Health system and community level interventions for improving antenatal care coverage and health outcomes

Lawrence Mbuagbaw; Nancy Medley; Andrea J Darzi; Marty Richardson; Kesso Habiba Garga; Pierre Ongolo‐Zogo

doi:10.1002/14651858.CD010994.pub2

. 2015 Dec 1;2015(12):CD010994. doi: 10.1002/14651858.CD010994.pub2

Health system and community level interventions for improving antenatal care coverage and health outcomes

Lawrence Mbuagbaw ^1,^2,^✉, Nancy Medley ³, Andrea J Darzi ⁴, Marty Richardson ⁵, Kesso Habiba Garga ¹, Pierre Ongolo‐Zogo ¹

Editor: Cochrane Pregnancy and Childbirth Group

PMCID: PMC4676908 PMID: 26621223

Abstract

Background

The World Health Organization (WHO) recommends at least four antenatal care (ANC) visits for all pregnant women. Almost half of pregnant women worldwide, and especially in developing countries do not receive this amount of care. Poor attendance of ANC is associated with delivery of low birthweight babies and more neonatal deaths. ANC may include education on nutrition, potential problems with pregnancy or childbirth, child care and prevention or detection of disease during pregnancy.

This review focused on community‐based interventions and health systems‐related interventions.

Objectives

To assess the effects of health system and community interventions for improving coverage of antenatal care and other perinatal health outcomes.

Search methods

We searched the Cochrane Pregnancy and Childbirth Group's Trials Register (7 June 2015) and reference lists of retrieved studies.

Selection criteria

We included randomised controlled trials (RCTs), quasi‐randomised trials and cluster‐randomised trials. Trials of any interventions to improve ANC coverage were eligible for inclusion. Trials were also eligible if they targeted specific and related outcomes, such as maternal or perinatal death, but also reported ANC coverage.

Data collection and analysis

Two review authors independently assessed trials for inclusion and risk of bias, extracted data and checked them for accuracy.

Main results

We included 34 trials involving approximately 400,000 women. Some trials tested community‐based interventions to improve uptake of antenatal care (media campaigns, education or financial incentives for pregnant women), while other trials looked at health systems interventions (home visits for pregnant women or equipment for clinics). Most trials took place in low‐ and middle‐income countries, and 29 of the 34 trials used a cluster‐randomised design. We assessed 30 of the 34 trials as of low or unclear overall risk of bias.

Comparison 1: One intervention versus no intervention

We found marginal improvements in ANC coverage of at least four visits (average odds ratio (OR) 1.11, 95% confidence interval (CI) 1.01 to 1.22; participants = 45,022; studies = 10; Heterogeneity: Tau² = 0.01; I² = 52%; high quality evidence). Sensitivity analysis with a more conservative intra‐cluster correlation co‐efficient (ICC) gave similar marginal results. Excluding one study at high risk of bias shifted the marginal pooled estimate towards no effect. There was no effect on pregnancy‐related deaths (average OR 0.69, 95% CI 0.45 to 1.08; participants = 114,930; studies = 10; Heterogeneity: Tau² = 0.00; I² = 0%; low quality evidence), perinatal mortality (average OR 0.96, 95% CI 0.89 to 1.03; studies = 15; Heterogeneity: Tau² = 0.01; I² = 45%; moderate quality evidence) or low birthweight (average OR 0.94, 95% CI 0.82 to 1.06; studies = five; Heterogeneity: Tau² = 0.00; I² = 5%; high quality evidence). Single interventions led to marginal improvements in the number of women who delivered in health facilities (average OR 1.08, 95% CI 1.02 to 1.15; studies = 10; Heterogeneity: Tau² = 0.00; I² = 0%; high quality evidence), and in the proportion of women who had at least one ANC visit (average OR 1.68, 95% CI 1.02 to 2.79; studies = six; Heterogeneity: Tau² = 0.24; I² = 76%; moderate quality evidence). Results for ANC coverage (at least four and at least one visit) and for perinatal mortality had substantial statistical heterogeneity. Single interventions did not improve the proportion of women receiving tetanus protection (average OR 1.03, 95% CI 0.92 to 1.15; studies = 8; Heterogeneity: Tau² = 0.01; I² = 57%). No study reported onintermittent prophylactic treatment for malaria.

Comparison 2: Two or more interventions versus no intervention

We found no improvements in ANC coverage of four or more visits (average OR 1.48, 95% CI 0.99 to 2.21; participants = 7840; studies = six; Heterogeneity: Tau² = 0.10; I² = 48%; low quality evidence) or pregnancy‐related deaths (average OR 0.70, 95% CI 0.39 to 1.26; participants = 13,756; studies = three; Heterogeneity: Tau² = 0.00; I² = 0%; moderate quality evidence). However, combined interventions led to improvements in ANC coverage of at least one visit (average OR 1.79, 95% CI 1.47 to 2.17; studies = five; Heterogeneity: Tau² = 0.00; I² = 0%; moderate quality evidence), perinatal mortality (average OR 0.74, 95% CI 0.57 to 0.95; studies = five; Heterogeneity: Tau² = 0.06; I² = 83%; moderate quality evidence) and low birthweight (average OR 0.61, 95% CI 0.46 to 0.80; studies = two; Heterogeneity: Tau² = 0.00; I² = 0%; moderate quality evidence). Meta‐analyses for both ANC coverage four or more visits and perinatal mortality had substantial statistical heterogeneity. Combined interventions improved the proportion of women who had tetanus protection (average OR 1.48, 95% CI 1.18 to 1.87; studies = 3; Heterogeneity: Tau² = 0.01; I² = 33%). No trial in this comparison reported on intermittent prophylactic treatment for malaria.

Comparison 3: Two interventions compared head to head. No trials found.

Comparison 4: One intervention versus a combination of interventions

There was no difference in ANC coverage (four or more visits and at least one visit), pregnancy‐related deaths, deliveries in a health facility or perinatal mortality. No trials in this comparison reported on low birthweight orintermittent prophylactic treatment of malaria.

Authors' conclusions

Implications for practice ‐ Single interventions may improve ANC coverage (at least one visit and four or more visits) and deliveries in health facilities. Combined interventions may improve ANC coverage (at least one visit), reduce perinatal mortality and reduce the occurrence of low birthweight. The effects of the interventions are unrelated to whether they are community or health system interventions.

Implications for research ‐ More details should be provided in reporting numbers of events, group totals and the ICCs used to adjust for cluster effects. Outcomes should be reported uniformly so that they are comparable to commonly‐used population indicators. We recommend further cluster‐RCTs of pregnant women and women in their reproductive years, using combinations of interventions and looking at outcomes that are important to pregnant women, such as maternal and perinatal morbidity and mortality, alongside the explanatory outcomes along the pathway of care: ANC coverage, the services provided during ANC and deliveries in health facilities.

Plain language summary

Health system and community level interventions for improving antenatal care coverage and health outcomes

What is the issue?

The World Health Organization recommends at least four antenatal visits for all pregnant women. Almost half of pregnant women worldwide miss out on this level of care, and this is more problematic in low‐ and middle‐income countries.

Why is this important?

Healthcare during pregnancy is a priority because poor antenatal attendance is associated with delivery of low birthweight babies and more newborn deaths. Antenatal care also provides opportunity for nutritional and health checks, such as whether a woman has a disease like malaria or has been exposed to infectious diseases such as HIV (human immunodeficiency virus) or syphilis.

What evidence did we find?

We reviewed randomised controlled trials that tested ways to improve the uptake of antenatal care during pregnancy. Some trials tested community‐based interventions (media campaigns, education on self and infant care or financial incentives for pregnant women to attend antenatal care), while other trials looked at health systems interventions (home visits for pregnant women or provision of equipment for clinics). We included 34 trials with approximately 400,000 women. Most trials took place in low‐ and middle‐income countries, and most trials were conducted in a way that made us feel confident about trusting the published reports. We assessed 30 of the 34 trials as of low or unclear overall risk of bias. The quality rating (high, moderate or low) shows our level of confidence that the result is robust and meaningful.

Trials comparing one intervention with no intervention

Single interventions only marginally improved the numbers of women attending four antenatal visits (high quality). Interventions did not improve rates of maternal death (low quality), baby deaths (moderate quality) or low birthweight (high quality). Even so, interventions led to modest improvements in the number of women who had at least one antenatal visit (moderate quality) and who delivered in a health facility (high quality). The number of women who received intermittent preventive treatment for malaria was not reported.

Trials comparing two or more interventions with no intervention

Combined interventions did not improve the number of women with four or more visits (low quality), or reduce maternal deaths (moderate quality). Nor did it increase the number of women who delivered in a health facility (moderate quality). However, more women who received combined interventions had one or more antenatal visits (moderate quality); there were also fewer baby deaths (moderate quality) and fewer low birthweight babies (moderate quality). The number of women who received intermittent preventive treatment for malaria was not reported.

We found no evidence that trials of community interventions worked differently from trials of health systems interventions.

Trials comparing one intervention with another intervention ‐ there were no trials for this comparison.

Trials comparing one intervention with a combination of interventions ‐ There was no difference in the number of women attending four or more antenatal visits (and at least one visit), maternal deaths, baby deaths, the number of deliveries in a health facility or the number of women who received intermittent preventive treatment for malaria.

What does this mean?

Single interventions may improve antenatal care coverage (women attending at least one visit and women attending four or more visits) and encourage women to give birth to their babies in health facilities. Combined interventions may also improve antenatal care coverage (at least one visit), reduce baby deaths and reduce the number of babies born with low birthweight.

We recommend that further studies of pregnant women and women in their reproductive years use combinations of interventions to maximise impact and look at outcomes that are important to the women themselves, such as maternal and baby deaths or ill health and the use of healthcare services.

Summary of findings

Background

In 2010, about 287,000 maternal deaths occurred worldwide (WHO 2013a). Despite considerable efforts to curb maternal mortality, close to 800 women continue to die every day due to complications of pregnancy and childbirth, and about 99% of these deaths occur in low‐ and middle‐income countries (WHO 2012c). In these settings, neonatal mortality rates are also high, despite the availability of evidence‐based interventions that could avert up to 72% of neonatal deaths (Darmstadt 2005). Interventions such as maternal immunisation against tetanus and skilled care at delivery can reduce both maternal and neonatal deaths (Lassi 2015).

Interventions to reduce maternal mortality may focus on three periods. The first is during pregnancy (antenatal care; ANC), the second is the intrapartum period, (i.e. during labour and delivery) and the third is in the postpartum period (after delivery). The intrapartum period is much shorter and less predictable than the longer more stable pregnancy period (Mbuagbaw 2011). It is also more challenging to provide adequate care in this period, especially in low‐ and middle‐income countries where human resource shortages and other health system weaknesses limit the availability of emergency obstetric care (Dogba 2009). ANC, on the other hand, is less resource‐intensive and its provision can be spread throughout the pregnancy period.

ANC generally comprises the following interventions (Kinzie 2004).

Health promotion: ANC is an opportunity to educate the woman about her health, pregnancy and childbirth, recognising danger signs, the benefits of good nutrition and exclusive breastfeeding, the harms of alcohol, tobacco and drugs, and other relevant issues.
Disease prevention: immunisation against tetanus, prophylactic treatment against malaria, and protection against iron‐deficiency anaemia are some conditions that can de addressed during ANC visits.
Early detection and treatment for complications and diseases: pregnant women can be screened for syphilis, human immunodeficiency virus (HIV) and other sexually transmitted infections. Complications of pregnancy such as pre‐eclampsia and eclampsia, infection and vaginal bleeding among others can be addressed.
Birth preparedness: the pregnant woman is counselled on her decision about where to deliver, choice of a skilled birth attendant and a caregiver (for herself or her other children at home). The ANC visit may cover planning for transportation to the hospital, costs of care and supplies for delivery.
Complication readiness: women are encouraged to have an emergency plan for complicated deliveries. This plan should include money for extra medical or surgical care and potential blood donors.

ANC may not address all the causes of maternal deaths; however, it is positively associated with receiving professional assistance at delivery (Bloom 1999; Mbuagbaw 2011; Mishra 2006; Oakley 2009) and improved pregnancy outcomes such as normal birthweight (Mbuagbaw 2011). In different regions, the effects of ANC on enhancing rates of delivery in a health facility are disparate (Mbuagbaw 2011; Raatikainen 2007).

Description of the condition

The World Health Organization (WHO) recommends at least four ANC visits for all pregnant women (WHO 2013).The first visit should take place during the first trimester (before the 12th week but no later than the 16th week), the second visit between the 24th and 28th week, and the third and fourth visits at 32 weeks and 36 weeks, respectively. Reports indicate that only 53% of pregnant women worldwide receive this amount of care (WHO 2013). Coverage is lower in low‐ and middle‐income countries where the use of maternal health care in general is limited and varies widely within and between countries (Say 2007). Poor attendance of ANC is associated with delivery of low birthweight infants (Mbuagbaw 2011; Raatikainen 2007; Showstack 1984; Siza 2008), and more neonatal deaths (Raatikainen 2007). ANC models with reduced visits may also be linked to higher perinatal mortality (Dowswell 2015; Vogel 2013).

Measuring antenatal care

Even though the WHO recommends four ANC visits during pregnancy, this is not a very informative measure (WHO 2013), as it gives no indication of the quality or timing of the visits. Furthermore, there is no measure of access. A comprehensive measure of ANC should include a measure of personal health‐seeking behaviour and also a measure of the availability of ANC services, as both are integral to effective ANC. More comprehensive measures have been proposed, which include the number and timing of visits, the provider of care and the adequacy of care provided (Delgado‐Rodriguez 1996; Mbuagbaw 2011). Well‐timed ANC visits are critical to the success of some interventions, as a systematic review has shown that adverse outcomes from syphilis can best be prevented by intervening in the first two trimesters (Hawkes 2013). The content of each ANC visit is also important, as some ANC interventions may not be beneficial, such as high does concomitant supplementation with vitamin C and vitamin E to prevent pre‐eclampsia in high‐risk women (Poston 2006). Irrespective of how it is measured, ANC is beneficial and represents an important point of contact with the health system for communication and pregnancy preparedness (Lassi 2015).

For the purposes of this review, coverage will be considered as the proportion of pregnant women who attend at least four ANC visits.

Description of the intervention

The fifth United Nations' Millenium Development Goal (MDG5) targets maternal health and explicitly calls for more ANC (United Nations 2013). The WHO now recommends a package of reduced visits with evidence‐based interventions through goal‐oriented clinic visits (WHO 2011). A variety of interventions can be used to increase the number of women who receive ANC. A systematic review on the effectiveness of interventions to improve early initiation of ANC in vulnerable populations identified two broad categories of interventions: outreach/community‐based interventions and alternative models of clinic‐based ANC. The former included the use of lay health workers and mobile health clinics, while the latter included adaptations of clinic‐based ANC to be more collaborative and comprehensive, and also to accommodate teens (Oakley 2009).

Community‐based interventions such as community support, mobilisation, education and home visits by trained community health workers can lead to significant reductions in maternal morbidity and neonatal mortality, and an increase in referrals to a health facility (Lassi 2015). In underserved areas, a community health van may improve access to adequate ANC (Edgerley 2007).

Other interventions, such as mass media campaigns, social mobilisation, information‐education‐communication (IEC) interventions, financial incentives, behaviour change interventions and policy interventions targeting health workers or pregnant women will also be investigated.

How the intervention might work

Interventions targeting the factors that reduce antenatal care coverage may be beneficial.

