for socio-economic categorisation (e.g. rural/urban). However, as indicated in the inclusion criteria, these were not within the scope of equity outcomes we had defined.   

Results of the search

The main literature search (for all financing strategies, not only conditional cash transfers) using electronic databases and websites resulted in more than 24,000 references to sift.

We identified 29 papers that were potentially relevant for the review. After further examination, 19 of these studies were excluded. The Characteristics of excluded studies table provides detailed reasons for exclusion. Most studies did not meet our study design criteria: they were primarily descriptive case studies, reviews, modelling or cross-sectional studies. Some studies were excluded on the grounds that their focus was not conditional cash transfers, but in-kind transfers, or non-conditional transfers.

Risk of bias in included studies

Methodological quality of included studies

Accounting for clustering effects in cluster randomised trials

Many of the studies were cluster-randomised trials, whose design and analysis needed to address clustering effects. Behrman 2005; Gertler 2004a; Morris 2004a; and Rivera 2004 all reported having taken clustering into account in their analyses (see Table 4 and Table 1). On the other hand, only Morris 2004a and Rivera 2004 mentioned that clustering was also addressed in the sample size calculation and design. Given the numbers of clusters and participants in most studies, however, it is unlikely that the statistical power of the analysis was seriously affected.

Table 4. Outcome measures and methods
Study ID/ InterventionTypes of outcomesMethods usedComments

Attanasio 2005

Colombia

Health services uptake:

Attendance of preventive care visits by children

Immunisation coverage:

Coverage of DPT vaccination (children)

Health outcomes:

Reported incidence of diarrhoea or respiratory diseases (children) 

Anthropometric or nutritional outcomes:

Height for height for age Z-score

Chronic malnourishment (children)

Estimation of DD estimators with a regression model accounting for clustering effects, controlling for a vector of individual, household and municipal variables.The two health utilisation outcomes are directly linked to the conditionalities of the programme.

Barham 2005a

Progresa

Immunisation coverage:

Coverage of  DPT and Measles vaccination (children)

Estimation of treatment effects (DD) with a regression model accounting for clustering, controlling for a vector of individual, household variables. 

Behrman 2005

Progresa

Anthropometric or nutritional outcomes:

Height increase

Estimation of treatment effect with a child-level fixed effects regression, allowing for clustering, applied to 1998 and 1999 data, controlling for observable differences at baseline (incl. health and nutritional status). 

Gertler 2000

Progresa

Health services uptake:

Daily visits in the nearby health facilities

Health outcomes:

Reported morbidity (children)

Estimation of treatment effects (DD) with regression models accounting for clustering, controlling for a vector of individual and household variables, using 4 waves of surveys (first one being the baseline). Health utilisation outcomes by provider type use the same methods but applied to only 2 survey waves (no baseline, only ‘after’ data). Finally  public clinic visit outcomes use similar models applied to facility data. 

Gertler 2004a

Progresa

Health outcomes:

Reported morbidity (children)

Anthropometric or nutritional outcomes:

Height increase

Prevalence of stunting

Regression models (logistic/linear) controlling for SES variables, using 5 waves of household survey for child morbidity (one before, 4 after) ,and another panel survey from 1998 and 2000 for objective health outcomes; clustering accounted for . The analysis is restricted to ‘eligible’ households only. 
Maluccio 2004

Health services uptake:

Attendance of preventive care visits by children

Immunisation coverage:

Reported up-to-date vaccination schedule (children)

Anthropometric or nutritional outcomes:

Prevalence of stunting, wasting and underweight (children under 5)

Height for Age Z-score (children under 5)

Prevalence of anaemia  

Estimation of treatment effects (DD) with a mixed effects regression model accounting for clustering effects and relating each outcome to intervention groups, time and interactions (+ control for individual and household characteristics).The  study  also included outcomes related to schooling, child labour, total expenditures (not health care expenditures) and expenditures by type of food. These were not included here, although some might be alluded to in the discussion.

Morris 2004a

Honduras

Health services uptake:

Attendance of preventive and prenatal care by women

Attendance of preventive care visits by children

Immunisation coverage:

Coverage for DPT, Measles (children under 3) and tetanus toxoid (mothers)

Estimation of treatment effects (DD) with a mixed effects regression model accounting for clustering effects and relating each outcome to intervention groups, time and interactions (no individual or household characteristics) ;

Results from interviews sometimes corroborated by objective data (clinic cards).

We reported only the results from the “household” intervention (i.e. pure CCT) as the other part of the intervention was only partly implemented, with difficulties. 

Morris 2004b

Brazil

Anthropometric or nutritional outcomes:

Height for Age Z-score (children)

Weight for Age Z-score (children)

Propensity Score matching technique are used to create controls as close as possible to beneficiaries. 

Rivera 2004

Progresa

Anthropometric or nutritional outcomes:

Prevalence of anaemia 

Height increase

Random intercept linear model for height (applied to 1998 and 2000 data) and Generalised Estimating Equation  model  for anaemia (applied to 1999 and 2000 data), both allowing for SES controls and accounting for clustering.

Nutrition supplements were provided along with cash incentives.

No baseline for Hb.

 

Thornton 2006

Malawi

Health services uptake:

Proportion of people who went back to get the results of their tests

Estimation of treatment effect with a regression model relating the outcome to intervention group, incentive amount, distance and other controls.None
Quality of randomisation and implications for the analysis

Some randomised trials did not provide a baseline (Table 4 and Table 1), and the EPOC quality criteria penalise this absence. The usual argument supporting the absence of the baseline is that if the randomisation is done well enough, it eliminates any potential differences between the control and intervention sites at the baseline. Therefore, the differences found after the intervention capture only the impact of the programme.

However, a methodological analysis of Progresa surveys (Behrman 1999) rejected the hypothesis of random assignment of Progresa cash transfers at household levels, despite random assignment at community level. Behrman 2001 further proved that similar problems hampered the nutritional sub-study (INSP surveys used by Behrman 2005; Gertler 2004a; and Rivera 2004). Both socio-economic characteristics and unobserved characteristics of households (e.g. level of concern of parents for their children, health status of children, etc.) may have influenced the eventual benefit received from the programme, and should therefore be accounted for in the analysis. Consequently all analyses of Progresa reporting results at individual level are susceptible to bias if they did not attempt to control for baseline differences.The nutritional sub-study that was done for Progresa to assess its impact on nutritional status took place after the beginning of the programme. We included these data whilst bearing in mind the potential bias stemming from the absence of a baseline. Other reports and surveys on the whole experiment provided enough details to inform potential flaws.

The biased distribution of financial incentives reported in Thornton 2006 also proves that it may be difficult to conform to the necessities of randomisation at all stages of the implementation of such a programme.

Leakage problems

In addition to the non-random assignment across households underlined by Behrman 1999, it was observed that the ‘papilla’ (the nutrition supplement provided by Progresa) was sometimes given to children who were not supposed to receive it (in control localities), and that supply-side bottlenecks led to discretionary choices from local administrators regarding the beneficiaries (Behrman 2005). Results from Rivera 2004 regarding intake of ‘papilla’ suggest that the allocation of the nutrition component of Progresa was far from being systematically followed: not only did they confirm leakage in control zones (with data from an INSP survey), but they also showed that less than 60% of children were actually consuming papilla regularly which may be due to the supply-side shortages mentioned earlier or some failure to comply with this condition from households. These factors would lead to an underestimation of the impact of the nutritional supplements by standard analyses.

This is another argument in favour of an individual-level analysis, as carried out by Behrman 2005, which actually confirms papilla intake (in addition to being a child from an eligible household residing in a treatment community).

Attrition bias

Attrition bias was found in the nutritional surveys done for the Progresa programmes. Attrition problems due to poor quality data and survey design problems are explained in detail in Behrman 2001. The magnitude of the attrition bias is confirmed by Rivera 2004. Although the authors tried to limit attrition bias by using datasets from 2000 and 1998, only 82% of the original cohort was assessed in 2000, and of this subgroup only 75% could be used for the analysis. Behrman 2001 showed that the attrition effect between 1998 and 1999 had resulted in an over-representation in the sample of children with a poor nutritional status. It is likely that a similar phenomenon occurred as a result of the attrition observed between 1998 and 2000. This is partially confirmed by the differences in mean height-for-age Z-score given in the two studies at ‘baseline’: while Behrman 2001 reported a mean of -0.24 for children aged 6-12 months old measured in 1998, Rivera 2004 reported a worse average nutritional status at baseline with a mean difference of -1.06 between the intervention group of children receiving Progresa and the comparison group.

This bias may have led to an over-estimation of the impact of Progresa by Rivera 2004, as individual characteristics were not accounted for and children who were worse off before the intervention benefited more from it. However, Behrman 2001 tried to compensate for these imbalances. 

Synthesis of quality assessment

Two studies present some minor limitations and were deemed as presenting moderate risks of bias (Maluccio 2004; Morris 2004a). All other studies presented high risks of bias after applying the quality criteria. However, the authors of the recent publications were in general aware of the limitations of the studies and made efforts to compensate statistically for potential biases, which were generally due not to poor design but to the fact that the health workers implementing the CCT schemes tried to ensure that the poorest received the benefits of the treatment even if this meant overriding the random assignment at the individual level (Progresa).

Effects of interventions

See: Summary of findings for the main comparison

Impact on uptake of health services

Thornton 2006 reported a positive impact of a financial incentive conditional on getting people who had been tested for HIV to return for their results. Controlling for distance, she found that the proportion of people who went to collect their results increased by 27 percentage points in the intervention group compared to the treatment group. This study also showed that there is no differential impact of monetary incentives according to
their amounts (from US$1 to US$31per result collection), although the numbers receiving the higher payment were limited.

Gertler 2000 showed that in areas where cash transfers were offered to the population, there was an increase of 2.09 in the number of daily outpatient visits to health facilities.  

Based on statistics from health centres in the control and the treatment areas, Morris 2004a found that use of services increased significantly for pre-school children but there was no significant increase in the uptake of antenatal care or 10-day postnatal check-ups (see Table 5).  Based on mothers’ reports, the same programme was found to have a significant impact on the uptake of antenatal care and routine well-child check-ups and growth monitoring visits for children (increased by 18, 19 and 15 percentage points respectively). However, there was no effect on the uptake of the 10-day check-up after delivery. The results on antenatal care uptake are at odds with registries.  

Table 5. Impact on health service utilisation
  1. Note: NP denotes results that were not presented in the articles reviewed. Blank cells denote that outcomes was either not available (eg no baseline date) or that outcomes do not apply.

     95% CI are shown in brackets

    *** indicates significance at the 1%level; ** at the 5% level; and * at the 10% level.

    ¶ results refer to percentage points. The treatment effect represent the net effect, e.g. taking into account the comparison with control groups.

    ¶¶ mean attendance of people without incentives was 0.39 ; treatment effect is estimated with a model controlling for the impact of distance to the VCT centre.

    ¶¶¶computed with surveys carried out after the beginning of the intervention only.

    indicate absolute value of t statistics and ††standard errors.

SourceOutcome description

 Initial outcome

(intervention areas)

Final outcome

(intervention areas)

Relative treatment effect

(difference in outcome measures between intervention and control sites, adjusting for baseline differences - e.g. net variations in percentage points or in the number of visits)

Malawi
Thornton 2006 9% of individuals who attended a VCT centre to learn their results-72¶¶

27.4***

(2.8)

Colombia - Familias en Acción

Attanasio 2005 12, Attanasio 2005b 20

 

 

% of children under 24 months with up-to-date schedule of preventive healthcare visitsNP40.0

22.8**

(0.067)††

% of children aged 24-48 months with up-to-date schedule of preventive healthcare visitsNP.66.8

33.2**

(0.115)††

% of children over 48 months with up-to-date schedule of preventive healthcare visitsNP40.4

1.5*

(0.008)††

Honduras - PRAF

Morris 2004a

 

 

 

% of women having completed more than 5 antenatal care visits37.9NP

18.7***

[7.4 ; 30.0]

% of women attending a 10-day post partum check-up17.8NP.

-5.6

[-015.6 ; 4.5]

% of children taken to a health centre at least once in the past month44.0NP

20.2**

[10.9 ; 29]

Nicaragua - Red de Protección Social

Maluccio 2004

 

 

% of children age 0-3 taken to a health centre at least once in the past 6 months69.892.7

11.0*

(5.9)††

% of children taken to health control and weighed in the past 6 months55.489.1

17.5**

(7.3)††

% of children taken to health control and weighed in the past 6 months - extremely poor groupNPNP

23.6**

(9.3)††

Mexico - Progresa

Gertler 2000

 

 

 

 

 

 

 

 

Number of daily consultations per public clinic in Progresa localities9.1112.84

2.09*

(0.067)††

Number of visits to a public clinic in the 4 weeks preceding the survey - children aged 0-2¶¶¶-0.066

-0.011

(-0.314)

Number of visits to a public clinic in the 4 weeks preceding the survey - children aged 3-5¶¶¶-0.075

0.027

(1.487)

Number of visits to a public clinic in the 4 weeks preceding the survey - children aged 6-17¶¶¶-0.034

0.015

(1.858)

Number of visits to a public clinic in the 4 weeks preceding the survey - adults aged 18-50¶¶¶-0.050

0.015

(1.624)

Number of visits to all facilities in the 4 weeks preceding the survey - children aged 0-2¶¶¶-0.081

-0.032

(-0.871)

Number of visits to all facilities in the 4 weeks preceding the survey - children aged 3-5¶¶¶-0.097

0.027

(1.439)

Number of visits to all facilities in the 4 weeks preceding the survey - children aged 6-17¶¶¶-0.041

0.016

(1.893)

Number of visits to all facilities in the 4 weeks preceding the survey - adults aged 18-50¶¶¶-0.071

0.011

(1.019)

Finally, another study from Nicaragua, displayed a positive impact on health care utilisation, with an increase by 19.5 percentage points after 1 year and 11 after 2 years in the proportion of infants (0-3 years old) taken to health centres in the past 6 months (Maluccio 2004) (see Table 5). The dip in estimated effect between the first and second year is due to an increase in the rates reported in the control group. The price year was not specified although it is probably 2005 (the year that the study was implemented).

Impact on health outcomes

Three studies reported health outcomes, measured as self-reported episode of illness in population surveys. See Table 6 for more details.

Table 6. Impact on health outcomes
  1. Note: NP denotes results that were not presented in the articles reviewed. Blank cells denote that outcomes was either not available (eg no baseline date) or that outcomes do not apply.

    95% CI are shown in brackets

    ¶ log-estimates  of the impact on the probability of illness (e.g. an estimate of 0.75 means that children benefiting from the treatment were 25% less likely than the control ones to be reported as ill) ; the sample was limited to potentially eligible households in treatment and control areas.