Health policy is a critical component of any health system and guides how resources (man power, money and material) are used. Policy can be applied at any level of the health system. Regional health managers are capable of making policy changes that influence the use of ANC services. Recent papers suggest that the effects of policy change in health outcomes should be explored in more detail (Dettrick 2013). Such policy changes may include capacity building in ANC to improve quality of care (Lassi 2015; Say 2007; van Eijk 2006), re‐organisation of services to include more midwives providing ANC (Dowswell 2015; Khan‐Neelofur 1998), and reduction of user fees to eliminate financial barriers (Lassi 2015; Mbuagbaw 2011; Say 2007; Titaley 2010; van Eijk 2006). Where coverage is better in the private sector (Cesar 2012), adopting their (private sector) model of care may be beneficial. Switching to individual counselling sessions may also improve the number of high‐risk women delivering in hospitals (Ballard 2013).

Mass media campaigns can be used to improve the utilisation of health services (Grilli 2002), and may also help to improve the use of ANC services. Social mobilisation ‐ engaging multiple stakeholders ‐ is an important way of bringing change in communities. If pregnant women receive the same consistent message on the benefits of ANC from health workers, community health workers and in other social gatherings, they may be more likely to take heed. Lack of awareness (Lassi 2015; Titaley 2010) and misconceptions (Agus 2012; Say 2007) about ANC can be addressed using IEC sessions. Financial incentives can be used to encourage pregnant women to attend ANC and cover costs including user fees and transportation costs where these problems exist (Lassi 2015; Mbuagbaw 2011; Say 2007; Titaley 2010; van Eijk 2006). They are most effective in the short term, and in resource‐limited settings (Marteau 2009). Behaviour change interventions are interventions derived from a specific model or theory of behaviour change and can play a role in improving health outcomes (Marteau 2006). Such interventions could play an important role in encouraging women to attend ANC.

Why it is important to do this review

Regions of the world with low ANC coverage can benefit from a comprehensive synthesis of the evidence surrounding the ways in which ANC coverage can be improved. In these places, low ANC coverage comes with low rates of deliveries in health facilities and assistance by skilled birth attendants. The latter two factors are associated with high materno‐fetal morbidity. This review will have important implications for reproductive health policy, the provision of services to women in reproductive ages and may highlight gaps in current evidence or openings for further research.

Objectives

To assess the effects of health system and community interventions for improving coverage of antenatal care and other outcomes.

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs), quasi‐randomised trials and cluster‐randomised trials. Trials of cross‐over design were not eligible. Trials reported in abstract form were eligible for inclusion in the review; however, we did not include any trial based on an abstract report alone.

Types of participants

This review included studies of stakeholders, providers of care and beneficiaries, including but not limited to:

professional health workers;
lay health workers;
community members;
pregnant women;
women of reproductive age.

Types of interventions

All interventions susceptible to improve coverage of ANC were eligible for inclusion in this review. These interventions could be aimed at the health system, the population or both. Owing to the potentially wide variety of interventions, there were no restrictions to duration or frequency of the intervention. For the purposes of this review, we classified these interventions into the following two main categories.

Interventions aimed at the health system

Policy changes.
Health worker education.
Re‐organisation of health services.

Interventions aimed at the community

Mass media campaigns.
Social mobilisation.
Information‐education‐communication (IEC).
Financial incentives.
Behaviour change interventions.

Types of outcome measures

Primary outcomes

Coverage of ANC: the proportion of pregnant women who attend at least four ANC visits during pregnancy.
Pregnancy‐related deaths: the proportion of women who die during pregnancy or 42 days after, irrespective of cause (WHO 2004).

Secondary outcomes

Coverage of ANC: the proportion of pregnant women who attend at least one ANC visit during pregnancy.
The proportion of pregnant women who initiate ANC in the first trimester.
The proportion of pregnant women who receive ANC from professional health workers.
The proportion of deliveries in health facilities.
The proportion of pregnant women with a written birth and emergency plan by 37 weeks of pregnancy.
The proportion of pregnant women who receive Intermittent Prophylactic Treatment (IPT) for malaria as per recommended guidelines (WHO 2012b).
The proportion of women with tetanus protection at birth.
The proportion of pregnant women who screen for syphilis.
The proportion of women who screen for asymptomatic bacteriuria.
The proportion of women who screen for HIV.
The proportion of women with HIV who receive a complete antiretroviral course for prevention of mother‐to‐child transmission of HIV.
Maternal near miss, defined as: "a woman who nearly died but survived a complication that occurred during pregnancy, childbirth or within 42 days of termination of pregnancy" (Pattinson 2009).
The proportion of women with preterm labour or delivery.
The proportion of low‐birthweight infants born.
The incidence of perinatal mortality.

We also considered combinations of the above outcomes if the data were not dissociable. For example, the proportion of women who have at least four well‐spaced ANC visits attended by a professional health worker (Mbuagbaw 2011).

Search methods for identification of studies

The following methods section of this review is based on a standard template used by the Cochrane Pregnancy and Childbirth Group.

Electronic searches

We searched the Cochrane Pregnancy and Childbirth Group’s Trials Register by contacting the Trials Search Co‐ordinator (7 June 2015).

The Cochrane Pregnancy and Childbirth Group’s Trials Register is maintained by the Trials Search Co‐ordinator and contains trials identified from:

monthly searches of the Cochrane Central Register of Controlled Trials (CENTRAL);
weekly searches of MEDLINE (Ovid);
weekly searches of Embase (Ovid);
monthly searches of CINAHL (EBSCO);
handsearches of 30 journals and the proceedings of major conferences;
weekly current awareness alerts for a further 44 journals plus monthly BioMed Central email alerts.

Details of the search strategies for CENTRAL, MEDLINE, Embase and CINAHL, the list of handsearched journals and conference proceedings, and the list of journals reviewed via the current awareness service can be found in the ‘Specialized Register’ section within the editorial information about the Cochrane Pregnancy and Childbirth Group.

Trials identified through the searching activities described above are each assigned to a review topic (or topics). The Trials Search Co‐ordinator searches the register for each review using the topic list rather than keywords.

Searching other resources

We searched the reference lists of retrieved studies and contacted authors and experts in the field. We did not apply any language or date restrictions.

Data collection and analysis

The following methods section of this review is based on a standard template used by the Cochrane Pregnancy and Childbirth Group.

Selection of studies

Two review authors (L Mbuagbaw (LM) and N Medley (NM)) independently assessed all the potential studies we identified for inclusion. We resolved any disagreement through discussion or, if required, by consulting a third author (P Ongolo‐Zogo (POZ)). Agreement on the inclusion of studies was estimated using the Kappa statistic (Viera 2005).

A Preferred Reporting Items for Systematic Reviews and Meta‐analyses (PRISMA) study flow diagram was created to map out the number of records identified, included and excluded (Liberati 2009).

Data extraction and management

We designed and tested a form to extract data. For eligible studies (abstract or full text), LM, AD and NM extracted the data using the agreed form. We resolved discrepancies through discussion or, if required, we consulted a third review author (POZ). We entered data into Review Manager software (RevMan 2014) and checked for accuracy.

When information regarding any of the above was unclear, we attempted to contact authors of the original reports to provide further details.

Assessment of risk of bias in included studies

Review authors LM and NM independently assessed risk of bias for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We resolved any disagreement by discussion or by involving a third assessor (POZ).

(1) Random sequence generation (checking for possible selection bias)

We described for each included study the method used to generate the allocation sequence in sufficient detail to allow an assessment of whether it should produce comparable groups.

We assessed the method as:

low risk of bias (any truly random process, e.g. random number table; computer random number generator);
high risk of bias (any non‐random process, e.g. odd or even date of birth; hospital or clinic record number);
unclear risk of bias.

(2) Allocation concealment (checking for possible selection bias)

We described for each included study the method used to conceal allocation to interventions prior to assignment and assessed whether intervention allocation could have been foreseen in advance of, or during recruitment, or changed after assignment.

We assessed the methods as:

low risk of bias (e.g. telephone or central randomisation; consecutively numbered sealed opaque envelopes);
high risk of bias (open random allocation; unsealed or non‐opaque envelopes, alternation; date of birth);
unclear risk of bias (cluster‐RCTs that reported no allocation concealment and RCTs that made no mention of allocation concealment).

(3.1) Blinding of participants and personnel (checking for possible performance bias)

We described for each included study the methods used, if any, to blind study participants and personnel from knowledge of which intervention a participant received. We considered that studies were at low risk of bias if they were blinded, or if we judged that the lack of blinding would be unlikely to affect results. We assessed blinding separately for different outcomes or classes of outcomes.

We assessed the methods as:

low, high or unclear risk of bias for participants;
low, high or unclear risk of bias for personnel.

(3.2) Blinding of outcome assessment (checking for possible detection bias)

We described for each included study the methods used, if any, to blind outcome assessors from knowledge of which intervention a participant received. We assessed blinding separately for different outcomes or classes of outcomes.

We assessed methods used to blind outcome assessment as:

low, high or unclear risk of bias. We entered "unclear risk" for studies that were not blinded for participants/personnel, or where blinding was not described, as most often blinding was not possible due to the nature of the intervention. It is unclear whether a lack of blinding presented a serious risk of bias in cluster‐RCTs with mortality outcomes.

(4) Incomplete outcome data (checking for possible attrition bias due to the amount, nature and handling of incomplete outcome data)

We described for each included study, and for each outcome or class of outcomes, the completeness of data including attrition and exclusions from the analysis. We stated whether attrition and exclusions were reported and the numbers included in the analysis at each stage (compared with the total randomised participants), reasons for attrition or exclusion where reported, and whether missing data were balanced across groups or were related to outcomes. Where sufficient information was reported, or was supplied by the trial authors, we re‐included missing data in the analyses which we undertook.

We assessed methods as:

low risk of bias (e.g. no missing outcome data; missing outcome data balanced across groups);
high risk of bias (e.g. numbers or reasons for missing data imbalanced across groups; ‘as treated’ analysis done with substantial departure of intervention received from that assigned at randomisation);
unclear risk of bias.

Attrition of 20% or more was considered as high risk of bias.

(5) Selective reporting (checking for reporting bias)

We described for each included study how we investigated the possibility of selective outcome reporting bias and what we found.

We assessed the methods as:

low risk of bias (where it is clear that all of the study’s pre‐specified outcomes and all expected outcomes of interest to the review have been reported);
high risk of bias (where not all the study’s pre‐specified outcomes have been reported; one or more reported primary outcomes were not pre‐specified; outcomes of interest are reported incompletely and so cannot be used; study fails to include results of a key outcome that would have been expected to have been reported);
unclear risk of bias.

(6) Other bias (checking for bias due to problems not covered by (1) to (5) above)

We described for each included study any important concerns we had about other possible sources of bias.

We assessed whether each study was free of other problems that could put it at risk of bias:

low risk of other bias;
high risk of other bias;
unclear whether there is risk of other bias.

(7) Overall risk of bias

We made explicit judgements about whether studies were at high risk of bias, according to the criteria given in the Handbook (Higgins 2011). With reference to (1) to (6) above, we assessed the likely magnitude and direction of the bias and whether we considered it was likely to impact on the findings. We explored the impact of the level of bias by undertaking sensitivity analyses ‐ seeSensitivity analysis.

For the cluster‐randomised trials we included, additional sources bias were considered (Higgins 2011), such as:

recruitment bias: whether individuals were recruited into the trial after the clusters had been formed;
baseline imbalances: due to the small numbers of clusters;
attrition of entire clusters;
analysis bias: methods of analysis ignoring the correlation between members of the same cluster;
their comparability with individually‐randomised trials.

Assessing the quality of the body of evidence using the GRADE approach

The quality of the evidence was assessed using the GRADE approach as outlined in the GRADE Handbook in order to assess the quality of the body of evidence relating to the following outcomes.

Coverage of ANC: the proportion of pregnant women who attend at least four ANC visits during pregnancy.
Pregnancy‐related deaths: the proportion of women who die during pregnancy or 42 days after, irrespective of cause (WHO 2004).
Coverage of ANC: the proportion of pregnant women who attend at least one ANC visit during pregnancy.
The proportion of deliveries in health facilities.
The incidence of perinatal mortality.
The proportion of low‐birthweight infants born.
The proportion of pregnant women who receive Intermittent Prophylactic Treatment (IPT) for malaria as per recommended guidelines (WHO 2012b).

The comparisons used to assess the quality of evidence were: Comparison 1: one intervention versus no intervention, and Comparison 2: multiple interventions versus no intervention.

We used the GRADEpro Guideline Development Tool to import data from Review Manager 5.3 (RevMan 2014) in order to create 'Summary of findings' tables. A summary of the intervention effect and a measure of quality for each of the above outcomes was produced using the GRADE approach. The GRADE approach uses five considerations (study limitations, consistency of effect, imprecision, indirectness and publication bias) to assess the quality of the body of evidence for each outcome. The evidence can be downgraded from 'high quality' by one level for serious (or by two levels for very serious) limitations, depending on assessments for risk of bias, indirectness of evidence, serious inconsistency, imprecision of effect estimates or potential publication bias.

We downgraded outcomes for substantial heterogeneity where the I² > 60%. Because lack of blinding of participants and personnel was nearly universal and would have been prohibitive to the implementation of our included trials, we did not downgrade evidence for lack of blinding during our GRADE assessments. Where we downgraded outcome evidence for risk of bias concerns, these design limitations had to do with the trials' problems with other bias domains, not lack of blinding.

Measures of treatment effect

Dichotomous data

For dichotomous data, we planned to present results as summary risk ratio with 95% confidence intervals. However, the trials in our review reported several different summary effect measures, and the majority presented cluster‐adjusted odds ratios. To make use of these trial data and retain the specific adjustments made by trial statisticians, we have presented our results as odds ratios, with 95% confidence intervals.

Continuous data

For continuous data, we planned to use the mean difference if outcomes were measured in the same way between trials. We would have used the standardised mean difference to combine trials that measured the same outcome, but used different methods.

Unit of analysis issues

Cluster‐randomised trials

We included cluster‐randomised trials in the analyses along with individually‐randomised trials. The review statistician (M Richardson) made appropriate adjustments according to the type of data available from trial reports and through correspondence with authors. Where possible, we adjusted a trials' sample sizes using the methods described in the Handbook using an estimate of the intracluster correlation co‐efficient (ICC) derived from the trial, from a similar trial or from a study of a similar population. We consulted Piaggio 2001, Pagel 2011, Prost 2013, Kidney 2009 and Souza 2013. Professor Souza further provided an ICC for maternal mortality via email. We conducted sensitivity analyses using two extreme values of ICC to investigate the effect of variation in the ICC for the primary outcome only. We have provided details of the specific ICCs used and of any additional methods required to analyse trial data in the additional tables. There is one table of adjustments made for each outcome relevant to the 'Summary of findings' table. Similar details of adjustments made for the analyses of other review outcomes are freely available upon request from the review team.

For cluster‐ and individually‐randomised trials, we considered it reasonable to combine the results from both, if there was little heterogeneity between the study designs and the interaction between the effect of intervention and the choice of randomisation unit was considered to be unlikely. We acknowledged heterogeneity in the randomisation unit, but we have not conducted sensitivity analysis by randomisation unit because all but five of the included trials are cluster‐randomised. Per outcome, there are not sufficient individually‐randomised trials for a subgroup (often just one or no individually‐randomised trial contributed). If in future updates we include several more individually‐randomised trials, we will perform sensitivity analysis, though at present all of the trials in ongoing studies are of cluster‐randomised design.

Other unit of analysis issues

No included trial specifically recruited multiple pregnancies. However, several if not most of the review's trials likely included women with multiple pregnancies. Where this information was provided in trial reports, we have noted this in the Characteristics of included studies table. Results were not reported separately for twins in any included trial. The proportion of twin pregnancies per trial was assumed to be low, and we have not made any adjustments to outcome data for their inclusion.

Some of the included studies had more than one intervention arm. Where appropriate, two intervention arms were combined, and pair‐wise comparisons conducted. Where necessary, we have also split trials. For example, we divided trials if trials had multiple arms and interventions and those interventions were relevant to separate review comparisons. Finally, we have split the control group when necessary to avoid double counting. Please see the included studies tables for Lewycka 2013a and Morris 2004a for specific details.

For quasi‐randomised studies, we planned to use estimates adjusted for baseline imbalances and other confounders.

Dealing with missing data

For included studies, we noted levels of attrition and explored the impact of including studies with high levels of missing data in the overall assessment of treatment effect by using sensitivity analysis.

For all outcomes, we carried out analyses, as far as possible, on an intention‐to‐treat basis, i.e. we attempted to include all participants randomised to each group in the analyses, and all participants were analysed in the group to which they were allocated, regardless of whether or not they received the allocated intervention. The denominator for each outcome in each trial was the number randomised minus any participants whose outcomes were known to be missing.

Assessment of heterogeneity

We assessed statistical heterogeneity in each meta‐analysis using the I², T² and Chi² statistics. We regarded heterogeneity as substantial if an I² was greater than 30% and either a T² was greater than zero, or there was a low P value (less than 0.10) in the Chi² test for heterogeneity. We have left the overall totals turned off only if the effect estimates of different trials were so disparate that it made little clinical sense to combine the trials (for example, if effect estimates fell on opposite sides of the line of no difference and the trials' confidence intervals do not overlap). Overall, even with high heterogeneity in several outcomes, the effect estimates were reasonably similar, and confidence intervals always overlapped. We have not turned off the totals for any outcome listed below.

Assessment of reporting biases

We investigated reporting biases (such as publication bias) using funnel plots. We assessed funnel plot asymmetry visually. If asymmetry was suggested by a visual assessment, we performed exploratory analyses to investigate it.

Data synthesis

We carried out statistical analysis using the Review Manager software (RevMan 2014). We planned to use fixed‐effect meta‐analysis for combining data where it was reasonable to assume that studies were estimating the same underlying treatment effect: i.e. where trials were examining the same intervention, and the trials’ populations and methods were judged sufficiently similar. In practice, we found considerable clinical heterogeneity between the included studies in the review and expected high statistical heterogeneity as a consequence. Therefore, we used random‐effects models to conduct all meta‐analyses. Where high heterogeneity in the overall summary was found, we left analyses pooled and advised readers to interpret the results with caution.

The random‐effects summary was treated as the average range of possible treatment effects and was reported as such, with 95% confidence intervals, and the estimates of T² and I².

We considered the following comparisons.

One intervention versus no intervention.
Two interventions compared.
One intervention versus a combination of interventions.
Combination of interventions versus no intervention.
Different combinations of interventions.

Subgroup analysis and investigation of heterogeneity

We planned to carry out the following subgroup analyses.

Urban versus rural settings.
High‐income versus lower‐income settings.
Interventions targeting the health system versus interventions targeting the population.

These subgroup analyses were limited to the primary outcomes of the review.

We were not able to conduct analyses for urban versus rural settings because our trials were overwhelmingly rural. For example, 14 of 19 trials contributing data for the primary outcome of ANC coverage (at least four visits) took place in rural settings, with just three in urban settings (one trial had mixed settings and one trial report was unclear).

We conducted subgroup analyses comparing trials in upper‐middle and high‐income countries with trials set in low‐ and lower‐middle‐income countries for the outcome of ANC coverage (at least four visits) only (World Bank 2015); there were insufficient data for this analysis for the primary outcome of maternal mortality, because just three of 10 trials contributing data were conducted in higher‐income settings.

We conducted subgroup analysis with trials targeting the health system versus trials targeting the population. All subgroup analyses will be re‐assessed when more trials are added during future updates of the review.

We assessed subgroup differences by interaction tests available within RevMan (RevMan 2014). We reported the results of subgroup analyses quoting the χ2 statistic and P value, and the interaction test I² value.

Sensitivity analysis

Sensitivity analyses were conducted to explore the effects of trial quality and trial design on the outcomes. In the first instance, we compared the results from the studies with high risk of bias with those at low risk of bias, and secondly we investigated the effect of the unit of randomisation (individual versus cluster) on the outcomes. Likewise, we also explored the effects of fixed‐effect or random‐effects analyses for outcomes with statistical heterogeneity and the effects of any assumptions made such as the value of the ICC used for cluster‐randomised trials.

Results

Description of studies

Results of the search

The search of the Cochrane Pregnancy and Childbirth Group's Trials Register retrieved 148 reports; we found three further reports during our own searches (see: Figure 1). We included 34 trials (79 reports) and excluded 30 trials (59 reports). Thirteen studies (13 reports) are ongoing (see Characteristics of ongoing studies), and we await trial completion or the publication of results. We have been unable to locate published data for any of the trials in ‘ongoing studies,’ as of 1 Sept, 2015. Two trials with multiple arms were divided for analyses (Lewycka 2014; Morris 2004), brining the total number of tables for included studies and risk of bias to 36.

Included studies

A total of 34 trials were identified as eligible for inclusion in this review. Trials involved approximately 400,000 women who were either of reproductive age or pregnant. Ten of the included trials aimed to evaluate health system interventions only (Basinga 2011; Darmstadt 2010; Kenyon 2012; Lund 2012; Majoko 2007; Penfold 2014; Persson 2013; Richter 2014; Villar 2001 (WHO 2001); Walker 2013). One trial with multiple arms aimed to evaluate health system and community interventions separately and together (Morris 2004a). Ten trials evaluated community interventions alone (Barber 2008; Fottrell 2013; Laken 1995; Lewycka 2013a; Manandhar 2004; Melnikow 1997; More 2012; Mori 2015; Mullany 2007; Tripathy 2010). The rest of the trials (13) aimed to evaluate a combination of health system and community interventions (Azad 2010; Baqui 2008; Bhutta 2011; Kirkwood 2013; Klerman 2001; Kumar 2008; le Roux 2013; Midhet 2010; Omer 2008; Villar 1992; Wahlstrom 2011; Waiswa 2015; Wu 2011).

Included trials took place in: Argentina (2), Bangladesh (4), Brazil, Cuba (2), Eastern China, Ghana, Honduras, India (3), Laos, Malawi, Mexico (3), Mongolia, Nepal (2), Pakistan (3), Rwanda, Saudi Arabia, South Africa (2), Southern Tanzania, Thailand, Uganda, United Kingdom, USA (3), Vietnam, Zanzibar and Zimbabwe. Two trials, Villar 1992 and Villar 2001 (WHO 2001), recruited clusters in multiple countries. Five trials were conducted in upper‐middle‐ or high‐income countries (Kenyon 2012; Klerman 2001; Laken 1995; Melnikow 1997; Villar 1992); the remaining trials took place in low‐ or lower‐middle‐income countries. Trials were overwhelmingly rural (22/34), with seven trials in mixed settings and four trials in urban settings; the setting of a final trial was unclear.

All trials were of cluster‐randomised design, apart from five (Kenyon 2012; Klerman 2001; Laken 1995; Melnikow 1997; Mullany 2007). Sample sizes ranged from a few hundred to more than 50,000, while cluster numbers ranged from eight to 506. Follow‐up times ranged from six to 72 months.

Interventions

Ten trials evaluated health system interventions (Basinga 2011; Darmstadt 2010; Kenyon 2012; Lund 2012; Majoko 2007; Penfold 2014; Persson 2013; Richter 2014; Villar 2001 (WHO 2001); Walker 2013). Three of the nine trials included interventions aimed at policy change (Basinga 2011; Lund 2012; Persson 2013). One trial provided financial support to the clusters (Basinga 2011); one trial sent text reminders and mobile credit vouchers (Lund 2012), and one trial held regular stakeholder problem‐solving meetings with generated actions (Persson 2013). Seven trials aimed for reorganisation of the health services (Darmstadt 2010; Kenyon 2012; Majoko 2007; Penfold 2014; Richter 2014; Villar 2001 (WHO 2001); Walker 2013). Four trials added home visits (Darmstadt 2010; Kenyon 2012; Penfold 2014; Richter 2014); two trials restructured the visits (Majoko 2007; Villar 2001 (WHO 2001)), and one trial added an obstetric nurse to rural clinics (Walker 2013).

In a four‐armed trial Morris 2004a tested a policy change, financial support and a combination of the two interventions against no intervention.

Ten trials evaluated community interventions (Barber 2008; Fottrell 2013; Laken 1995; Lewycka 2013a; Manandhar 2004; Melnikow 1997; More 2012; Mori 2015; Mullany 2007; Tripathy 2010). The majority of trials (7/10) were based on Information‐education‐communication (IEC) interventions (Fottrell 2013; Lewycka 2013a; Manandhar 2004; More 2012; Mori 2015; Mullany 2007; Tripathy 2010). Of the seven trials, five held participatory women’s groups (Fottrell 2013; Lewycka 2013a; Manandhar 2004; More 2012; Tripathy 2010); one trial had private education sessions (Mullany 2007), and one trial provided women with their own case notes to carry to antenatal visits (Mori 2015). The other four trials gave financial incentives (Barber 2008; Laken 1995; Melnikow 1997; Morris 2004a).

Thirteen trials were based on combined health system and community interventions (Azad 2010; Baqui 2008; Bhutta 2011; Kirkwood 2013; Klerman 2001; Kumar 2008; le Roux 2013; Midhet 2010; Omer 2008; Villar 1992; Wahlstrom 2011; Waiswa 2015; Wu 2011). Several trials provided IEC as a community intervention. These included: participatory women’s groups (Azad 2010), community awareness meetings (Baqui 2008; Bhutta 2011; Klerman 2001; Kumar 2008; Wahlstrom 2011); the provision of information booklets and cassettes and mass media campaigns (Midhet 2010; Villar 1992; Wu 2011); and home education sessions (Kirkwood 2013; le Roux 2013; Omer 2008). Azad 2010, Bhutta 2011, Kirkwood 2013, Midhet 2010, Waiswa 2015, and Wu 2011 provided healthcare worker training as a health system package to the intervention group. Baqui 2008 offered surveillance and home visits (two ANC, three perinatal mortality rate (PNC)) by community health workers (CHWs) and the provision of iron and folic acid supplements. Four other trials also added home visits as a health system intervention (Kumar 2008; le Roux 2013; Villar 1992; Waiswa 2015), while Klerman 2001 added additional clinic visits and extended time with clinicians. Kumar 2008 provided home visits with thermo‐spot and home visits without thermo‐spot as health system interventions. Wu 2011, Wahlstrom 2011 and Waiswa 2015 each also provided equipment and/or medicines for health centres.

Outcomes

Apart from the trial primary outcome, the outcomes listed in the Characteristics of included studies tables refer to this review's prespecified primary and secondary outcomes. All trials reported additional outcomes not relevant to our review beyond those listed in the Characteristics of included studies tables.

Many trials had multiple primary outcomes including different combinations of the following priorities: neonatal and perinatal mortality, antenatal care coverage, healthcare utilisation, newborn care practices (including exclusive breastfeeding), maternal physical and mental health (including postpartum anaemia, treated urinary tract infection, indicators of social support, maternal depression, use of folic acid during pregnancy, tetanus immunisation, birth preparedness and satisfaction with care) and delivery practices (including assisted delivery, hospital delivery). Our 34 included trials had 23 different primary outcomes. Even with this diversity, our included trials most commonly targeted antenatal care coverage and neonatal mortality. Sixteen trials had primary outcomes with a specific antenatal care coverage component (Basinga 2011; Darmstadt 2010; Laken 1995;, Majoko 2007; Melnikow 1997; Midhet 2010; More 2012; Mori 2015; Morris 2004, Mullany 2007; Omer 2008; Richter 2014; Wahlstrom 2011; Waiswa 2015; Walker 2013; Wu 2011), and 12 trials targeted neonatal or perinatal mortality (Azad 2010; Baqui 2008; Bhutta 2011; Fottrell 2013; Kirkwood 2013; Kumar 2008; Lewycka 2013; Manandhar 2004; Midhet 2010; More 2012; Persson 2013; Tripathy 2010).

One trial was included based on unpublished data only (Wahlstrom 2011). Two included trials contributed no usable outcome data to the review (Melnikow 1997; Omer 2008). The trial statistician for Mori 2015 provided unpublished, cluster‐adjusted odds ratios for our use. See the Included studies tables for further details.

Excluded studies

Thirty trials were excluded from the review. Briefly, 27 trials did not assess an increase in antenatal care coverage as a central objective of the trial (Baqui 2009; Bhutta 2008; Colbourn 2013; Dance 1987; Doyle 2014; Ellard 2012; Ford 2001; Gokcay 1993; Homer 2001; Ickovics 2007; Jennings 2010; Kafatos 1989; Koniak‐Griffin 1991; Koniak‐Griffin 2000; Kusulasai 1993; Leung 2012; Miller 2012; Munjanja 1996; Olds 1986; Olds 1995a; Rodriguez‐Angulo 2012; Schellenberg 2011; Srinivasan 1995; Tomlinson 2014; Tough 2006; Tough 2007; Turan 2001). Three studies were excluded for being of observational study design (Alisjahbana 1995; Foord 1995; Magriples 2008). See the Characteristics of excluded studies table for more details.

Risk of bias in included studies

Because two included trials with multiple arms have been divided (Lewycka 2013 and Morris 2004), the total number of studies assessed for bias below is 36, rather than 34.

For the most part, risk of bias was low or unclear. See Figure 2 and Figure 3. In our overall risk assessment only four studies were at high risk of bias: Omer 2008 (data not reported according to randomisation, unclear denominators); Walker 2013 (all data based on provider recall); Wu 2011 (unclear denominators, other interventions in trial area and poor implementation) and Wahlstrom 2011 (unpublished data, unclear analyses). In 16 trials, overall risk of bias was unclear (Basinga 2011; Darmstadt 2010; Fottrell 2013; Laken 1995; le Roux 2013; Lewycka 2013b; Lewycka 2013a; Melnikow 1997; Midhet 2010; More 2012; Morris 2004b; Morris 2004a; Persson 2013; Richter 2014; Tripathy 2010; Waiswa 2015), and it was low in 16 trials (Azad 2010; Baqui 2008; Barber 2008; Bhutta 2011; Kenyon 2012; Kirkwood 2013; Klerman 2001; Kumar 2008; Lund 2012; Majoko 2007; Manandhar 2004; Mori 2015; Mullany 2007; Penfold 2014; Villar 1992; Villar 2001 (WHO 2001)).

'Risk of bias' graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.

Allocation

Random sequence generation

Random sequence generation was adequate (low risk of bias) in 30 trials (Azad 2010; Baqui 2008; Barber 2008; Basinga 2011; Bhutta 2011; Darmstadt 2010; Kenyon 2012; Kirkwood 2013; Kumar 2008; Lewycka 2013a; Lewycka 2013b; Manandhar 2004; Melnikow 1997; Midhet 2010; More 2012; Mori 2015; Morris 2004a; Morris 2004b; Mullany 2007; Omer 2008; Penfold 2014; Persson 2013; Richter 2014; Tripathy 2010; Villar 1992; Villar 2001 (WHO 2001); Wahlstrom 2011; Waiswa 2015; Walker 2013; Wu 2011), and unclear in six trials (Fottrell 2013; Klerman 2001; Laken 1995; le Roux 2013; Lund 2012; Majoko 2007).