    *** indicates significance at the 1% level; ** at the 5% level; and * at the 10% level.

    indicates t statistics, ††p-value and †††Standard errors

SourceOutcome description

 Initial outcome

(intervention areas)

Final outcome

(intervention areas)

Relative treatment effect

(difference in outcome measures between intervention and control sites, adjusting for baseline differences - e.g. net variations in percentage points or probability)

Colombia - Familias en Acción

Attanasio 2005

 

 

 

 

 

 

 

 

 

 

 

Probability of diarrhoea being reported, for children in rural areas, under 24 months oldNPNP

-0.106*

(0.059)†††

Probability of diarrhoea being reported, for children in rural areas, 24-48 months oldNPNP

-0.109**

(0.037)†††

Probability of diarrhoea being reported, for children in rural areas, over 48 months oldNPNP

-0.015

(0.026)†††

Probability of diarrhoea being reported, for children in urban areas, under 24 months oldNPNP

0.150

(0.103)†††

Probability of diarrhoea being reported, for children in urban areas, 24-48 months oldNPNP

-0.033

(0.041)†††

Probability of diarrhoea being reported, for children in urban areas, over 48 months oldNPNP

-0.042

(0.026)†††

Probability of respiratory disease symptoms being reported, for children in rural areas, under 24 months oldNPNP

-0.056

(0.083)†††

Probability of respiratory disease symptoms being reported, for children in rural areas, 24-48 months oldNPNP

-0.005

(0.054)†††

Probability of respiratory disease symptoms being reported, for children in rural areas, over 48 months oldNPNP

-0.012

(0.056)†††

Probability of respiratory disease symptoms being reported, for children in urban areas, under 24 months oldNPNP

-0.094

(0.103)†††

Probability of respiratory disease symptoms being reported, for children in urban areas, 24-48 months oldNPNP

0.034

(0.101)†††

Probability of respiratory disease symptoms being reported, for children in urban areas, over 48 months oldNPNP

-0.010

(0.080)†††

Mexico - Progresa

Gertler 2000

 

% of children  whose mother reported that they were ill in the past 4 weeks - under age 3 at baseline0.402NP

-4.7***

(-2.368)

% of children whose mother reported that they were ill in the past 4 weeks - age 3-5  at baseline0.280NP

-3.2***

(-2.591)

Gertler 2004a

 

 

 

 

 

Likelihood of children (aged under 3 years old at baseline) to be reported ill in the past 4 weeks - global impact ¶--

0.777***

(0.000)††

Likelihood of children (aged under 3 years old at baseline) to be reported ill in the past 4 weeks - impact after 2 months of programme¶--

0.940

(0.240)††

 

Likelihood of children (aged under 3 years old at baseline) to be reported ill in the past 4 weeks - impact after 8 months of programme¶--

0.749***

(0.000)††

Likelihood of children (aged under 3 years old at baseline) to be reported ill in the past 4 weeks - impact after 14 months of programme¶--

0.836***

(0.005)††

Likelihood of children (aged under 3 years old at baseline) to be reported ill in the past 4 weeks - impact after 20 months of programme¶¶--

0.605***

(0.000)††

Likelihood of children (aged under 3 years old at baselin) to be reported ill in the past 4 weeks - global impact ¶--

0.747**

(0.013)††

Attanasio 2005 reported mixed results regarding the impact on the probability of children suffering from diarrhoea. While Familias in Accion seems to have reduced the probability of reported diarrhoea symptoms for children aged under 48 months living in rural areas, older groups did not display any changes. The study also failed to detect any effect on the probability of respiratory symptoms being reported by children.

The analysis of a Nicaraguan programme by Maluccio 2004 found it did not have an impact on anaemia or mean haemoglobin among infants aged 6 to 59 months old.

The analyses performed by Gertler 2004a concluded that Progresa led to a 22% decrease in the probability of children younger than 3 years of age having been ill in the past month. It also showed that the longer the children have received the programme, the greater that beneficial effect.

In summary, available existing evidence shows that CCT programmes can have a positive impact on children’s health outcomes, but this is neither systematic nor consistent across all age groups.

Impact on immunisation coverage

Four studies reported effects on immunisation coverage. See Table 7 for more details.

Table 7. Impact on immunisation coverage
  1. Note: NP denotes results that were not presented in the articles reviewed. Blank cells denote that outcomes was either not available (eg no baseline date) or that outcomes do not apply.

    95% CI are shown in brackets

    *** indicates significance at the 1%level; ** at the 5% level; and * at the 10% level.

    ¶ results refer to percentage points for proportion and point estimates for scores. The treatment effect represent the net effect, e.g. taking into account the comparison with control groups.

    indicates standard errors in parentheses and ††absolute value of t statistics

SourceOutcome description

 Initial outcome

(intervention areas)

Final outcome

(intervention areas)

Relative treatment effect

(difference in outcome measures between intervention and control sites, adjusting for baseline differences - e.g. net variations in percentage points or probability)

Colombia - Familias en Acción

Attanasio 2005

 

 

Probability  of compliance with DPT vaccination, for children under 24 months oldNPNP

8.9*

(0.047)

Probability  of compliance with DPT vaccination, for children 24-48 months oldNPNP

3.5

(0.026)

Probability  of compliance with DPT vaccination, for children, over 48 months oldNPNP

3.2

(0.039)

Honduras - PRAF

Morris 2004a

 

 

 

% of children under age 3 vaccinated with DPT1/pentavalent72NP

6.9***

[1; 12.8]

% of children under age 3 vaccinated for Measles84NP

-0.2

[-9.4 ; 9.0]

%of mothers vaccinated for tetanus toxoid56NP

4.2

[-9.7 ; 18.2]

Nicaragua - Red de Protección Social

Maluccio 2004

 

% of children aged 12-23 months old with up-to-date vaccinations36.471.7

6.1

(10.2)

Mexico - Progresa

Barham 2005a

Evolution after 6 months

% of children under 12 months old (at baseline) vaccinated for TB8889

5.2***

(2.07)††

% of children aged 12-23 months old (at baseline) vaccinated for Measles9296

3.0**

(2.03)††

Impact after 12 months% of children under 12 months old (at baseline) vaccinated for TB8892

1.6

(0.66)††

% of children aged 12-23 months old (at baseline) vaccinated for Measles9291

2.8

(1.00)††

Barham 2005a reported mixed results of Progresa on immunisation coverage. The difference-in-difference estimators in OLS regressions showed a difference of 5 percentage points in TB immunisation coverage for Progresa children aged 12 to 23 months old (baseline of May 1998), 6 months after the beginning of the intervention. However, this was due to a decrease in coverage in control zones, and the difference was no longer significant once the ‘control’ children recovered from the drop 12 months after baseline. Measles vaccination increased by 3 percentage points for children aged 12 to 23 months old 6 months after the beginning of the programme, and by 6 percentage points after 12 months in low coverage villages.

Results from the study in Honduras (Morris 2004a) showed an increase of 6.9% in the coverage of the first dose DTP/pentavalent vaccine among children but not in tetanus immunisation among pregnant women, nor in measles vaccination among children. Familias en Accion in Colombia also increased the probability that 24-month-old children had complied with DPT vaccination schedule (Attanasio 2005).

The Red de Proteccion Social (RPS) programme in Nicaragua had no significant impact on vaccination coverage. However it seems that this was mainly due to a concurrent increase in vaccination coverage in both the intervention and the control areas due to, which both benefited from supply-side incentives strengthening the procurement of vaccines.  

Impact on anthropometric or nutritional outcomes

Six papers reported outcomes on anthropometric measures and nutritional status from four different CCT programmes.

The results obtained by Attanasio 2005 on the short-term impact of Familias in Accion in Colombia showed mixed conclusions about the impact of monetary transfers on the nutritional status of children. They show a positive impact on nutritional status for children under 24 months (see Table 8), and an increase of 0.58 kg in newborn weight in the urban areas of treatment localities. However, no impact was detected on the nutritional status of children older than 24 months, or on newborn weight in rural areas.

Table 8. Impact on anthropometric and nutritional outcomes
  1. Note: NP denotes results that were not presented in the articles reviewed. Blank cells denote that outcomes were either not available (eg no baseline date) or that outcomes do not apply.

    95% CI are shown in brackets

    *** indicates significance at the 1% level; ** at the 5% level; and * at the 10% level.

    ¶ results refer to percentage points for proportion and point estimates for scores. The treatment effect represent the net effect, e.g. taking into account the comparison with control groups.

    ¶¶  The intervention group include children aged under 6 months old at baseline (Aug 1998) and exposed to 2 years of Progresa, while the control group is a crossover group (e.g. it includes children without treatment for a year and then exposed to 1 year of Progresa when they are 12-18 months old and after)

    ¶¶¶  the difference was computed by the reviewers, using data from control and intervention groups from the article; statistical significance of the difference was computed by the authors of the article.