Allocation concealment

Seventeen trials were at low risk of bias for allocation concealment (Baqui 2008; Fottrell 2013; Kenyon 2012; Kirkwood 2013; Klerman 2001; Kumar 2008; le Roux 2013; Manandhar 2004; Melnikow 1997; Mori 2015; Morris 2004a; Morris 2004b; Penfold 2014; Richter 2014; Villar 1992; Villar 2001 (WHO 2001); Waiswa 2015). Risk of bias was unclear in 19 trials, where allocation concealment was unclear or not described at all (Azad 2010; Barber 2008; Basinga 2011; Bhutta 2011; Darmstadt 2010; Laken 1995; Lewycka 2013a; Lewycka 2013b; Lund 2012; Majoko 2007; Midhet 2010; More 2012; Mullany 2007; Omer 2008; Persson 2013; Tripathy 2010; Wahlstrom 2011; Walker 2013; Wu 2011).

Blinding

Generally, blinding was not feasible in the trials included in this review. Large cluster‐randomised trials with community or health‐system relevant interventions will not achieve adequate blinding of women and staff. Most trials did make an attempt to prevent contamination between clusters, though. Where trial authors have reported lack of blinding, we have judged this to be of unclear risk of bias because we do not know for certain what kinds of bias may or may not have been introduced. We determined that it was unfair to judge studies that reported a lack of blinding as having a high risk of bias and studies that did not report at all as unclear risk of bias, when blinding might not have been feasible. The risk of bias in all trials due to lack of blinding will differ per outcome, and the impact of no blinding in cluster trials measuring mortality outcomes is also unclear. Where trials did not report blinding at all in published reports, we judged the trial as of unclear risk of bias due to lack of information. Overall, included trials did a much better job of attempting to blind outcomes assessors than participants and personnel.

Blinding of participants and personnel

Only one trial had adequate blinding of participants and low risk of bias (Basinga 2011). Thirty‐five trials had unclear risk of bias; these trials were either not blinded or the blinding of participants was not reported. (Azad 2010; Barber 2008; Baqui 2008; Bhutta 2011; Darmstadt 2010; Fottrell 2013; Kenyon 2012; Kirkwood 2013; Klerman 2001; Kumar 2008; Laken 1995; le Roux 2013; Lewycka 2013a; Lewycka 2013b; Lund 2012; Majoko 2007; Manandhar 2004; Melnikow 1997; Midhet 2010; More 2012; Mori 2015; Morris 2004a; Morris 2004b; Mullany 2007; Omer 2008; Penfold 2014; Persson 2013; Richter 2014; Tripathy 2010; Villar 1992; Villar 2001 (WHO 2001); Walker 2013; Wahlstrom 2011; Waiswa 2015; Wu 2011).

Blinding of outcome assessors

Twelve trials were at low risk of bias due to explicit blinding of data collectors and interviewers (Basinga 2011; Bhutta 2011; Fottrell 2013; Kenyon 2012; Klerman 2001; le Roux 2013; Lewycka 2013a; Lewycka 2013b; Manandhar 2004; Melnikow 1997; More 2012; Villar 1992). Twenty‐four trials were judged to have unclear risk of bias due to either no mention of blinding or lack of blinding of those collecting data (Azad 2010; Baqui 2008; Barber 2008; Darmstadt 2010; Kirkwood 2013; Kumar 2008; Laken 1995; Lund 2012; Majoko 2007; Midhet 2010; Mori 2015; Morris 2004a; Morris 2004b; Mullany 2007; Omer 2008; Penfold 2014; Persson 2013; Richter 2014; Tripathy 2010; Villar 2001 (WHO 2001); Wahlstrom 2011; Waiswa 2015; Walker 2013; Wu 2011).

Incomplete outcome data

Twenty‐three trials were at low risk of bias with clear reporting of losses to follow‐up (Basinga 2011; Bhutta 2011; Darmstadt 2010; Fottrell 2013; Kenyon 2012; Kirkwood 2013; Klerman 2001; Kumar 2008; Lund 2012; Majoko 2007; Manandhar 2004; Midhet 2010; More 2012; Mori 2015, Morris 2004a; Morris 2004b; Mullany 2007; Penfold 2014, Tripathy 2010; Villar 1992; Villar 2001 (WHO 2001), Wahlstrom 2011; Waiswa 2015). Risk of bias was unclear in 11 trials due to confusing reporting or lack of reporting of follow‐up (Azad 2010; Baqui 2008; Barber 2008; Laken 1995; le Roux 2013; Lewycka 2013a; Lewycka 2013b; Omer 2008; Persson 2013; Walker 2013; Wu 2011). One small trial was at high risk of bias due to 32% attrition (22/69) in the intervention arm lost to follow‐up and excluded (Melnikow 1997). Another trial (Richter 2014) also had poor follow‐up and was assessed as of high risk of bias (follow‐up: 70% at post‐birth interview; 57% at six months; 24% at 12 months).

Selective reporting

Twenty‐four studies were at low risk of bias (Azad 2010; Baqui 2008; Barber 2008; Basinga 2011; Bhutta 2011; Kenyon 2012; Kirkwood 2013; Klerman 2001; Kumar 2008; le Roux 2013; Lewycka 2013a; Lewycka 2013b; Lund 2012; Majoko 2007; Manandhar 2004; More 2012; Mori 2015, Morris 2004a; Morris 2004b; Mullany 2007; Persson 2013; Villar 1992; Villar 2001 (WHO 2001); Wahlstrom 2011), and in 11, risk of bias was unclear (Darmstadt 2010; Fottrell 2013; Laken 1995; Melnikow 1997; Midhet 2010; Omer 2008; Penfold 2014; Richter 2014; Tripathy 2010; Walker 2013; Wu 2011). One trial (Waiswa 2015) was assessed as of high risk because the authors have not yet published mortality data; preterm birth was also mentioned in the publication but no data reported; finally, data for low birthweight were not reported according to intervention arm in Waiswa 2015.

Other potential sources of bias

Recruitment bias

Twenty‐six trials were at low risk of bias (Baqui 2008; Barber 2008; Basinga 2011; Bhutta 2011; Darmstadt 2010; Fottrell 2013; Kenyon 2012; Kirkwood 2013; Klerman 2001; Kumar 2008; Laken 1995; Lewycka 2013a; Lund 2012; Majoko 2007; Manandhar 2004; Melnikow 1997; Morris 2004a; Mullany 2007; Omer 2008; Penfold 2014; Persson 2013; Tripathy 2010; Villar 1992; Villar 2001 (WHO 2001); Waiswa 2015; Walker 2013). Risk of bias was unclear for six trials (Azad 2010; Midhet 2010; More 2012; Mori 2015; Richter 2014; Wu 2011) and high for two trials: le Roux 2013 (recruiters were redeployed to control clusters due to 22% fewer women in standard care) and Wahlstrom 2011 (authors reported problems with recruitment in the intervention clusters for the post‐intervention survey).

Analysis bias

Twenty‐two trials were at low risk of bias with clear reporting of methods used for adjustments and analysis (Azad 2010; Baqui 2008; Barber 2008; Bhutta 2011; Kenyon 2012; Kirkwood 2013; Kumar 2008; Lewycka 2013a; Lewycka 2013b; Lund 2012; Majoko 2007; Melnikow 1997; More 2012; Mori 2015; Morris 2004a; Morris 2004b: Mullany 2007; Penfold 2014; Persson 2013; Richter 2014; Villar 2001 (WHO 2001); Waiswa 2015). Risk of bias was unclear in 13 trials, largely due to missing ICCs and unclear methods of adjusting for clustering (Basinga 2011; Darmstadt 2010; Fottrell 2013; Klerman 2001; Laken 1995; le Roux 2013; Manandhar 2004; Midhet 2010; Tripathy 2010; Villar 1992; Wahlstrom 2011; Walker 2013; Wu 2011), and high in one trial, where results were not reported according to the randomisation group (Omer 2008).

Other bias

In this section we assessed the trials for issues such as baseline imbalances, respect of allocation assignment, similarities between clusters, information bias, parallel non‐trial interventions, changes in general living conditions and implementation of the intervention. Thirteen trials did not seem to have any other sources of bias (Baqui 2008; Barber 2008; Bhutta 2011; Darmstadt 2010; Kirkwood 2013; Klerman 2001; Kumar 2008; Majoko 2007; Manandhar 2004; Melnikow 1997; Mullany 2007; Richter 2014; Villar 2001 (WHO 2001)). In 17 trials it was unclear how these issues might have introduced bias (Azad 2010; Fottrell 2013; Kenyon 2012; Laken 1995; le Roux 2013; Lewycka 2013a; Lewycka 2013b; Lund 2012; Midhet 2010; More 2012; Mori 2015; Omer 2008; Penfold 2014; Persson 2013; Tripathy 2010; Villar 1992; Waiswa 2015). Five trials were at high risk of bias from one or more of these sources: Basinga 2011 (allocation assignment was not respected due to government restructuring); Morris 2004a and Morris 2004b (unsuccessful implementation of service‐level package in relevant clusters); Wahlstrom 2011 (baseline differences and recall bias for ANC visits outcome); Walker 2013 (baseline differences, recall and reporting bias); and Wu 2011 (baseline differences, unclear implementation)

Effects of interventions

See: Table 1; Table 2

Summary of findings for the main comparison. One intervention versus no intervention.

Comparison 1: One intervention versus no intervention
Patient or population: improving antenatal care coverage and health outcomes among pregnant women  Setting: Argentina, Bangladesh, Brazil, Cuba, Ghana, Honduras, India, Malawi, Mexico, Mongolia, Nepal, Rwanda, South Africa, Tanzania, UK, Vietnam, Zanzibar, Zimbabwe  Intervention: One intervention  Comparison: No intervention
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect  (95% CI)	№ of participants  (studies)	Quality of the evidence  (GRADE)	Comments
	Risk with no intervention	Risk with One intervention
ANC coverage: four or more visits	Moderate		Average OR 1.11  (1.01 to 1.22)	45022  (10 RCTs)	⊕⊕⊕⊕  HIGH¹	This is the primary analysis, ICC 0.02.
	529 per 1000	555 per 1000  (531 to 578)
Pregnancy‐related deaths	Moderate		Average OR 0.69  (0.45 to 1.08)	114930  (10 RCTs)	⊕⊕⊝⊝  LOW ^{2 3}
	700 per 1000000	483 per 1000000  (315 to 756)
ANC coverage: one or more visits	Moderate		Average OR 1.68  (1.02 to 2.79)	19281  (6 RCTs)	⊕⊕⊕⊝  MODERATE ⁴
	490 per 1000	617 per 1000  (495 to 728)
Deliveries in a health facility	Moderate		Average OR 1.08  (1.02 to 1.15)	74299  (10 RCTs)	⊕⊕⊕⊕  HIGH
	645 per 1000	662 per 1000  (650 to 676)
Perinatal mortality	Moderate		Average OR 0.96  (0.89 to 1.03)	189164  (15 RCTs)	⊕⊕⊕⊝  MODERATE ^{2 5}
	40 per 1000	38 per 1000  (36 to 41)
Low birthweight	Moderate		Average OR 0.94  (0.82 to 1.06)	27154  (5 RCTs)	⊕⊕⊕⊕  HIGH
	125 per 1000	118 per 1000  (105 to 132)
Intermittent Prophylactic Treatment for malaria	Study population		not pooled	00  (0 study)		No trial included in this review reported this outcome.
	not pooled	not pooled
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk ratio; OR: Odds ratio. Denominators for the calculation of the absolute comparative effects have been taken from individual trial reports or from Prost 2013. Where different denominators are stated in different reports, we have taken the larger. The median control group risk has been calculated from event and participant raw data, where this was available. If we found no raw event and participant data in published reports, these trials were not included in the calculation of the median control group risk. Both the participant totals and the median control group risk are for illustrative purposes only. In the majority of the trials in this review, the final odds ratio presented will not correspond with raw event and participant data due to adjustments made for the effects of cluster design. We have designated the control risk as moderate because it is based on the median of a wide range of baseline rates in control groups.
GRADE Working Group grades of evidence   High quality: We are very confident that the true effect lies close to that of the estimate of the effect  Moderate quality: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different  Low quality: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect  Very low quality: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

Comparison 2: Combination of interventions versus no intervention
Patient or population: improving antenatal care coverage and health outcomes among pregnant women  Setting: Eastern China, Honduras, India, Laos, Malawi, Pakistan, South Africa, USA  Intervention: Combination of interventions  Comparison: No intervention
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect  (95% CI)	№ of participants  (studies)	Quality of the evidence  (GRADE)	Comments
	Risk with no intervention	Risk with Combination of interventions
ANC coverage: four or more visits	Moderate		Average OR 1.48  (0.99 to 2.21)	7840  (6 RCTs)	⊕⊕⊝⊝  LOW ^{1 2 3}	This is the primary analysis, ICC 0.02.
	430 per 1000	528 per 1000  (428 to 625)
Pregnancy‐related deaths	Moderate		Average OR 0.70  (0.39 to 1.26)	13756  (3 RCTs)	⊕⊕⊕⊝  MODERATE ³
	600 per 100000	421 per 100000  (235 to 755)
ANC coverage: one or more visits	Moderate		Average OR 1.79  (1.47 to 2.17)	12426  (5 RCTs)	⊕⊕⊕⊝  MODERATE ³
	580 per 1000	712 per 1000  (670 to 750)
Deliveries in a health facility	Moderate		Average 1.53  (0.96 to 2.43)	12314  (5 RCTs)	⊕⊕⊕⊝  MODERATE ³
	165 per 1000	252 per 1000  (158 to 401)
Perinatal mortality	Moderate		Average 0.74  (0.57 to 0.95)	39130  (5 RCTs)	⊕⊕⊕⊝  MODERATE ⁴
	90 per 1000	67 per 1000  (51 to 58)
Low birthweight	Moderate		Average 0.61  (0.46 to 0.80)	2084  (2 RCTs)	⊕⊕⊕⊝  MODERATE ¹
	165 per 1000	101 per 1000  (76 to 132)
Intermittent Prophylactic Treatment for malaria	Study population		not pooled	00  (0 study)		No trial eligible for this comparison reported this outcome
	not pooled	not pooled
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).CI: Confidence interval; RR: Risk ratio; OR: Odds ratio. Denominators for the calculation of the absolute comparative effects have been taken from individual trial reports or from Prost 2013. Where different denominators are stated in different reports, we have taken the larger. The median control group risk has been calculated from event and participant raw data, where this was available. If we found no raw event and participant data in published reports, these trials were not included in the calculation of the median control group risk. Both the participant totals and the median control group risk are for illustrative purposes only. In the majority of the trials in this review, the final odds ratio presented will not correspond with raw event and participant data due to adjustments made for the effects of cluster design. We have designated the control risk as moderate because it is based on the median of a wide range of baseline rates in control groups.
GRADE Working Group grades of evidence   High quality: We are very confident that the true effect lies close to that of the estimate of the effect  Moderate quality: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different  Low quality: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect  Very low quality: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