    ¶¶¶¶ log-estimates  of the impact on the probability of illness (e.g. an estimate of 0.75 means that children benefiting from the treatment were 25% less likely than the control ones to be affected)

    indicates absolute value of t statistics, ††p-value and †††standard errors

SourceOutcome description

Initial outcome

(intervention areas)

Final outcome

(intervention areas)

Relative treatment effect

(difference in outcome measures between intervention and control sites, adjusting for baseline differences: eg. net variations in percentage points or scores)

Colombia - Familias en Acción

Attanasio 2005

 

 

 

 

 

Height-for-Age Z-score of children under 24 months oldNPNP

0.161*

(0.085)†††

Height-for-Age Z-score of children aged 24-48 monthsNPNP

0.011

(0.055)†††

Height-for-Age Z-score of children over 48 months oldNPNP

0.012

(0.033)†††

Probability of chronic malnourishment for children under 24 months oldNPNP

-0.069**

(0.034)†††

Probability of chronic malnourishment for children aged 24-48 monthsNPNP

0.004

(0.022)†††

Probability of chronic malnourishment for children over 48 months oldNPNP

-0.021

(0.014)†††

Nicaragua - Red de Protección Social  (evolution 2000-2002)

Maluccio 2004

 

 

 

 

 

 

Height-for-Age Z score for children under 5

 

-1.79-1.65

0.17**

(0.08)

% of children under age 5 who are stunted

 

41.9

 

37.1

-5.3*

(3.1)

% of children under age 5 who are underweight15.310.4

-6.0**

(2.6)

% of children under age 5 who are wasted

 

1.00.4

-0.4

(0.5)

Hemoglobin for children 6-to-59 months of age

 

11.211.4

-0.1

(0.2)

% of children 6-to-59 months of age with anemia

 

33.732.8

-0.2

(6.8)

Brazil - Bolsa Alimentação 

Morris 2004b

 

 

 

 

 

 

 

 

 

Height-for-Age Z score for children under 24 months old--0.68

-0.25

(± 0.13)†††

Height-for-Age Z score for children under 24-48 months old--0.75

-0.11

(± 0.10)†††

Height-for-Age Z score for children aged 4-7 years old--0.77

-0.08

(± 0.08)†††

Mean Height-for-Age Z score for children under 7 years old--0.75

-0.13**

(± 0.06)†††

Weight-for-Age Z score for children under 24 months old--0.90

-0.11

(± 0.13)†††

Weight-for-Age Z score for children under 24-48 months old--0.85

-0.19

(± 0.11)†††

Weight-for-Age Z score for children aged 4-7 years old--0.95

-0.04

(± 0.09)†††

Mean Weight-for-Age Z score for children under 7 years old--0.90

-0.11

(± 0.06)†††

Mexico - Progresa

Rivera 2004

 

 

 

 

 

Growth (cm) of children aged under 6months old (at baseline), from poorest households¶¶-26.4

1.1**

(0.046)††

Growth (cm) of children aged 6-12 months old (at baseline), from poorest households¶¶-19.7

-0.6

NS

Mean hemoglobin (g/dL) value among children (after a year of Progresa vs. no exposure in the control group) 11.12

0.37**

(0.01)††

Prevalence (%) of anemia (after a year of Progresa vs. no exposure in the control group)-44.3

10.6**

(0.03)††

Prevalence (%) of anemia (after 2 years of Progresa vs. 1 year in the control group)-25.8

-2.8

(0.40)††

Behrman 2005

 

 

 

Height of children (cm) aged between 4-12 months old (at baseline in Aug. 1998)--

0.503

(0.96)

Height of children (cm) aged between 12-36 months old  (at baseline in Aug. 1998)--

1.016**

(2.55)

Height of children (cm) aged between 24-36 months old  (at baseline in Aug. 1998)--

1.224**

(2.05)

Height of children (cm) aged between 36-48 months old (at baseline in Aug. 1998)--

-0.349

(0.66)

Gertler 2000

 

Height (in cm) of children aged 12-36 months old (in Sept 1999)-80.7

0.959***

(0.004)††

Likelihood of children aged 12-36 months old (in Sept 1999) to be stunted ¶¶¶¶-NP

0.914

(0.495)††

The analysis of a Nicaraguan programme by Maluccio 2004 showed positive effects. It reduced the magnitude of stunting (net average improvement of the height-for-age score by 0.17) and the proportion of under-weight children aged 0 to 5 years old (a net impact of 6 percentage points after 2 years). However, it did not have an impact on the proportion of wasted children aged 0 to 5 years old.

The evaluation of the Brazilian programme (Morris 2004b) showed no effect on height-for-age measures and even a negative impact on weight-for-age for children under 7 years old (see discussion).

Finally, three papers reported findings for the Mexican programme, using different groups of reference for their analyses.

Rivera 2004 provided results mainly comparing a group having received Progresa for 2 years against a “crossover” one that received it for one year only. Their analysis showed a significant impact on growth for the youngest children in the poorest households (aged less than 6 months old at baseline in 1998): these infants gained 1.1 cm more after 2 years of the Progresa programme than those in the “crossover” group. However, they found no difference for older children (aged 6 to 12 months at baseline) or for the youngest coming from less poor families. The authors also mentioned that, based on results from the 1999 survey, anaemia prevalence among children who have received Progresa for a year (44.3%) is significantly inferior to that among the “crossover” group (54.9%). In contrast, this difference had disappeared once the “crossover” group had entered the programme for a year.

Using linear regression models applied to post-intervention data only, Gertler 2004a reported a similar result on height gain. They found that beneficiary children aged 12 to 36 months in October to December 1999 were 0.96 cm taller than other children, and were 25% less likely to be anaemic. No difference in stunting was detected between treatment and control groups.

Behrman 2005 also found equivalent results on child growth with a different method, controlling for several sources of bias. Their findings showed a positive effect of Progresa on the height of children 12 to 36 months old: they indicated a growth gain of 1.02 cm more than children in the same age range who did not receive Progresa. They further showed that the programme appeared to have a significant effect for children aged 24 to 36 months (through a height increase of 1.22 cm) but not for other age groups.

Discussion

Despite its widely acclaimed design, we found that the Progresa trial was undermined by a number of methodological issues that hampered the interpretation of its results. However, it was certainly a milestone that influenced the implementation and evaluation of many other programmes. Unlike other financing schemes, conditional cash transfer programmes have in general been evaluated by well-designed and executed evaluations, compared for example with user fees where the quality of studies was much poorer (Lagarde 2006). No less than 4 out of the 6 evaluated programmes we included had been designed to be evaluated by randomised trials. The Progresa data were analysed by different groups using different approaches and sometimes failing to reference each others’ publications. Multiple statistical comparisons were undertaken without adjustment and this could have led to spurious ‘statistically significant differences’. In addition, some conclusions were based on sub-group analyses (e.g. the effects of nutritional interventions at different ages). This again can lead to spurious ‘significant’ results. However, when weaknesses or bias arose, evaluators sometimes made strenuous efforts to correct them or to account for confounding factors (Attanasio 2005; Behrman 2005; Gertler 2004a; Maluccio 2004).  We therefore concluded that, despite the remaining problems, the overall risk of bias is relatively moderate, particularly in light of the consistent effects in a number of different settings and especially compared to the evaluation of the effectiveness of other mechanisms.  Overall this body of evidence finds that conditional cash transfers can be effective means to increase health service utilisation, health outcomes and nutritional status of children, although the significance and size of effects varies (see Summary of findings for the main comparison).

CCTs have been widely introduced as pro-poor policies, in particular because in many Latin American countries they only target the poorest groups of the population. This might be one of the reasons why equity of the programmes were not particularly studies (we did not find adequate detailed analysis of outcome measures by socio-economic groups, only some hints at results per sub-groups reported here). However, a couple of studies alluded to findings per sub-groups, and reported mixed results. In Nicaragua, the study reported that the increase in household expenditures was greatest for the poorest group as was the uptake of preventive services for infants (Maluccio 2004). However, nutritional benefits drawn from Progresa were greater for children whose mother had more than five years of schooling (Behrman 2001), which can be used as proxy to distinguish more disadvantaged groups.