Study	Measure of effect	Statistical approach¹	Intervention	Control
Barber 2008	Beta coefficient: 0.0235 (Cash transfer, instrumental variable model; p. 1411, Barber 2008)	Calculate exp(beta) to get OR and CI	712	180
Basinga 2011	Beta (95% CI): 0.008 (‐0.063 to 0.079)	Calculate exp(beta) to get OR and CI	Reported only the total n, stated as = 2223
Fottrell 2013	Adjusted Odds Ratio (95%CI) 1.37 (0.99 to 1.88)	Cluster adjusted, straight into RevMan	9106	8834
Kenyon 2012	RR = 1.01 (95% CI 0.91 to 1.13)	Non‐cluster trial, calculated OR	322/599	320/604
Kirkwood 2013	RR 1·02 (0·96 to 1·09)	Adjusted using ICC 0.02	5975/7859	5988/8121
Lund 2012	Adjusted OR 2.39 (1.03 to 5.55) adjusted for cluster effect and significant variables	Cluster adjusted, straight into RevMan	574/1311	385/1239
Mori 2015	Adjusted OR 1.25 (0.31 to 5.00) trial statistician (H Noma) calculated unpublished OR for this systematic review. OR adjusted for cluster effect and significant variables.	Cluster adjusted, straight into RevMan	243/252	237/248
Morris 2004   (Household package arm and service package arm added together)	Calculated from change scores, Table 2 Program effects p. 2034 of main report.	Adjusted using ICC 0.02	166/293 + 112/232	151/313
Walker 2013	Adjusted OR 95% CI 1.8(1.2 to 2.8). OR adjusted various characteristics and cluster ('in accordance with WHO standards' from Table 3, p. 1203 Walker 2013)	Cluster adjusted, straight into RevMan	78/1129	39/924

Trial	Measure of effect	Statistical approach¹	Intervention	Control
Kenyon 2012	RR 0.92 (0.74, 1.14) P = 0.43	Not cluster trial ‐ calculated OR	127/605	141/613
Mori 2015	Adjusted OR 0.65 (0.26 to 1.59) trial statistician (H Noma) calculated unpublished OR adjusted for cluster effects and significant variables	Straight into RevMan	8/251	12/247
Richter 2014		Adjusted	67/377	52/414
Villar 1992	OR 0.88 (95% CI 0.67 to 1.16)	Straight into RevMan	115/1033	130/1040
Villar 2001		Adjusted	910/11534	852/11040

Trial	Effect Measure	Statistical approach¹	Intervention	Control
Baqui 2008		Adjusted	2552/32279	1260/15914
Darmstadt 2010		Adjusted	224/4800	255/5472
Fottrell 2013	Risk Ratio (95% CI) adjusted without Tea‐garden residents 0.87 (0.62 to 1.22)	Calculated OR	474/9106	503/8834
Kenyon 2012	RR 2.04 (0.51 to 8.12) P = 0.30	Not a cluster trial. Calculated adjusted OR	6/604	3/616
Kirkwood 2013		Adjusted	288/8035	355/8294
Lewycka 2013a peer counselling only		Adjusted	150/4690	103/2529
Lewycka 2013 women's groups only		Adjusted	173/4773	104/2530
Lund 2012	Adjusted OR 0.50 (0.27 to 0.93) adjusted for clustering and covariates associated with perinatal mortality	Straight into RevMman	25/1300	44/1236
Majoko 2007	OR 1.11 (0.89 to 1.39)		185/6614	161/6384
Manandhar 2004		Adjusted as below	123/2972	147/3303
More 2012		Adjusted	205/8017	173/7844
Mori 2015	Unpublished OR 0.49 (0.05 to 4.54) provided by trial statistician (H Noma) adjusted for cluster effect and significant variables	Straight into RevMan	1/253	2/248
Persson 2013		Adjusted	283/11906	290/10655
Villar 1992	OR 0.89 (95% CI 0.55 to 1.47)	Straight into RevMan	3.6% of 1033	4% of 1040
Villar 2001 (WHO 2001)	From Vogel 2013 Adjusted RR 1.18 (1.01 to 1.37). Adjusted for clustering and for risk strata, smoking, education less than primary, hospital admission in last pregnancy, previous surgery on reproductive tract, late booking, vaginal bleeding in the first trimester, age, previous low birth weight, parity.	Calculated adjusted OR	234/11672	190/11121

Study	Measure of effect	Statistical approach	Intervention	Control
Kumar 2008		Adjusted	12/2749	10/1142
Laken 1995	0.9713 (0.0191 to 49.4211)	Straight into RevMan	0/104	0/101
Lewycka 2013		Adjusted	23/4601	29/5059

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 ANC coverage: four or more visits	10		Odds Ratio (Random, 95% CI)	Subtotals only
1.1 Primary analysis (ICC 0.02 for studies Kirkwood and Morris)	10	45022	Odds Ratio (Random, 95% CI)	1.11 [1.01, 1.22]
1.2 Sensitivity analysis using ICC 0.08	10	45022	Odds Ratio (Random, 95% CI)	1.11 [1.00, 1.22]
2 Pregnancy‐related deaths	10	114930	Odds Ratio (Random, 95% CI)	0.69 [0.45, 1.08]
3 ANC coverage: one or more visits	6		Odds Ratio (Random, 95% CI)	1.68 [1.02, 2.79]
4 Pregnant women initiating ANC in first trimester	1		Odds Ratio (Random, 95% CI)	1.20 [0.99, 1.45]
5 Pregnant women receiving ANC from health professional	1		Odds Ratio (Random, 95% CI)	1.13 [0.84, 1.52]
6 Deliveries in a health facility	10		Odds Ratio (Random, 95% CI)	1.08 [1.02, 1.15]
7 Intermittent Prophylactic Treatment for malaria	0	0	Odds Ratio (M‐H, Random, 95% CI)	0.0 [0.0, 0.0]
8 Proportion of women with tetanus protection at birth	8		Odds Ratio (Random, 95% CI)	1.03 [0.92, 1.15]
9 Proportion of women treated for syphilis	2		Odds Ratio (Random, 95% CI)	1.46 [0.94, 2.26]
10 Proportion of women with HIV who receive a complete antiretroviral course for prevention of mother‐to‐child transmission of HIV	1		Odds Ratio (Random, 95% CI)	0.44 [0.26, 0.74]
11 Preterm labour	4		Odds Ratio (Random, 95% CI)	1.00 [0.93, 1.09]
12 Low birthweight	5		Odds Ratio (Random, 95% CI)	0.94 [0.82, 1.06]
13 Perinatal mortality	15		Odds Ratio (Random, 95% CI)	0.96 [0.89, 1.03]

Outcome or subgroup title	Statistical method	Effect size
1 ANC coverage: four or more visits	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
2 Pregnancy‐related deaths	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
3 ANC coverage: one or more visits	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
4 Deliveries in a health facility	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
5 Perinatal mortality	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
6 Low birthweight	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
7 Intermittent Prophylactic Treatment for malaria	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 ANC coverage: four or more visits	2		Odds Ratio (Random, 95% CI)	0.99 [0.70, 1.40]
2 Pregnancy‐related deaths	2		Odds Ratio (Random, 95% CI)	1.00 [0.52, 1.96]
3 ANC coverage: one or more visits	3		Odds Ratio (Random, 95% CI)	0.86 [0.61, 1.20]
4 Deliveries in a health facility	3		Odds Ratio (Random, 95% CI)	0.95 [0.69, 1.30]
5 Perinatal mortality	2		Odds Ratio (Random, 95% CI)	0.88 [0.72, 1.07]
6 Proportion of women with tetanus protection at birth	2		Odds Ratio (Random, 95% CI)	1.07 [0.80, 1.43]
7 Low birthweight	0	0	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
8 Intermittent Prophylactic Treatment for malaria	0	0	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 ANC coverage: four or more visits	1	383	Odds Ratio (M‐H, Random, 95% CI)	0.77 [0.41, 1.43]
2 Pregnancy‐related deaths	0	0	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
3 ANC coverage: one or more visits	0	0	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
4 Deliveries in a health facility	1	383	Odds Ratio (M‐H, Random, 95% CI)	1.55 [0.71, 3.37]
5 Perinatal mortality	0	0	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
6 Low birthweight	0	0	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
7 Intermittent Prophylactic Treatment for malaria	0	0	Odds Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]

Outcome or subgroup title	No. of studies	Statistical method	Effect size
1 Health systems vs Population ANC coverage: four or more visits	9	Odds Ratio (Random, 95% CI)	Subtotals only
1.1 Health system interventions	5	Odds Ratio (Random, 95% CI)	1.13 [0.96, 1.34]
1.2 Population interventions	4	Odds Ratio (Random, 95% CI)	1.04 [0.96, 1.13]
2 Health systems vs Population Pregnancy‐related deaths	11	Odds Ratio (Random, 95% CI)	Subtotals only
2.1 Health system interventions	4	Odds Ratio (Random, 95% CI)	1.22 [0.41, 3.65]
2.2 Population intervention	7	Odds Ratio (Random, 95% CI)	0.69 [0.46, 1.03]
3 Country Income Low vs High ANC at least 4 visits	18	Odds Ratio (Random, 95% CI)	1.14 [1.04, 1.25]
3.1 Low or lower middle income countries	11	Odds Ratio (Random, 95% CI)	1.21 [1.04, 1.40]
3.2 High or higher middle income countries	7	Odds Ratio (Random, 95% CI)	1.12 [0.95, 1.32]

Methods	Parallel arm cluster‐RCT conducted at 18 sites in Bangladesh between Feb 2005 and Dec 2008.
Participants	Sample size: 18 clusters (6389 women). Clusters: purposive sampling was performed in 3 different divisions in Bangladesh on the basis of the districts having active Diabetic Association of Bangladesh (BADAS) offices. Within these districts, sub districts (upazilas) and unions (the lowest level administrative units in rural Bangladesh) were also purposefully sampled by use of recommendations from BADAS representatives, the main criteria being perceived limited access to perinatal health care in those unions, and a feasible travelling distance from BADAS district headquarters. Individuals: women were eligible to participate in the study if they were aged 15–49 years, residing in the project area, and had given birth during the study period.
Interventions	Target: health system (re‐organisation of health services intervention) and community (education or IEC intervention). Arm 1 (9 clusters, 17,514 births ITT): in intervention clusters, a facilitator convened 18 groups every month to support participatory action and learning for women, and to develop and implement strategies to address maternal and neonatal health problems. 5 of the 9 clusters became TBA intervention clusters and 4 became controls. 482 TBAs were given basic training in undertaking clean and safe deliveries, providing safe delivery kits, recognising danger signs in mothers and infants, making emergency preparedness plans, accompanying women to facilities, and undertaking mouth‐to‐mouth resuscitation. They also received additional training in neonatal resuscitation with bag valve‐mask. Arm 2 (9 clusters, 18,599 births ITT): health services strengthening intervention and basic training of TBAs.
Outcomes	Trial primary outcome: neonatal mortality rate. Review outcomes reported: Primary: ANC coverage (at least 4 visits), maternal mortality. Secondary: health facility deliveries, tetanus protection, perinatal mortality, neonatal mortality.  Follow‐up: outcomes measured at 1, 2, and 3 years. We have used mortality data from Table 2 (Azad 2010 p. 1197). We used Years 1‐3 combined, excluding the "temporary and tea garden residents" who may not have received the full intervention. We calculated our own cluster adjusted ORs for antenatal care outcomes using the percentages from Table 4, p. 1200 and the denominators from years 1‐3 in Table 2, p. 1197 (all births: intervention n = 15,696 and control n = 15,257).
Notes	Funders: Women and Children First, the UK Big Lottery Fund, Saving Newborn Lives, and the UK Department for International Development.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"The allocation sequence was decided upon by the project team before drawing" pg 1194 "and was based on clusters rather than  individuals."
Allocation concealment (selection bias)	Unclear risk	Not clear how allocation was concealed.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Not reported.
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Not reported.
Recruitment bias (for cluster RCTs)	Unclear risk	"Additionally, about 10% of mothers in our study area were temporary residents and mainly came into the cluster areas to give birth, since the tradition is for women to go to their mothers’ home just before delivery. These temporary residents were not exposed to the women’s group intervention, and often had returned to their marital homes outside the study area before the postnatal interview." Presumably this would have affected all clusters.
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Not reported.
Selective reporting (reporting bias)	Low risk	Most relevant outcomes reported.
Analysis bias	Low risk	Analysis appropriate for clusters; ICC reported; ITT analysis performed.
Other bias	Unclear risk	Baseline imbalances not reported.
Overall risk assessment	Low risk	No serious risk of bias concerns.

Methods	Parallel 3‐arm cluster‐RCT conducted at 24 sites in Bangladesh (Sylhet district) between Jul 2003 and Dec 2005.
Participants	Sample size: 24 clusters (113816 women; 46,444 live births analysed). Clusters: 24 clusters (with a population of about 20,000 each) in Sylhet district, a district with poor access to health care, about 15,000 livebirths per year, and the presence of non‐government organisations with the ability to scale‐up the intervention. The area also has the highest neonatal mortality in Bangladesh. Individuals: ever‐married women of reproductive age (15–49 years old).
Interventions	Target: health system (addition of home visits) and community (IEC). Arm 1 (8 clusters, 36,059 women): (Home care) the CHWs identified pregnancies through routine surveillance during visits to each household once every 2 months; promoted birth and newborn care preparedness through 2 scheduled antenatal and 3 early postnatal home visits; and provided iron and folic acid supplements during birth and newborn‐care preparedness visits. Arm 2 (8 clusters, 40,159 women): (Community care) in the community‐care arm, female volunteers called community resource people were recruited in each village to identify pregnant women, encourage them to attend community meetings held by the community mobilisers, receive routine ANC, and seek care for signs of serious illness in mothers or newborns. Arm 3 (8 clusters, 37,598 women): (Usual care) families received the usual health services provided by the government, non‐government organisations, and private providers.  Refresher training sessions for management of maternal and newborn complications were provided for government health workers in all 3 study arms.
Outcomes	Trial primary outcome: reduction in neonatal mortality. Review outcomes reported: Primary: not reported. Secondary: tetanus protection, at least 1 ANC visit, neonatal mortality.  Follow‐up: 9, 16, and 24 months. We have analysed these data by combining the 2 arms with individual interventions (home care or community care) compared to the control arm of standard care. Outcome data are included in our Comparison 1.
Notes	Funders: United States Agency for International Development and saving newborn lives programme by Save the Children (US) with a grant from Bill and Melinda Gates Foundation.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"computer generated pseudo random number sequence."
Allocation concealment (selection bias)	Low risk	"The computer‐generated randomisation was implemented by a study investigator who had no role in the implementation of the study."
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	"nature of the intervention meant masking was unachievable."
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Not reported.
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Missing data described in study flow chart.
Selective reporting (reporting bias)	Low risk	Most relevant outcomes reported.
Analysis bias	Low risk	Analysis appropriate for clusters; ICC reported; ITT analysis performed.
Other bias	Low risk	No baseline imbalances.
Overall risk assessment	Low risk	No serious risk of bias concerns.

Methods	Parallel‐arm cluster‐RCT conducted at 506 sites in Mexico between 1997 and 2003.
Participants	Sample size: 506 clusters randomised (individuals not reported), 173 clusters analysed. Clusters: "The rural programme established eligibility in two stages: poor communities were first identified, and low‐income households were identified within those communities". Communities and households were randomly selected based on a probability sample proportionate to the number of women of reproductive age women (15–49 years). Individuals: the sample included women who experienced a singleton live birth between 1997 and 2003, were designated as poor and eligible for Oportunidades, and lived in the original treatment and control communities
Interventions	Target: community (financial incentive intervention). Arm 1 (97 clusters, 810 women): Progresa or Opportunidades is a conditional cash transfer program established in 1997 in Mexico, with the dual aim of immediate poverty relief and long‐term impact on the generational transfer of poverty. Every 2 months intervention families received a cash transfer representing approximately a 25% increase in household income (Gertler 2000, p. 3). The cash transfer required specific health behaviours of all members of households. Pregnant women were required to have 5 prenatal visits beginning in the first trimester of pregnancy. Beneficiary births are those births that occurred after the household received their first cash transfer. Households in intervention areas began receiving benefits during the summer of 1998. Arm 2 (61 communities, 215 women): non‐beneficiary births are those that occurred among eligible women prior to receiving the first cash transfer. Households in control clusters began receiving benefits in November 1999.
Outcomes	Trial primary outcome: birthweight. Review outcomes reported: Primary: ANC coverage (at least 4 visits). Secondary: at least 1 ANC visit, health facility deliveries, tetanus protection, low birthweight infants.  Follow‐up: once.
Notes	The Mexican social welfare program Oportunidades (now Prospera) has multiple citations. We have incorporated data from a specific analysis conducted on a small sample of women in households involved in this large poverty relief program (Barber 2008). Funders: National Institutes of Health Fogarty International Center TW006084 and National Institute of Child Health and Human Development.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	For the initial cluster‐randomisation, "random assignment was generated at the community level without weighting by use of the randomisation commands in Stata version 2.0" (Fernald 2008). For the survey, areas were randomly assigned "based on a probability sample proportionate to the number of women of reproductive age". p. 20 Barber 2009.
Allocation concealment (selection bias)	Unclear risk	Not reported.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Due to the nature of the intervention participants could not be blinded.
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Not reported.
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Loss to follow‐up described but sample sizes vary in the different reports.
Selective reporting (reporting bias)	Low risk	Most relevant outcomes reported.
Analysis bias	Low risk	Analysis appropriate for clusters; ITT analysis performed.
Other bias	Low risk	No baseline imbalances.
Overall risk assessment	Low risk	No serious risk of bias concerns.

Methods	Parallel‐arm cluster‐RCT conducted at 166 sites in Rwanda between June 2006 and Oct 2006.
Participants	Sample size: 166 clusters (2563 women). Clusters: districts without pre‐existing P4P schemes managed by non‐governmental organisations. Individuals: not described.
Interventions	Target: health system (financial intervention). Arm 1 (80 clusters, 1242 women): P4P scheme to supplement primary health centres’ input‐based budgets. In this P4P scheme, payments are made directly to facilities and are used at each facility’s discretion. Arm 2 (86 clusters, 1321 women): control facilities would continue to receive traditional input‐based financing for an additional 23 months until the rollout of the scheme was complete.
Outcomes	Trial primary outcomes: prenatal care visits and institutional deliveries. Review outcomes reported: Primary: ANC coverage (at least 4 visits). Secondary: ANC coverage (at least 1 visit), health facility deliveries, tetanus protection, use of child preventative care.  Follow‐up: baseline and 25 months.
Notes	Funders: World Bank’s Bank‐Netherlands Partnership Program, the British Economic and Social Research Council, the Government of Rwanda, and the World Bank’s Spanish Impact Evaluation Fund; Global Development Network and the MacNamara Foundation.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"The remaining districts were then grouped into eight blocks based on rainfall, population density, and livelihood data from the 2002 Census.15 Blocks covered between two and 4 districts, depending on district characteristics and size. The blocks were then divided into two sides, and one side of each block was randomly assigned to either the intervention or control group. Randomisation was done by coin toss."
Allocation concealment (selection bias)	Unclear risk	Not reported.
Blinding of participants and personnel (performance bias)   All outcomes	Low risk	Women interviewed in households would not have been aware of their local facility's group assignment. Women attending facilities should also not have been aware of the funding scheme in operation at her local health clinic.
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	"All surveys were done by trained enumerators hired by external firms specialised in data collection who were masked to whether they were interviewing in an intervention or control area."
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	2.1 % of intervention and 1.9% of control households refused to participate. 12% loss to follow‐up between baseline and end of trial surveys. 11.8% attrition in each treatment arm between first and second interviews. Incomplete household surveys were dropped from the sample after each round.
Selective reporting (reporting bias)	Low risk	Most relevant outcomes are reported.
Analysis bias	Unclear risk	Analysis appropriate for clusters, ICC and ITT not reported.
Other bias	High risk	Allocation assignment not respected due to government restructuring.
Overall risk assessment	Unclear risk	Due to uncertainties raised above.

Methods	Parallel arm cluster‐RCT conducted in Mirzapur, Bangladesh, between Dec 2003 and Dec 2006.
Participants	Sample size: 12 clusters (21,140 individuals randomised, 10,700 women with at least 1 pregnancy during 10 preceding months analysed). Clusters: rural unions surrounding an urban central union (excluded from the study) served by a 750 bed private referral‐level hospital. Individuals: all married women of reproductive age (i.e. 15–49 years) in the intervention arm were eligible for enrolment. Women in the survey were eligible if they had had a pregnancy outcome in the last 3 years.
Interventions	Target: health system (addition of home visits). Arm 1: 2 home visits (12‐16 and 32‐34 weeks); they were given a labour card for women to present upon arrival at hospital for delivery and 3 postnatal visits on days 2, 5 and 8. CHWs facilitated free‐of‐charge transfer of ill neonates to hospital.The purpose of the antenatal component of the intervention was to increase uptake of ANC (3 visits taking place at home or at a health centre or satellite clinic ‐ distinct from the 2 antenatal CHW home visits), tetanus toxoid vaccination, general pregnancy and newborn care education, and birth preparedness (including delivery at a health facility). Arm 2: standard ANC.
Outcomes	Trial primary outcomes: antenatal and immediate newborn care behaviours, knowledge of danger signs, care seeking for neonatal complications, and neonatal mortality. Review outcomes reported: Primary: not reported. Secondary: ANC coverage (at least 1 visit), health facility delivery, IPT for malaria, neonatal mortality.  Follow‐up: data collection at delivery and during pre and postnatal home visits. Endline survey Jan ‐ May 2006, before the end of the trial.
Notes	Funders: The Wellcome Trust: Burroughs Wellcome Fund Infectious Disease Initiative 2000 and the Office of Health, Infectious Diseases and Nutrition, Global Health Bureau, United States Agency for International Development (USAID) through the Global Research Activity Cooperative agreement with the Johns Hopkins Bloomberg School of Public Health (award HRN‐A‐00‐96‐90006‐00). Support for data analysis and manuscript preparation was provided by the Saving Newborn Lives program through a grant by the Bill & Melinda Gates Foundation to Save the Children‐US. The study was registered at clinicaltrials.gov, No. NCT00198627.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation of clusters with a computer‐generated randomisation sequence.
Allocation concealment (selection bias)	Unclear risk	Not described.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Not blinded.
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Not blinded.
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Study flow chart with details of exclusions. Response rate for the endline survey reported as 87.8% (11731/16771).
Selective reporting (reporting bias)	Unclear risk	Data for proportion of women who received 2 tetanus toxoid immunisations were not reported. The authors report falling rates during the trial and attribute this to a national shortage of vaccine.
Analysis bias	Unclear risk	Method of analysing clusters not clearly described apart from stating ITT analysis performed.
Other bias	Low risk	Baseline characteristics similar between arms.
Overall risk assessment	Unclear risk	Due to uncertainties raised above.

Methods	Parallel‐arm cluster‐RCT conducted at 18 sites/unions in Bangladesh between Jan 2009 and June 2011.
Participants	Sample size: 18 clusters (532,996 population). Clusters: purposeful selection of the 3 districts on the basis of having active Diabetic Association of Bangladesh offices and somewhat representing the social and geographical diversity of Bangladesh..basis of perceived limited access to perinatal health care and feasible accessibility from Diabetic Association of Bangladesh district headquarters. Individuals: women whose childbirths or deaths were recorded in the study areas.
Interventions	Target: community (IEC intervention). Arm 1 (9 clusters, 12,135 women/births): women's participation groups; effect of monthly participatory learning and action cycle focus on maternal and newborn health. Arm 2 (9 clusters, 13,459 women/births): control not described (presumably no women's participation groups).
Outcomes	Trial primary outcome: neonatal mortality rate. Review outcomes reported: Primary: ANC coverage (at least 4 visits). Secondary: health facility deliveries, perinatal and neonatal mortality.  Follow‐up: data collected monthly for 24 months. 1 of the control areas (with 3 clusters) included "tea‐garden estates". Residents on these estates were described as having more social and economic disadvantage, and separate analyses were carried out including and excluding these areas. For the analyses in this review, we have used the outcome data that excludes these tea garden residents.
Notes	Funders: Big Lottery Fund International Strategic Grant, Wellcome Trust Strategic Award. For the outcome of perinatal mortality we have used stillbirths plus early neonatal deaths. We calculated our own OR using an ICC because the adjusted perinatal deaths OR (without Tea Garden residents) is asymmetrical and would not go into RevMan. See Fottrell 2013 (Table 3, p. 823).
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Paper stated that the sequence "had been decided before drawing the papers" (containing the allocation).
Allocation concealment (selection bias)	Low risk	Allocated "by blindly pulling pieces of paper, each representing 1 union from a bottle".
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	The intervention was not masked, it is not clear how lack of blinding might affect outcomes reported.
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	The implementation and in‐country monitoring and evaluation teams were blind to the allocation arms" during interim analysis (June 2011).
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Study flow chart included displaying reasons for exclusions. Missing data described as 13% on home delivery practice and 0.8% on other secondary outcomes.
Selective reporting (reporting bias)	Unclear risk	Relevant outcomes reported although separate analyses for some control group births meant that results were more difficult to interpret.
Analysis bias	Unclear risk	Analysis appropriate for clusters but ICC not reported and ITT analysis only performed for primary outcomes.
Other bias	Unclear risk	1 of the control areas (with 3 clusters) included "tea‐garden estates"; residents on these estates were described as having more social and economic disadvantage and separate analyses were carried out including and excluding these areas. For the analyses in this review, we have used the outcome data that excludes these tea garden residents.
Overall risk assessment	Unclear risk	We were uncertain whether some of the above might have significantly biased the results.

Methods	Parallel‐arm individual‐randomised RCT conducted at 3 primary care trusts in Birmingham, UK between Jul 2010 and Oct 2011. Trial name: ELSIPS.
Participants	Sample size: 1324 women. Inclusion criteria: nulliparous women < 28 weeks' gestation assessed by a midwife as having specific social risk. (Risk factors included housing problems, lack of social support, smoking, low maternal weight or obesity, teenage, late booking for ANC.) Exclusion criteria: women under 16 years of age, or teenage mothers recruited to another national trial of additional support during pregnancy.
Interventions	Target: health system (re‐organisation of health services: home visits). Arm 1 (662 women): POW provided support, including home visits, in addition to standard ANC and PNC. The POW organised antenatal visits and advised on lifestyle changes. In addition to emotional and health‐related support, the POW helped with financial, legal or benefits problems and with housing. The POW also provided support with care of the newborn, including breastfeeding. Arm 2 (662 women): women in the control group received standard ANC and PNC.
Outcomes	Trial primary outcome: Edinburgh Postnatal Depression Scale1 (EPDS) 8–12 weeks postpartum and antenatal visits attended. Review outcomes reported: Primary: ANC coverage (at least 10 contacts). Secondary: preterm birth (< 34 weeks), low birthweight infants, perinatal mortality. Other: depression scores.  Follow‐up: intervention involved individual support during pregnancy and follow‐up to 8‐12 weeks postpartum. We did not include data for preterm birth < 34 weeks because our review's definition of preterm birth < 37 weeks. Outcome data from unpublished paper obtained from author: SL Kenyon, s.kenyon@bham.ac.uk.
Notes	Funders: this work was funded by the National Institute for Health Research (NIHR) through the Collaborations for Leadership in Applied Health Research and Care for Birmingham and Black Country (CLAHRC‐BBC) programme. The views expressed in this publication are not necessarily those of the NIHR, the Department of Health.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation generated by the trial statistician using computer‐generated lists with random block sizes stratified by area.
Allocation concealment (selection bias)	Low risk	Telephone randomisation using a registered trial unit ensured allocation concealment.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Inadequate ‐ blinding not possible.
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	Outcomes were recorded in maternity care notes by staff providing care. Those who collected and entered data were blind to group assignment.
Recruitment bias (for cluster RCTs)	Low risk	Not applicable. Not a cluster‐randomised trial.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Data on antenatal outcomes available for 100% and 99%. Data for the EPDS at 8‐12 weeks postpartum were available for 82% and 85% of the intervention and control arms. respectively.
Selective reporting (reporting bias)	Low risk	All outcomes stated in protocol are reported in the unpublished paper, with the exception of "engagement with other services, as required (e.g. smoking cessation services)".
Analysis bias	Low risk	ITT analysis performed.
Other bias	Unclear risk	Baseline characteristics similar between treatment groups.
Overall risk assessment	Low risk	No serious risk of bias concerns for primary outcomes.

Methods	Parallel‐arm cluster‐RCT conducted in Ghana between Nov 2008 and Dec 2009.
Participants	Sample size: 98 clusters (18,609 individuals). Clusters: residential zones. Individuals: all pregnant women and newborn babies living in the Newhints zones, where pregnancies ended between November 2008 and December 2009.
Interventions	Target: health system (added home visits by community‐based surveillance volunteers) and community (IEC). Arm 1 (49 clusters, 9174 women): training of community‐based surveillance volunteers to identify pregnant women in their community and to undertake 2 home visits during pregnancy and 3 visits after birth on days 1, 3, and 7, to promote essential newborn‐care practices, and to assess and refer sick newborn babies. Arm 2 (49 clusters, 9435 women): control (no intervention).
Outcomes	Trial primary outcome: neonatal mortality rate and coverage of key essential newborn‐care practices. Review outcomes reported: Primary: ANC coverage (at least 4 visits). Secondary: health facility deliveries, neonatal mortality. Other: coverage of key essential newborn care practices.  Follow‐up: 2 home visits during pregnancy and 3 visits after birth on days 1, 3, and 7.
Notes	Funders: WHO, Save the Children’s Saving Newborn Lives Programme from the Bill & Melinda Gates Foundation, and the UK Department for International Development.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"computer generated randomisation."
Allocation concealment (selection bias)	Low risk	An independent epidemiologist conducted the randomisation.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Not described.
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Not described.
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	3 groups of pregnancies were not included in the analysis of NMR: 908 (5%) women were lost to follow‐up during pregnancy; 1216 (7%) had pregnancies that ended early and did not result in a livebirth or stillbirth; and 156 (< 1%) women moved, resulting in a change of treatment groups.
Selective reporting (reporting bias)	Low risk	Most relevant outcomes reported.
Analysis bias	Low risk	Analysis appropriate for clusters; ICC reported; ITT analysis performed.
Other bias	Low risk	No baseline imbalances noted.
Overall risk assessment	Low risk	No serious risk of bias concerns.