There might be a danger that unanticipated perverse effects may occur, as illustrated by the  study by Morris 2004b where some unexpected decrease in health outcomes amongst children  have been explained by a seeming misunderstanding of the requirements by mothers, who would have kept their child malnourished in order to retain eligibility for the programme. Considering that this study was not a pure C-RCT design, it is also possible that this effect was subject to bias, and it should not necessarily be trusted. However, this underlines an issue that has been highlighted in the literature on incentives, where gaming strategies and unanticipated consequences have often been observed (Courty 2004; Propper 2003).

The good functioning of CCTs relies on different elements. First, in many countries, it has relied on the efficient targeting of poorer groups. However, this is not necessarily a requirement for a Conditional Cash Transfer Programme. For example, some south Asian countries have recently started to offer cash incentives to all pregnant women who would go and deliver in a health facility.  Another element that is key to the good functioning of CCTs is the capacity of the programme implementer to monitor whether the requirements are met or not by the beneficiaries. To do this, it is essential to define as requirements some behaviours or actions that are easily controlled. The size and timing of the incentive is another dimension of the CCT programmes. Because they were initially conceived as social transfer programmes, incentives in many Latin American countries were calculated in relation to the poverty level. In some more recent programmes (not reviewed here), they are being designed in relation to the financial costs (indirect or direct) linked to accessing the health care services (Ministry of Health and Family Welfare 2005; Nepal 2005). Finally, CCT programmes assume that the provision of the health services are adequate, and ready to address the potential increase in the demand of services induced by the cash transfer programme. In fact, some programmes in Latin America (Maluccio 2004) also introduced some supply-side interventions to strengthen the delivery of health services in intervention areas.

In theory, the impact of Conditional Cash transfer programmes can be altered by a number of factors.

The success of conditional cash transfers is probably dependent on the magnitude of the barriers to accessing services on the demand-side. If the main reasons for poor uptake of health services are linked to financial barriers, then CCTs are likely to be effective mechanisms. However if there are few obstacles impeding access to care, as demonstrated for example by high levels of uptake of health services, CCTs will be less, if at all, successful. This is what the experiences in Latin America suggest, where CCTs failed to have an impact on immunisation rates, when the rates were initially quite high. 

Similarly, if the obstacles to health care utilisation by the population are on the supply side (lack of drugs, low density of facilities) CCTs will be less effective.

In fact, the quality and availability of health services is probably a pre-requisite to the success of CCT. There is ample evidence in the health services literature of households avoiding health services for their poor quality. It is likely that even financial incentives would not be sufficient (nor necessarily recommended) to encourage the use of poor quality services.

Secondly, the relative size of effects of CCT programmes is certainly linked to the nature and types of requirement required. For example, some programmes asked the beneficiaries to attend nutrition and health education workshops as part of the requirements, while others did not. It is likely that attending such workshops could have an impact on health behaviours amongst households, thereby influencing health outcomes.

Finally, the level of the financial incentive used might modify the size of the effect. It is likely that the larger the financial incentive the greater the likelihood of individuals to comply with the requirement, hence the more likely the entire targeted population will be reached.

The lack of empirical evidence for these issues, which are key to understanding under which conditions CCTs might work more or less efficiently, continues to shape the agenda for future research.

Authors' conclusions

Implications for practice

The CCT programmes implemented so far have not been targeted at reducing barriers to access to curative health services, as their health focus was on preventive services or health promotion, or both. Based on the evidence reviewed here, it seems that CCT programmes can be effective strategies to increase the uptake of preventive services which were already free. Their implementation in a context where services are not free remains to be tested.

Some results also suggest the limitations of CCT to achieve some results. For example, even with important financial incentives, some CCT programmes failed to improve vaccination coverage. It is essential for policy-makers, before embarking on a CCT programme that might be very costly, to analyse carefully the various barriers to health services faced by the population. If supply-side barriers (e.g. lack of vaccines or drugs) are responsible for a low uptake of services, CCTs are unlikely to provide a relevant solution, or at least not the only one.

Despite the highly publicised success of CCT programmes, at least partly confirmed by this review, several questions remain regarding their feasibility in poorer settings, and policy-makers in such environments should be aware of a number of issues. First, policy-makers willing to introduce CCTs should probably ensure a minimum quality on the supply side, so that the intervention can effectively address the demand-side obstacles. Most successful CCT programmes reported in this review have been implemented in middle-income countries, which have relatively well-functioning health systems. Second, it is likely that the success of existing programmes has relied on effective mechanisms to target and monitor beneficiaries, as well as to transfer the money in a timely fashion.  There are several examples of strategies (e.g. exemption policies) in which such elements have failed to work correctly in low-income settings, and it is important to acknowledge their importance in the success of CCTs. Finally, policy makers should carefully study the cost implications of CCT programmes, in particular if no targeting mechanism is put in place. Indeed, some have shown that not targeting the groups who have the least access to health services will increase the marginal cost per person covered (Lagarde 2007), and therefore increase the opportunity cost of CCT programmes.

Implications for research

The first area for further research on CCT relates to the question of their cost-effectiveness. Given the financial constraints of most Sub-Saharan African countries, providing schools and health care facilities may be a more effective allocation of public spending than cash transfers. There has not been any study in the CCT literature that tried to address that issue. A careful analysis of the costs and benefits of these programmes versus other traditional delivery of health care services is urgently needed.

A second area of research on the effects of conditional cash transfers relates to a better understanding of the different pathways through which CCTs work. The relatively well-designed evaluations we presented do not necessarily explain why such monetary incentives are working, and in particular whether they effectively help overcome all financial barriers (would they have worked with non-free services?) and/or other types of obstacles such as cultural barriers, e.g. access to child and maternal health services in cultures where access is controlled by male members of the family.

Furthermore, so far, the positive impact of CCT programmes is generally linked only to the presence of the financial incentives. But there are other potential reasons explaining why health outcomes have increased. As underlined by Gertler 2004a, the multiple components of the programmes may play a role, and their respective weight and role has so far not been isolated from that of the financial incentive. It would be interesting to know for example, if nutritional outcomes amongst children have been improved by the wealth effect of the financial bonus (allowing to buy more and better food as it was proved), or if these were improved by attendance of health and nutrition workshops in some programmes and by nutritional supplements provided to some children in other programmes.

A final area for future research is the issues of the relative effect of CCTs for different levels of incentives or different socio-economic groups. So far, only one of the evaluated programmes varied the size of cash transfers (Thornton 2006), but it did not assess the relative impact of those levels for different socio-economic groups. A better knowledge of the presence and the existence of threshold effects, and the potential marginal positive effects of the cash transfer for various income groups is needed.

Acknowledgements

We gratefully acknowledge:

  • the Bill and Melinda Gates Foundation for funding this work;

  • Andy Oxman, Jessie McGowan and two anonymous referees for their useful comments on the protocol;

  • Andy Oxman, Elizabeth Paulsen, Luke Vale and three anonymous reviewers for their support and comments on the review;

  • Sandra Russell for her help in retrieving and copying of papers.

Data and analyses

Download statistical data

This review has no analyses.

Appendices

Appendix 1. Search strategy used for Pubmed

The search in PubMed was also restricted to all the developing countries listed on the World Bank website, by selecting all relevant geographical categories as exploded terms. 

Some pilot searches led us to use quite general (exploded) MeSH terms, as it was noticed that several relevant articles were indexed with generic MeSH terms, or not particularly appropriate ones. For example, a study on Ghana would not be referenced under “Ghana” but under “Africa”. Besides, since including “Africa[MeSH]” would also include all MeSH terms of lower levels, it was decide to include mainly higher level MeSh terms for delimiting the geographic scope of the study (see #1 below). A few countries were excluded (see #6).

A similar approach was taken for specifying the topic filters of the search. Generic MeSh terms were used (see #2), and more selective terms that are currently used in the literature were added as free text references (see #3). However, because this was potentially return a large number of irrelevant studies, it was decided to limit this by excluding some irrelevant studies (see #4).

These different filters were then rearranged together (see #7, #8 and #9).

1Search "Developing countries"[MeSH] OR "Africa"[MeSH] OR "Central America"[MeSH] OR "South America"[MeSH] OR "Latin America"[MeSH] OR "Mexico"[MeSH] OR "Asia"[MeSH] OR "Commonwealth of Independent States"[MeSH] OR "Pacific Islands"[MeSH] OR "Indian Ocean Islands"[MeSH] OR "Europe, Eastern"[MeSH]
2Search ("Economics"[MeSH] OR "Economics"[SH] OR "socioeconomic factors"[MeSH]) AND ("Delivery of health care"[MeSH] OR "health services research"[MeSH] OR "health planning"[MeSH] OR "health services "[MeSH] OR "utilization"[SH])
3Search "Fees and charges"[MeSH] OR user fee[TIAB] OR user fees[TIAB] OR social insurance[TIAB] OR health insurance[TIAB] OR community-based insurance[TIAB] OR prepayment plan[TIAB] OR prepayment plans[TIAB] OR prepayment scheme[TIAB] OR prepayment schemes[TIAB] OR conditional cash transfers[TIAB] OR cost recovery[TIAB] OR prepayment[TIAB] OR contracting out [TIAB] OR output-based contract[TIAB] OR pay for performance [TIAB]
4Search "Personnel Downsizing"[MeSH] OR "workplace"[MeSH] OR "health planning guidelines"[MeSH] OR "patient freedom of choice laws "[MeSH] OR "preferred provider organizations"[MeSH] OR "provider-sponsored organizations"[MeSH] OR "emergency Medical Service Communication Systems"[MeSH] OR "Genetic Services"[MeSH] OR "Medical Errors"[MeSH] OR Chemicals and Drugs Category[MAJR] OR "Drug industry"[MAJR] OR "epidemiology"[MAJR] OR "Patents"[MAJR] OR "War"[MAJR] OR Anatomy Category[MAJR] OR "Child Abuse"[MeSH] OR ("Technology and Food and Beverages Category"[MAJR] NOT "food supply"[MeSH])
5Search Practice Guideline[ptyp] OR Letter[ptyp] OR Editorial[ptyp] "Clinical Trial"[ptyp] OR "Clinical Trial, Phase I"[ptyp] OR "Clinical Trial, Phase II"[ptyp] OR "Clinical Trial, Phase III"[ptyp] OR "Clinical Trial, Phase IV"[ptyp]
6Search "Japan"[MeSH] OR "Korea"[MeSH] OR "Taiwan"[MeSH] OR "New Zealand"[MeSH] OR "Singapore"[MeSH] OR "Israel"[MeSH]
7Search #1 AND #2 NOT #4 NOT #5 NOT #6
8Search #1 AND #3 NOT #4 NOT #5 NOT #6
9Search #8 OR #7

 

Ovid MEDLINE(R) 1950 to April Week 4 2009

Searched 05.05.2009

 

1. "Fees and Charges"/

2. Fees, Dental/

3. Fees, Medical/

4. Fees, Pharmaceutical/

5. Prescription Fees/

6. Hospital Charges/

7. Capitation Fee/

8. Fee-for-Service Plans/

9. "Cost Sharing"/

10. Contract Services/

11. Outsourced Services/

12. Prepaid Health Plans/

13. Prospective Payment System/

14. Insurance, Health/

15. ((medical or dental or pharmac$ or dispensing or drug or drugs or medicament? or medicine? or prescript$ or consultation? or treatment? or registration? or hospital? or care) adj3 (fee? or charge?)).tw.

16. ((user? or patient? or outpatient? or inpatient?) adj3 (fee? or charge? or pay$)).tw.

17. fee for service?.tw.

18. capitation.tw.

19. ((pay$ or cash or money or monetary or economic or financial) adj3 incentive?).tw.

20. (pay$ adj3 performance).tw.

21. p4p.tw.

22. ((result? or performance) adj based).tw.

23. ((result? or performance or output or out put) adj2 (financ$ or pay$ or incentive? or initiative? or bonus$)).tw.

24. ((cash or pay$) adj3 (condition$ or contingent or requirement?)).tw.

25. ((cash or pay$ or monetary ot money) adj3 transfer$).tw.

26. cost sharing.tw.

27. cost recover$.tw.

28. price change?.tw.

29. (contract or contracts or contracting).tw.

30. (outsourc$ or out sourc$).tw.

31. (risk sharing or shared risk?).tw.

32. (prospective adj (pay$ or reimbursement?)).tw.

33. (prepay$ or pre pay$ or prepaid or pre paid).tw.

34. ((health or medical) adj insurance?).tw.

35. ((social or community) adj3 (insurance? or financ$)).tw.

36. demand side.tw.

37. supply side.tw.

38. (financ$ adj (strategy or strategies)).tw.

39. or/1-38

40. Developing Countries/

41. Medically Underserved Area/

42. exp Africa/ or exp "Africa South of the Sahara"/ or exp Asia/ or exp South America/ or exp Latin America/ or exp Central America/

43. (Africa or Asia or South America or Latin America or Central America).tw.

44. (American Samoa or Argentina or Belize or Botswana or Brazil or Bulgaria or Chile or Comoros or Costa Rica or Croatia or Dominica or Equatorial Guinea or Gabon or Grenada or Hungary or Kazakhstan or Latvia or Lebanon or Libya or Lithuania or Malaysia or Mauritius or Mexico or Micronesia or Montenegro or Oman or Palau or Panama or Poland or Romania or Russia or Seychelles or Slovakia or South Africa or "Saint Kitts and Nevis" or Saint Lucia or "Saint Vincent and the Grenadines" or Turkey or Uruguay or Venezuela or Yugoslavia).sh,tw. or Guinea.tw. or Libia.tw. or libyan.tw. or Mayotte.tw. or Northern Mariana Islands.tw. or Russian Federation.tw. or Samoa.tw. or Serbia.tw. or Slovak Republic.tw. or "St Kitts and Nevis".tw. or St Lucia.tw. or "St Vincent and the Grenadines".tw.

45. (Albania or Algeria or Angola or Armenia or Azerbaijan or Belarus or Bhutan or Bolivia or "Bosnia and Herzegovina" or Cameroon or China or Colombia or Congo or Cuba or Djibouti or Dominican Republic or Ecuador or Egypt or El Salvador or Fiji or "Georgia (Republic)" or Guam or Guatemala or Guyana or Honduras or Indian Ocean Islands or Indonesia or Iran or Iraq or Jamaica or Jordan or Lesotho or "Macedonia (Republic)" or Marshall Islands or Micronesia or Middle East or Moldova or Morocco or Namibia or Nicaragua or Paraguay or Peru or Philippines or Samoa or Sri Lanka or Suriname or Swaziland or Syria or Thailand or Tonga or Tunisia or Turkmenistan or Ukraine or Vanuatu).sh,tw. or Bosnia.tw. or Cape Verde.tw. or Gaza.tw. or Georgia.tw. or Kiribati.tw. or Macedonia.tw. or Maldives.tw. or Marshall Islands.tw. or Palestine.tw. or Syrian Arab Republic.tw. or West Bank.tw.

46. (Afghanistan or Bangladesh or Benin or Burkina Faso or Burundi or Cambodia or Central African Republic or Chad or Comoros or "Democratic Republic of the Congo" or Cote d'Ivoire or Eritrea or Ethiopia or Gambia or Ghana or Guinea or Guinea-Bissau or Haiti or India or Kenya or Korea or Kyrgyzstan or Laos or Liberia or Madagascar or Malawi or Mali or Mauritania or Melanesia or Mongolia or Mozambique or Myanmar or Nepal or Niger or Nigeria or Pakistan or Papua New Guinea or Rwanda or Senegal or Sierra Leone or Somalia or Sudan or Tajikistan or Tanzania or East Timor or Togo or Uganda or Uzbekistan or Vietnam or Yemen or Zambia or Zimbabwe).sh,tw. or Burma.tw. or Congo.tw. or Kyrgyz.tw. or Lao.tw. or North Korea.tw. or Salomon Islands.tw. or Sao Tome.tw. or Timor.tw. or Viet Nam.tw.