Methods	Parallel‐arm individually‐randomised RCT conducted in the USA between Mar 94 and Jun 98.
Participants	Sample size: 656 women. Inclusion criteria: African American, eligible for Medicaid, less than 26 weeks' gestation, at least 16 years old, score of 10 or higher on a risk assessment scale. Exclusion criteria: alcoholism and substance abuse, asthma, cancer, diabetes, epilepsy, high blood pressure, sickle cell disease and HIV/AIDS.
Interventions	Target: health system (additional and longer appointments) and community (IEC). Arm 1: educational intervention informing women about their risk conditions and what behaviours might improve their pregnancy outcome. Augmented care included educationally oriented peer groups, additional appointments, extended time with clinicians, and other supports. Arm 2: control (no intervention).
Outcomes	Trial primary outcomes: pregnancy outcomes, women's knowledge of risks, satisfaction with care. Review outcomes reported: Primary: not reported. Secondary: preterm birth, low birthweight infants. Other: average no. of ANC visits.  Follow‐up: every 2 weeks, until the last month of pregnancy then every week.
Notes	Funders: Federal Agency for Health Care Policy and Research to the University of Alabama at Birmingham, USA.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Sequence generation not described.
Allocation concealment (selection bias)	Low risk	"sealed envelopes."
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Not described.
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	"interviewers blinded" to treatment allocation. Additional outcome data taken from clinic records, data collection forms and a computerised database.
Recruitment bias (for cluster RCTs)	Low risk	Not applicable.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Missing data < 10%. 656 women enrolled, but data available for 619; 12 women with fetal deaths excluded from analysis (intervention group: 3 before 20 weeks and 4 after; controls: 3 before 20 weeks and 2 after).
Selective reporting (reporting bias)	Low risk	Relevant outcomes reported.
Analysis bias	Unclear risk	ITT not stated.
Other bias	Low risk	No baseline imbalances.
Overall risk assessment	Low risk	No serious risk of bias concerns.

Methods	Parallel 3‐arm cluster‐RCT conducted in India between Jan 2004 and May 2005.
Participants	Sample size: 39 clusters (3891 individuals analysed). Clusters: administrative units. Individuals: all mothers who had delivered during the study period and were available for interview.
Interventions	Target: health system (home visits) and community (IEC). Arm 1 (13 clusters): a preventive package of interventions for essential newborn care (birth preparedness, clean delivery and cord care, thermal care [including skin‐to‐skin care], breastfeeding promotion, and danger sign recognition). The strategy included 2 prenatal (60 days and 30 days before expected date of delivery) and 2 postnatal (day 0 and day 3) home visits, community meetings and folk‐song meetings, maternal and newborn health stakeholder meetings, and meetings for community volunteers. Arm 2 (13 clusters): received same package of essential newborn care plus use of a liquid crystal hypothermia indicator (ThermoSpot; a sticker that indicates hypothermia in the newborn by changing colour). Arm 3 (13 clusters): received the standard care available from government and NGO providers in the area.
Outcomes	Trial primary outcome: newborn care practices and neonatal mortality rate. Review outcomes reported: Primary: not reported. Secondary: maternal mortality (up to 6 weeks postpartum), ANC coverage (at least 1 visit), health facility deliveries, tetanus protection, stillbirths, neonatal mortality. Other: essential newborn care measures, breastfeeding.  Follow‐up: 2 prenatal assessments at 60 days and 30 days before expected date of delivery; 2 postnatal assessments at day 0 and day 3.
Notes	Funders: The United States Agency for International Development, Delhi Mission, and the Saving Newborn Lives program of Save the Children US through a grant from the Bill and Melinda Gates Foundation. Perinatal mortality included neonatal deaths up to 28 days after birth.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Stratified cluster‐randomisation conducted at Johns Hopkins University using a computer program.
Allocation concealment (selection bias)	Low risk	Allocation performed remotely.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Blinding not possible.
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Preliminary analysis (2005) of neonatal mortality rate was said to be masked.
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Loss to follow‐up described in study flow diagram with missing data < 20%.
Selective reporting (reporting bias)	Low risk	Relevant outcomes reported.
Analysis bias	Low risk	Analysis appropriate for clusters; no ICC reported; ITT analysis performed.
Other bias	Low risk	No baseline imbalances.
Overall risk assessment	Low risk	No serious risk of bias concerns.

PERMALINK

Health system and community level interventions for improving antenatal care coverage and health outcomes

Lawrence Mbuagbaw

Nancy Medley

Andrea J Darzi

Marty Richardson

Kesso Habiba Garga

Pierre Ongolo‐Zogo

Abstract

Background

Objectives

Search methods

Selection criteria

Data collection and analysis

Main results

Authors' conclusions

Plain language summary

Summary of findings

Background

Description of the condition

Measuring antenatal care

Description of the intervention

How the intervention might work

Why it is important to do this review

Objectives

Methods

Criteria for considering studies for this review

Types of studies

Types of participants

Types of interventions

Interventions aimed at the health system

Interventions aimed at the community

Types of outcome measures

Primary outcomes

Secondary outcomes

Search methods for identification of studies

Electronic searches

Searching other resources

Data collection and analysis

Selection of studies

Data extraction and management

Assessment of risk of bias in included studies

(1) Random sequence generation (checking for possible selection bias)

(2) Allocation concealment (checking for possible selection bias)

(3.1) Blinding of participants and personnel (checking for possible performance bias)

(3.2) Blinding of outcome assessment (checking for possible detection bias)

(4) Incomplete outcome data (checking for possible attrition bias due to the amount, nature and handling of incomplete outcome data)

(5) Selective reporting (checking for reporting bias)

(6) Other bias (checking for bias due to problems not covered by (1) to (5) above)

(7) Overall risk of bias

Assessing the quality of the body of evidence using the GRADE approach

Measures of treatment effect

Dichotomous data

Continuous data

Unit of analysis issues

Cluster‐randomised trials

Other unit of analysis issues

Dealing with missing data

Assessment of heterogeneity

Assessment of reporting biases

Data synthesis

Subgroup analysis and investigation of heterogeneity

Sensitivity analysis

Results

Description of studies

Results of the search

1.

Included studies

Interventions

Outcomes

Excluded studies

Risk of bias in included studies

2.

3.

Allocation

Random sequence generation

Allocation concealment

Blinding

Blinding of participants and personnel

Blinding of outcome assessors

Methods	3‐arm, individually‐randomised RCT conducted in Detroit, USA; recruitment dates not reported.
Participants	Sample size: 205 individuals. Inclusion criteria: low‐income women who entered prenatal care at a local clinic before 32 weeks' gestation and who delivered at a tertiary‐level hospital. Exclusion criteria: not reported.
Interventions	Target: community (financial incentive intervention). Arm 1 (51 women): women received gift certificates for each prenatal appointment. Arm 2 (53 women): women received gift certificates and a chance to win in a $100 raffle. Arm 2 (101 women): no financial incentive. Women in all 3 groups were offered $10 for the postnatal interview.
Outcomes	Triap Primary Outcome: kept appointments for antenatal care and postpartum care Review outcomes reported: Primary: ANC coverage (at least 4 visits). Secondary: maternal mortality, health facility deliveries, perinatal mortality.  Follow‐up: 6‐8 weeks postpartum.
Notes	Funders: Michigan Health Care Education and Research Foundation. Authors provided unpublished data for coverage, mortality and health facility deliveries by email on 16/1/15.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"random numbers were used" for group assignment.
Allocation concealment (selection bias)	Unclear risk	Not described.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Clinic staff members were blind to assignment, but women would have been aware of their own assignment.
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Not reported.
Recruitment bias (for cluster RCTs)	Low risk	Not applicable.
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Missing data < 20% for birth outcomes (low risk) but high loss to follow‐up at the postnatal interview (45%).
Selective reporting (reporting bias)	Unclear risk	Relevant outcomes reported, but insufficient data provided.
Analysis bias	Unclear risk	Methods not reported in sufficient detail.
Other bias	Unclear risk	Insufficient information to make a judgement.
Overall risk assessment	Unclear risk	Insufficient information to make a judgement.

Methods	Parallel‐arm cluster RCT conducted at 26 sites in South Africa between May 2009 and Sept 2010.
Participants	Sample size: 26 clusters (24 clusters and 1238 individuals analysed). Clusters: neighbourhoods were matched. Eligible neighbourhoods had 450‐600 households, with formal and informal housing, that were within 5 km of health clinics; had 5 to 7 alcohol bars; were noncontiguous or separated by natural barriers; had similar numbers of child care centres, informal shops, and schools; and had households with similar length of residence. Individuals: pregnant women were recruited at an average 26 weeks of pregnancy (range, 3–40 weeks). 94 women in 10 of the 12 standard care neighbourhoods were enrolled post‐birth. Approximately 25% of women in each treatment arm were living with HIV.
Interventions	Target: health system (added home visits) and community (IEC). Arm 1 (12 clusters, 644 women): Philani Intervention Program, home visits by CHWs in addition to standard care. Arm 2 (12 clusters, 594 women): standard care, comprehensive healthcare at clinics.
Outcomes	Trial primary outcomes: composite of maternal and child health and well being measures. Review outcomes reported: Primary: ANC coverage (at least 4 visits). Secondary: HIV screening, complete antiretroviral course, low birthweight infants.  Follow‐up: once during pregnancy and twice after birth: at 1 week, 6 months; additional trial report for 18 months follow‐up. Mortality data for women and infants are reported in the primary trial report. However, these data represent all deaths within the particular time frame of data collection (e.g. all deaths to 6 months post birth, p. 1464 of the 2013 trial report). We consulted with trial authors, but we cannot recalculate the mortality data to fit standard definitions for pregnancy‐related deaths or perinatal deaths. We were particularly concerned that maternal deaths may have been unrelated to pregnancy. We were therefore unable to use mortality data in meta‐analysis.
Notes	Funders: NIAAA Grant # 1R01AA017104 and supported by NIH grants MH58107, 5P30AI028697, and UL1TR000124. M.T. is supported by the National Research Foundation (South Africa).
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Sequence generation not described. Method described as simple randomisation.
Allocation concealment (selection bias)	Low risk	Neighbourhoods were randomised in matched pairs using simple randomisation. Randomisation conducted by an independent research team (UCLA).
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	"interviewers were blinded but may have known from participants about CHWs."
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	A driver transported all participants to a central assessment site, allowing interviewers to be blind to group allocation.
Recruitment bias (for cluster RCTs)	High risk	"Initially, however, we identified 22% fewer pregnant women in standard care. By redeploying recruiters, we identified an additional 94 women in 10 of the 12 standard care neighbourhoods who were pregnant during the recruitment period (median of 7 late‐entry participants per neighbourhood; range, 3–24). These women were enrolled post‐birth when their infants were a mean age of 9 months old (range, 1–18 months). The final sample (n = 1238) consisted of a median of 51 pregnant women per neighbourhood (range, 23–72)." Analyses were conducted with and without late‐entry participants and results were similar.
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	1 matched cluster pair was excluded after 6 months due to poor recruitment.
Selective reporting (reporting bias)	Low risk	Relevant outcomes were reported.
Analysis bias	Unclear risk	Analysis appropriate for clusters; no ICC reported; ITT analysis not performed.
Other bias	Unclear risk	Not reported.
Overall risk assessment	Unclear risk	We were uncertain what impact potential biases mentioned above had on results.

Methods	2 by 2 factorial cluster‐RCT conducted in Malawi between 2005 and 2009.
Participants	Sample size: 42 clusters (18960 pregnancies, 18,744 livebirths analysed). Clusters: the unit of randomisation was a cluster of villages and not an individual village. Cluster design was based on census enumeration areas with population of approximately 3000, surrounded by a buffer zone to reduce contamination. The target population was rural communities; the urban administrative centre of the district was excluded. Individuals: all women aged 10‐49 who were willing to participate were enrolled. Women who had a terminal family planning procedure were excluded from the final sample, but not from participating in the intervention.
Interventions	Target: community (IEC). Arm 1 (12 clusters, 4557 pregnancies): facilitator initiated women's groups to discuss issues of pregnancy, childbirth and newborn and infant health, as well as peer counselling (infant feeding and care counselling via 5 home visits during and after pregnancy (3rd trimester, week after birth, at 1, 3 and 5 months). Arm 2 (12 clusters, 4722 pregnancies): facilitated women's groups. Arm 3 (12 clusters, 4660 pregnancies): peer counselling via home visits. Arm 4 (12 clusters, 5021 pregnancies): no intervention. All clusters benefited from training of staff in health facilities in essential newborn care.
Outcomes	Trial primary outcomes: maternal, perinatal, neonatal and infant mortality rates, and exclusive breastfeeding. Review outcomes reported: Primary: ANC coverage (at least 4 visits), maternal mortality. Secondary: ANC coverage (at least 1 visit), health facility deliveries, IPT for malaria, tetanus protection, HIV screening, perinatal mortality, neonatal mortality.  Follow‐up: data were gathered monthly between December 2004 and December 2010. All pregnancies, births and deaths were identified, and surviving mothers and infants were followed for up to 1 year.
Notes	Funders: Saving Newborn Lives, UK Department for International Development, Wellcome Trust, Institute of Child Health, and UNICEF Malawi. The primary trial report presents several different analyses, including 1 where Interventions were combined, in order to evaluate the effect of women's groups (arm 1 + 2 combined versus arm 3 + 4 combined) and the effect of peer counselling (arm 1 + 3 combined versus arm 2 + 4 combined) separately. For the analysis in our review's Comparison 1: Lewycka 2013a refers to the women's group intervention only. Lewycka 2013b refers to the peer counselling intervention only. These 2 single‐intervention arms are compared to the arm with no intervention. For the analysis in our review's Comparison 2: Lewycka 2013a refers to the trial arm that received both women's groups and peer counselling. This arm is compared to the arm with no intervention.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation done with computer program Stata.
Allocation concealment (selection bias)	Unclear risk	Not described.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Not blinded.
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	Group assignment was masked for data analysis. Data collection was conducted independently of program implementation and was not fed back to inform the intervention.
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Women with miscarriages were excluded from analysis. Loss to follow‐up about 20%. Miscarriage rates varied across study arms and were more frequent in the combined intervention cluster.
Selective reporting (reporting bias)	Low risk	Relevant outcomes reported.
Analysis bias	Low risk	Analysis appropriate for clusters; no ICC reported; ITT analysis performed.
Other bias	Unclear risk	The authors discuss an interaction between the 2 interventions and baseline imbalances after randomisation across several outcomes.
Overall risk assessment	Unclear risk	We were concerned that the exclusion of women with miscarriages might bias maternal death rates.

Methods	2 by 2 factorial cluster RCT conducted in Malawi between 2005 and 2009. Lewycka 2013b describes the same trial as Lewycka 2013a above, and all of the descriptions and risk of bias are identical to that above. We have had to duplicate the 'Risk of bias' judgements below due to RevMan requirements.
Participants	For the analysis in our review's Comparison 1: Lewycka 2013a refers to the women's group intervention only. Lewycka 2013b refers to the peer counselling intervention only. These 2 single‐intervention arms are compared to the arm with no intervention.
Interventions	See Lewycka 2013a.
Outcomes	See Lewycka 2013a.
Notes	See Lewycka 2013a.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation done with computer program Stata.
Allocation concealment (selection bias)	Unclear risk	Not described.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Not blinded.
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	Group assignment was masked for data analysis. Data collection was conducted independently of program implementation and was not fed back to inform the intervention.
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Women with miscarriages were excluded from analysis. Loss to follow‐up about 20%. Miscarriage rates varied across study arms and were more frequent in the combined intervention cluster.
Selective reporting (reporting bias)	Low risk	Relevant outcomes reported.
Analysis bias	Low risk	Analysis appropriate for clusters; no ICC reported; ITT analysis performed.
Other bias	Unclear risk	The authors discuss an interaction between the 2 interventions and baseline imbalances after randomisation across several outcomes.
Overall risk assessment	Unclear risk	We were concerned that the exclusion of women with miscarriages might bias maternal death rates.