47. ((developing or less$ developed or third world or under developed or middle income or low income or underserved or under served or deprived or poor$) adj (count$ or nation? or state? or population?)).tw.

48. (lmic or lmics).tw.

49. or/40-48

50. randomized controlled trial.pt.

51. random$.tw.

52. intervention$.tw.

53. control$.tw.

54. evaluat$.tw.

55. effect?.tw.

56. or/50-55

57. Animals/

58. Humans/

59. 57 not (57 and 58)

60. 56 not 59

61. 39 and 49 and 60

Appendix 2. Quality criteria used for appraising quality of included studies

This appendix presents the detail of all of the criteria used in the appraisal of included studies.

CBA studies:

In the following list, criteria one, two and four are directly taken from the list of standard criteria of the EPOC Group.

Criteria three and five are adapted from the original criteria to make them more relevant to the specificities of the studies included in this review. Standards to judge the risk of exclusion or selection bias were rephrased to be more adapted to the types of population-based studies that might be included in the review. The criterion on quality and reliability of data was also adapted to reflect better the risks of bias relating to the type of outcomes that were the primary focus of the review.

Criteria six was added following preliminary findings which showed that statistical significance of studies was not systematically computed or available in the studies found.

Finally, we omitted a standard criterion of the Cochrane Collaboration textbook on the blinded assessment of primary outcomes. We judged that this was not relevant for the types of outcomes this review focused on.

  1. Baseline outcome characteristics: DONE if outcomes were measured prior to the intervention, and no significant differences were present across study groups (e.g. where multiple pre intervention measures describe similar trends in intervention and control groups); NOT CLEAR if baseline measures are not reported, or if it is unclear whether baseline measures are significantly different across study groups; NOT DONE if there are differences at baseline in main outcome measures likely to undermine the post intervention differences (e.g. are differences between the groups before the intervention similar to those found post intervention?)

  2. Equivalent control sites: DONE if characteristics of study and control sites are reported and similar (in terms of 1/population 2/facilities and 3/external influence characteristics); NOT CLEAR if it is not clear in the paper e.g. characteristics are mentioned in the text but no data are presented; NOT DONE if there is no report of characteristics either in the text or a table OR if baseline characteristics are reported and there are differences between study and control providers.

  3. Protection against exclusion or selection bias: DONE if outcome measures obtained from the whole population or a representative sample of the population (and the control group) was studied; NOT CLEAR if not specified in the paper; NOT DONE if outcome measures were not obtained from a representative sample.

  4. Protection against contamination: DONE if allocation was by community, institution, or practice and is unlikely that the control group received the intervention; NOT CLEAR if communication (i.e  individuals present in one control group cannot move and benefit from the interventions in experimental areas) between treatment and control group was likely to occur; NOT DONE if it is likely that the control group received the intervention (e.g. cross-over studies or if patients rather than providers were randomised).

  5. Quality/reliability of outcome measures: scored DONE if the outcome is obtained from some automated system (e.g. length of hospital stay) or comes from another objective source; NOT CLEAR if reliability is not reported for outcome measures that are obtained by chart extraction or collected by an individual (will be treated as NOT DONE if information cannot be obtained from the authors); and NOT DONE if the primary data is reportedly of a poor quality.

  6. Appropriate analysis: DONE if statistical significance of differences in outcomes was tested and/or statistical analysis was appropriate. NOT CLEAR if statistical significance of results is not specified in the paper or if the analysis chosen was not appropriate; NOT DONE if statistical significance of results was not tested.

 Randomised Controlled Trials

All the following criteria are taken from the standard EPOC criteria (EPOC 2002), except for criteria three and four. Indeed, we judged important to add specific criteria for cluster-randomised for two reasons. Firstly because interventions of interest would be more likely to be implemented at community level, they would require such study designs. Secondly, issues regarding sampling and analysis have identified as particular concerns that might lead to biases when analysing cluster-randomised trials (Ukoumunne 1999). We also omitted one criteria on exclusion bias concerning the follow-up of professionals. It was judged not relevant for the focus of our review (where studies are all focusing on populations).

  1. Concealment of allocation: DONE if the unit of allocation was by institution, team or professional and any random process is described explicitly, e.g. the use of random number tables or coin flips; OR the unit of allocation was by patient or episode of care and there was some form of centralised randomisation scheme, an on-site computer system or sealed opaque envelopes were used. NOT CLEAR if the unit of allocation is not described explicitly OR the unit of allocation was by patient or episode of care and the authors report using a ‘list’ or ‘table’, ‘envelopes’ or ‘sealed envelopes’ for allocation. NOT DONE if the authors report using alternation such as reference to case record numbers, dates of birth, day of the week or any other such approach (as in CCTs) OR the unit of allocation was by patient or episode of care and the authors report using any allocation process that is entirely transparent before assignment such as an open list of random numbers or assignments OR allocation was altered (by investigators, professionals or patients).

  2. Protection against exclusion bias: DONE if outcome measures obtained for 80-100% of subjects randomised (or a biased sample) or for patients who entered the trial (do not assume 100% follow up unless stated explicitly); NOT CLEAR if not specified in the paper; NOT DONE if outcome measures obtained for less than 80% of subjects randomised (or a biased, non-representative sample).

  3. Sampling (for cluster-randomised trials): DONE if sampling took cluster effects/bias into account or if the sample is large enough to provide robust results; NOT CLEAR if not specified in the paper; NOT DONE if the sampling is too small to provide robust results.

  4. Appropriate Analysis (for cluster-randomised trials): DONE if the analysis accounted for cluster effects/bias; NOT CLEAR if not specified in the paper; NOT DONE if the analysis did not account for cluster effects/bias.

  5. Quality/reliability of the data: scored DONE if the outcome is obtained from some automated system (e.g. length of hospital stay) or comes from another objective source; NOT CLEAR if reliability is not reported for outcome measures that are obtained by chart extraction or collected by an individual (will be treated as NOT DONE if information cannot be obtained from the authors); and NOT DONE if the primary data is reportedly of a poor quality.

  6. Protection against detection bias: DONE if the authors state explicitly that the primary outcome variables were assessed blindly OR the outcome variables are objective, e.g. length of hospital stay, drug levels as assessed by a standardised test; NOT CLEAR if not specified in the paper; NOT DONE if the outcome(s) were not assessed blindly.

  7. Baseline Measurement: DONE if performance or patient outcomes were measured prior to the intervention, and no substantial differences were present across study groups (e.g. where multiple pre intervention measures describe similar trends in intervention and control groups); NOT CLEAR if baseline measures are not reported, or if it is unclear whether baseline measures are substantially different across study groups; NOT DONE if there are differences at baseline in main outcome measures likely to undermine the post intervention differences (e.g. are differences between the groups before the intervention similar to those found post intervention?).

  8. Protection against contamination: DONE if allocation was by community, institution or practice and it is unlikely that the control received the intervention; NOT CLEAR if professionals were allocated within a clinic or practice and it is possible that communication between experimental and group professionals could have occurred; NOT DONE if it is likely that the control group received the intervention (e.g. cross-over trials or if patients rather than professionals were randomised).

 ITS analyses

1.      Protection against changes: scored as DONE if the intervention occurred independently of other changes over time; NOT CLEAR if not specified (NOT DONE if information cannot be obtained from the authors); NOT DONE if reported that the intervention was not independent of other changes in time.

2.      Appropriate analysis: DONE if ARIMA (Auto-Regressive Integrated Moving Average) models were used, OR time series regression models were used to analyse the data and serial correlation was adjusted/tested for, OR if reanalysis performed; NOT CLEAR if not specified; NOT DONE if it is clear that neither of the conditions above are met.

3.      No selection bias in the sample framing: DONE if outcome measures are obtained from the whole population or a representative sample of the population studied; NOT CLEAR if not specified (treated as NOT DONE if information cannot be obtained from the authors); NOT DONE if data set is not drawn from a representative sample.

4.      Quality/reliability of outcome data: DONE if the outcome is obtained from an automated system (e.g. length of hospital stay) or comes from another objective source; NOT CLEAR if reliability is not reported for outcome measures that are obtained by chart extraction or collected by an individual (treated as NOT DONE if information cannot be obtained from the authors); and NOT DONE if the primary data are reportedly of a poor quality.

5.      Number of points specified: DONE if  3 or more data points before and 3 or more data points recorded after the intervention. Score NOT CLEAR if not specified in paper e.g. number of discrete data points not mentioned in text or tables (will be treated as NOT DONE if information cannot be obtained from the authors). Score NOT DONE if less than 3 data points recorded before and 3 data points recorded after intervention.

6.      Intervention effect specified: DONE if point of analysis was the point of intervention OR a rational explanation for the timing of intervention effect was given by the author(s).

7.      Detection bias: DONE if it is reported that the intervention itself was unlikely to affect data collection (for example, sources and methods of data collection were the same before and after the intervention).

History

Review first published: Issue 4, 2009

Contributions of authors

AH and NP secured the funding. ML and NP prepared the protocol, ML conducted the searches, ML and AH applied the inclusion criteria, assessed the quality and extracted the data for the included studies. ML prepared the report and NP and AH edited the draft.

Declarations of interest

None known.

Sources of support

Internal sources

  • London School of Hygiene and Tropical Medicine, UK.

External sources

  • Bill and Melinda Gates Foundation, Not specified.

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Attanasio 2005

MethodsCBA
Participants

Country: Colombia

Program: FA (Familias en Acion)

Eligible households (poorest in selected municipalities).

InterventionsCash incentives conditional on health and nutrition interventions for under 7 years old and school attendance for 8-18 years old.
Outcomes

Health services uptake:

Attendance of preventive care visits by children

Immunisation coverage:

Coverage of DPT vaccination (children)

Health outcomes:

Reported incidence of diarrhoea or respiratory diseases (children) 

Anthropometric or nutritional outcomes:

Height for height for age Z-score

Chronic malnourishment (children)

Notes 

Barham 2005a

MethodsC-RCT
Participants

Country: Mexico

Program: Progresa

Same as Gertler 2000.

Interventions

Same as Gertler 2000.

Up-to-date immunisation was part of the health requirements to get the monetary transfers.

Outcomes

Immunisation coverage:

Coverage of  DPT and Measles vaccination (children)

Notes 

Behrman 2005

MethodsC-RCT
Participants

Country: Mexico

Program: Progresa

Same as Gertler 2000.

InterventionsSame as Gertler 2000.
Outcomes

Anthropometric or nutritional outcomes:

Height increase

Notes

Progresa reanalysed: importance of baseline measurement and unobserved characteristics.

Subversion of randomisation.

Gertler 2000

MethodsC-RCT
Participants

Country: Mexico

Program: Progresa

Eligible households among selected communities (selected on poverty grounds).

InterventionsFamilies enrolled received two types of cash transfers: universal (dependent on attendance at health facilities for all family members) and specific (associated with school attendance of school-aged children).
Outcomes

Health services uptake:

Daily visits in the nearby health facilities

Health outcomes:

Reported morbidity (children)

NotesThe trial was funded by the Inter-American Development Bank.

Gertler 2004a

MethodsC-RCT
Participants

Country: Mexico

Program: Progresa

Same as Gertler 2000.

InterventionsSame as Gertler 2000.
Outcomes

Health outcomes:

Reported morbidity (children)

Anthropometric or nutritional outcomes:

Height increase

Prevalence of stunting

Notes

Progresa.

Re-analysed without taking into account baseline data because some variables were not measured at baseline.

Maluccio 2004

MethodsC-RCT
Participants

Country: Nicaragua

Program: RPS (Red de Protecion Social)

42 comarcas chosen to participate in the pilot phase (see Table 2): ½ randomly selected for intervention.

Interventions

Monetary transfer for children under 5 conditional on attendance at educational workshop and bringing child to preventive health programme (‘bono alimentario’).

Monetary transfer conditional on school attendance for 7-13 year old children (‘bono escolar’).

Outcomes

Health services uptake:

Attendance of preventive care visits by children

Immunisation coverage:

Reported up-to-date vaccination schedule (children)

Anthropometric or nutritional outcomes:

Prevalence of stunting, wasting and underweight (children under 5)

Height for Age Z-score (children under 5)

Prevalence of anaemia

NotesLimited external validity at national scale due to the purposive selection of areas (Table 2).

Morris 2004a

MethodsC-RCT
Participants

Country: Honduras

Program: PRAF (Programa de Asignacion Familiar)

Children and women from poor households, living in the beneficiary municipalities.

InterventionsEither or both: 1/ two types of monetary incentives (for health and education); 2/nutrition interventions + resources for local health teams.
Outcomes

Health services uptake:

Attendance of preventive and prenatal care by women

Attendance of preventive care visits by children

Immunisation coverage:

Coverage for DPT, Measles (children under 3) and tetanus toxoid (mothers)

NotesProgramme created in 1990 to mitigate the effects of structural adjustment.

Morris 2004b

MethodsCBA
Participants

Country: Brazil

Program: Bolsa Alimentação

Pregnant and lactating women and children  under 7 from low-income households.

InterventionsMothers received capped monthly transfers based on the number of beneficiaries.
Outcomes

Anthropometric or nutritional outcomes:

Height for Age Z-score (children)

Weight for Age Z-score (children)

Notes 

Rivera 2004

MethodsC-RCT
Participants

Country: Mexico

Program: Progresa

Same as Gertler 2000.

InterventionsSame as Gertler 2000.
Outcomes

Anthropometric or nutritional outcomes:

Prevalence of anaemia 

Height increase

NotesSub-study on a cohort of infants to investigate the nutritional impact.

Thornton 2006

MethodsC-RCT
Participants

Country: Malawi

Individuals who tested for STD.

InterventionsVoucher randomly given to individuals and exchangeable for cash if they come back to get their test results.
Outcomes

Health services uptake:

Proportion of people who went back to get the results of their tests

Notes 

Characteristics of excluded studies [ordered by study ID]

StudyReason for exclusion
Ahmed 2003Very broad intervention: unconditional cash transfer, free provision of health and hygiene services (latrine, pregnancy care, etc.)
Attanasio 2005bSummary of Attanasio 2005.
Barham 2005bBased on Progresa C-RCT but the analysis uses a mix of other data and ends up with a modelling study, not a (quasi) experimental one.
Behrman 2001Working paper of Berhman 2005; we used the published version.
Behrman 2004Outcome variables related to education.
Borghi 2005Voucher scheme for STI clinic attendance and treatment for sex workers; no cash transfer and not appropriate design (cost effectiveness).
Chase 2001The programme described does not focus on CCT.
Coady 2001Modelling study on Progresa, outcomes not of interest.
Dupas 2005Subsidised nets not cash transfer.
Gertler 2001Same as Gertler 2000 but with an additional wave that took place after Progresa began to be offered to the former control areas.
Gertler 2004bOutcome variables not relevant for our review (children’s development).
Levy 2003Methodology report on a forthcoming evaluation of a CCT programme in Jamaica.
Mushi 2003Targeted subsidy voucher scheme for bednets.
Pritchett 2002Targeted subsidies for health, no conditional cash transfers.
Saadah 2001Case study.
Sandiford 2005Case study on a voucher scheme.
Savedoff 2000Case study on health reform.
Schubert 2005Case study, study design not relevant.
Weeden 1986Excluded on the grounds that the nature of the transfer did not meet our definition of “conditional cash transfer”. Indeed, participants in this programme did not receive direct incentives to modify their uptake of contraceptive methods. The financial incentive was indirect as the programme provided easier access to small loans on the basis of village-level and individual-level contraception uptake.

Ancillary