Methods	A parallel arm cluster‐RCT conducted on the island of Unguja, Zanzibar, Tanzania between Mar 2009 and Mar 2010.
Participants	Sample size: 24 clusters (2367 individuals randomised, 2550 analysed). Clusters: government‐run primary healthcare facilities, 4 per district, were selected, based on 2 inclusion criteria: highest number of ANC clients in 2008 and the availability of at least 1 midwife in the facility. All included facilities had mobile phone network coverage. Individuals: women who attended ANC at 1 of the 24 selected healthcare facilities were included on their first ANC visit and followed until 42 days after delivery. Women were eligible for study participation irrespective of their mobile phone and literacy status. Exclusion criteria: PIH, anaemia, multiple pregnancy and malpresentation.
Interventions	Target: health system (policy/practice change). Arm 1 (12 clusters, 1351 women): women received an automated text messaging service for health information and appointment reminders, mobile phone vouchers to enable women to contact health services. The content of messages depended upon gestational age. Women received 2 messages/month < 36 weeks and then 2 per week. Only women with registered phone numbers received text messages; women without received only vouchers with mobile credit. Arm 2 (12 clusters 1286 women): women attending control health facilities received standard ANC, with the goal of at least 4 ANC visits.
Outcomes	Trial primary outcome: skilled delivery attendance. Review outcomes reported: Primary: ANC coverage (at least 4 visits), maternal mortality. Secondary: tetanus protection, perinatal mortality, neonatal mortality. Other: skilled birth attendant (midwife/doctor/nurse) at delivery. Follow‐up: women were offered at least 4 antenatal visits and a postnatal home visit within 48 hours after delivery. Women were interviewed for demographics at trial entry at 6 weeks after delivery. The trial definition of perinatal mortality is non‐standard, stated as stillbirth and death of the infant up to 42 days.
Notes	Funders: Danish International Development Cooperation.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Randomisation described as 'simple randomisation' but sequence generation not described.
Allocation concealment (selection bias)	Unclear risk	Not described.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Clusters and study participants were not masked due to the nature of the intervention requiring overt participation.
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Not described.
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Exclusions/withdrawals < 20% were due to exclusion criteria (development of complications) or loss to follow‐up.
Selective reporting (reporting bias)	Low risk	Relevant outcomes reported.
Analysis bias	Low risk	Analysis appropriate for clusters, ITT analysis was performed. No ICC reported.
Other bias	Unclear risk	No baseline imbalances noted. The trial definition of perinatal mortality is non‐standard, stated as stillbirth and death of the infant up to 42 days.
Overall risk assessment	Low risk	No serious risk of bias concerns.

Methods	Parallel arm cluster‐RCT conducted in a rural setting in Zimbabwe between Jan 1995 and Oct 1997.
Participants	Sample size: 23 clusters (13517 individuals). Clusters: health centres in a rural setting. Gutu district was chosen as the study area because the utilisation of maternity services and reproductive health status of the community had been previously studied.The district had 25 health facilities, comprising a district hospital and 24 rural health centres (RHCs) serving a population of 195 000. Individuals: all mothers booking for ANC since 01/12/94.
Interventions	Target: health system (re‐organisation of ANC). Arm 1: an experimental package of ANC with reduced procedures, clear goals and symphysio‐fundal height measurements. Visits scheduled according to a 5 visit program with reduced routine procedures at these visits. First visit: risk assessment, health education and delivery plan, Hb, rapid plasma reagin, tetanus vaccination and do urinalysis. Visit 2 at 24–28 weeks: exclude multiple pregnancy, check for hypertensive disorders, and do urinalysis. Visit 3 at 32–34 weeks: Exclude anaemia, check fetal growth and review delivery plans, check Hb and do urinalysis. Visit 4 at 36–38 weeks: check fetal growth, exclude abnormal presentation, discuss labour and do urinalysis. Visit 5 at 40–41 weeks: check fetal well being, referral for post‐term induction at 42 weeks and do urinalysis. Arm 2: the control arm followed the standard schedule with a visit every 4 weeks from booking until 28 weeks, every 2 weeks between 28 and 36 weeks and weekly after 36 weeks until delivery. Risk assessment was performed at the booking and subsequent visits, and referral for hospital delivery was made using a list of risk markers recommended by the Zimbabwe Ministry of Health and Child Welfare. Blood pressure, body weight and urinalysis were measured at each visit, while Hb and syphilis test (RPR) were performed at the first visit. Women who tested positive for syphilis had treatment initiated at the booking visit. Oral iron supplementation was provided to all women in both models.
Outcomes	Trial primary outcomes: number of antenatal visits, referrals for antenatal, intrapartum or postpartum problems, place of delivery and low birthweight infant (< 2500 g). Review outcomes reported: Primary: maternal mortality. Secondary: health facility deliveries, preterm birth, perinatal mortality, neonatal mortality.  Follow‐up: women were followed up at ANC visits.
Notes	Funders: Swedish International Development Cooperation Agency (Sida/SAREC) through the Sida–University of Zimbabwe Reproductive Health Research Programme.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Sequence generation not described. Randomisation described as stratified according to the availability of telecommunication for referrals.
Allocation concealment (selection bias)	Unclear risk	Not described.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Not blinded.
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Not blinded.
Recruitment bias (for cluster RCTs)	Low risk	None noted.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	13,517 women randomised. Full records were available for 78% of women with curtailed follow‐up for an additional 20%. Communication with the authors has clarified the numbers used for perinatal deaths, adding back in many women whose records were not retrieved.
Selective reporting (reporting bias)	Low risk	Relevant outcomes reported. The authors were contacted to clarify the numbers used to calculate perinatal death.
Analysis bias	Low risk	Analysis appropriate for clusters; no ICC reported; ITT analysis performed.
Other bias	Low risk	No baseline imbalances.
Overall risk assessment	Low risk	No serious risk of bias concerns after contacting authors with data queries.

Methods	Parallel arm, individually‐randomised RCT conducted at 5 sites in the USA between Jan 94 and Dec 95.
Participants	Sample size: 104 women. Inclusion criteria: all pregnant women planning to attend prenatal care at 1 of 5 participating family planning and women's health clinics. All randomised women were eligible for the state of California's Medicaid program (assistance with healthcare costs). Exclusion criteria: women planning prenatal care at another location or women considering abortion.
Interventions	Target: community (financial incentive). Arm 1: 34 women were randomised to receive at taxi voucher and 35 received a coupon for a baby blanket. Arm 2: the control group received no incentive to attend the first antenatal appointment.
Outcomes	Trial primary outcome: compliance with the first prenatal appointment. Review outcomes reported: Primary: not reported. Secondary: not reported.  Follow‐up: the primary study outcome was compliance with the first prenatal appointment. It appears that additional outcomes were not collected.
Notes	Funders: funded in part by a grant from the American Academy of Family Physicians' Foundation. No usable data.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Sequence from a table of random numbers.
Allocation concealment (selection bias)	Low risk	Sealed envelopes were prepared remotely from clinics.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Women were not told that the study had to do with appointment compliance. It is not clear if staff were aware of group assignment.
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	Outcomes assessors for the primary outcome were blind to group assignment.
Recruitment bias (for cluster RCTs)	Low risk	Not applicable.
Incomplete outcome data (attrition bias)   All outcomes	High risk	22/69 (32%) in the intervention group lost to follow‐up and excluded; 8/35 in controls.
Selective reporting (reporting bias)	Unclear risk	No review outcomes reported.
Analysis bias	Low risk	ITT analysis performed.
Other bias	Low risk	No baseline imbalances noted.
Overall risk assessment	Unclear risk	High loss to follow‐up with no relevant review outcomes.

Methods	Parallel arm cluster‐RCT conducted in Balochistan, Pakistan. A baseline survey took place in Aug‐Sept 1998. Intervention package in place by March 2000; follow‐up survey was conducted between March and April 2002.
Participants	Sample size: 32 clusters (2561 individuals analysed). Clusters: Balochistan is an underdeveloped and poor region of Packistan with the highest maternal mortality rate. Each eligible village cluster had between 5‐15 villages. The project area was divided into 3 zones based on distance from the district hospital. Randomisation took place within each zone. Individuals: women who had had a pregnancy in the last 12 months.
Interventions	Target: health system (health worker education and re‐organisation of services ‐ transport) community (IEC intervention). Arm 1: women were provided information on safe motherhood through pictorial booklets and audiocassettes; TBAs were trained in clean delivery and recognition of obstetric and newborn complications; and emergency transportation systems were set up. The intervention was delivered to women only in 1 group and to both women and husbands in another. Arm 2: the project provided training for health professionals at the district hospital, who provided care for women from both intervention and control clusters. Government healthcare providers also trained staff in primary health facilities throughout the study area.
Outcomes	Trial primary outcomes: perinatal or neonatal death; use of iron‐folic acid.  during pregnancy; prenatal care; tetanus immunisation; and delivery in the district hospital Review outcomes reported: Primary: not reported. Secondary: ANC coverage (at least 1 visit), health facility deliveries, tetanus protection, perinatal mortality, neonatal mortality.  Follow‐up: household survey was conducted 2 years after intervention. The perinatal death outcome seems to have been calculated with live births rather than all births. For our analyses, we have used data from the 2002 follow‐up survey only.
Notes	Funders: NICHD, USAID, UNICEF, World Health Organization, British Council, Government of Japan and The Asia Foundation, and implemented by The Asia Foundation’s Islamabad office. Residual impact survey conducted 2 years after project ended, in 2004; on a sample of 900 women randomly selected from immunisation records at the district health office. We have used data from the original cluster‐randomised trial only.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation by "blindly drawing village cluster names written on folded chits".
Allocation concealment (selection bias)	Unclear risk	Not described.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Not blinded.
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Not described.
Recruitment bias (for cluster RCTs)	Unclear risk	At the 2002 follow‐up survey, intervention clusters were expanded, resulting in 47% increase in size of the control arm. At the 2004 follow‐up survey, refusals or locked households in a selected cluster were replaced by the nearest available household.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	The follow‐up survey interviews were completed for 95.2% of visited households.
Selective reporting (reporting bias)	Unclear risk	Relevant outcomes were reported.
Analysis bias	Unclear risk	Analysis appropriate for clusters; ICC reported; ITT not stated.
Other bias	Unclear risk	Information bias. Most of the results are from the 2002 follow‐up survey, but authors state that some data are from the 2004 survey.
Overall risk assessment	Unclear risk	We were uncertain whether the risk of bias concerns above might have impacted the results.

Methods	A parallel arm cluster‐RCT conducted in India between Oct 2006 and Sept 2009.
Participants	Sample size: 48 clusters (18,197 individuals). Clusters: eligible clusters were communities in urban slums in Mumbai for which a perinatal vital registration was set up as part of the City Initiative for Newborn Health in 2005. The wards were selected purposively for the 2005 Initiative based on accessibility and relative infant mortality rates. Communities with transient populations and areas where resettlement was being negotiated were both excluded. Individuals: women of all ages residing in intervention clusters, whether pregnant or not pregnant, were invited to attend women's groups.
Interventions	Target: community (IEC intervention). Arm 1: women were invited to weekly meetings that emphasized knowledge of local health services, perinatal health care, and negotiating optimal care with family and health providers. Arm 2: no weekly meetings. Surveillance data were collected in both intervention and control areas. 12 interviewers collected these data at 6 weeks postpartum. Unwell mothers or infants in either arm were referred and treatment expedited.
Outcomes	Trial primary outcome: perinatal care, maternal morbidity, and extended perinatal mortality. Review outcomes reported: Primary: maternal mortality. Secondary: professional ANC, health facility delivery, perinatal mortality, neonatal mortality.  Follow‐up: weekly women's group meetings and attendance records. Interviews took place 6 weeks postpartum. Retrospective census at end of trial to pick up any missed births in all 48 clusters.
Notes	Funders: ICICI Foundation for Inclusive Growth – Centre for Child Health and Nutrition, and the Wellcome Trust. DO was funded by a Wellcome Trust Fellowship (081052/Z/06/Z).
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation by "drawing of lots".
Allocation concealment (selection bias)	Unclear risk	Allocation not concealed.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Not blinded.
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	Interviewers may have been aware of the assignment of their particular area, but the authors argue that they "were focused on their task (surveillance) and did not dwell on the comparative nature of the trial." Data analysts were blinded.
Recruitment bias (for cluster RCTs)	Unclear risk	9 clusters were expanded for insufficient births, and 2 clusters reduced for excess births.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition was less than 20% in each arm, and authors have provided a study flow diagram with documented reasons for loss to follow‐up.
Selective reporting (reporting bias)	Low risk	Relevant outcomes were reported.
Analysis bias	Low risk	Analysis appropriate for clusters; ITT analysis performed; ICC not reported.
Other bias	Unclear risk	Other initiatives during the trial period include outreach services by health volunteers, birth registration and pulse polio campaigns and infectious disease surveillance. Conditions in slums improved over the trial period.
Overall risk assessment	Unclear risk	We were uncertain whether the risk of bias concerns above might have impacted the results.

Methods	Cluster‐RCT in Bulgan, Mongolia.
Participants	Sample size: 501 women randomised. Clusters: the unit of randomisation was the Soum and bag, small geographic areas in Mongolia. Each Soum has a healthcare facility where women must register their newborn. 18 geographic areas were randomised, after selection for administrative convenience and to avoid contamination. Individuals: pregnant women living in Bulgan, Mongolia.
Interventions	Target: community. Arm 1: distribution of maternal and child health handbooks during pregnancy. The MCH handbook logged maternal health and personal information, pregnancy, delivery and postpartum health and weight, dental health, parenting classes, child developmental milestones from 0‐6 years, immunisation records and height and weight charts for children. Arm 2: women received standard care.
Outcomes	Trial primary outcome: number of antenatal visits; proportion of women attending 6 or more antenatal visits. (The national standard for ANC in Mongolia is 6 visits.) Review outcomes reported: Primary: ANC coverage of at least 4 visits, maternal mortality Secondary: maternal outcomes: morbidity during pregnancy, mode of delivery, breastfeeding initiation, maternal depression and health (EPDS and GHQ). Infant outcomes: birthweight, Apgar score, NICU admission, neonatal mortality at discharge. Maternal healthy behaviours. Follow‐up: data collection at 1 month postpartum.
Notes	Funders: this study was funded by the National Center for Global Health and Medicine, Tokyo, Japan. Significant group differences noted for distances travelled to nearest health centre (greater in the intervention group) and for wealth index (the control group was poorer). The authors report that travel time did not function as an effect modifier; however, women from a higher socioeconomic background attended more ANC visits. Trial authors provided unpublished outcome data upon request. The trial statistician (HN) calculated ORs and 95% confidence intervals using the generalised estimating equations (GEE) method to adjust for cluster design and baseline differences, including wealth.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Sequence according to the shuffling of sealed envelopes.
Allocation concealment (selection bias)	Low risk	Allocation was concealed in sealed envelopes at time of randomisation. All areas were randomised at the same time.
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	Masking was not possible for this intervention.
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Masking was not possible for this intervention.
Recruitment bias (for cluster RCTs)	Low risk	No problems with recruitment are reported.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	3 randomised areas were excluded; 1 was the subject of a pilot study, and 2 areas were included in another health study. 9 clusters each received the intervention or the control. Missing outcome data for individual women is reported and minimal.
Selective reporting (reporting bias)	Low risk	Prespecified outcomes have been reported. Addtional analyses were obtained from the authors upon request. The trial data file has been published online with the trial report.
Analysis bias	Low risk	Analyses were undertaken with methods appropriate for cluster trials; the authors used GEE methods to adjust for the effects of cluster design and baseline variables. A sample size calculation was undertaken and met.
Other bias	Unclear risk	The authors reported baseline imbalances between clusters for travel time to health centre and wealth. The authors reported that recall bias may exist due to data collection at 1 month after birth.
Overall risk assessment	Low risk	Overall the trial was well planned and conducted.