Impact of packaged interventions on neonatal health: a review of the evidence

Abstract

A disproportionate burden of infant and under-five childhood mortality occurs during the neonatal period, usually within a few days of birth and against a backdrop of socio-economic deprivation in developing countries. To guide programmes aimed at averting these 4 million annual deaths, recent reviews have evaluated the efficacy and cost-effectiveness of individual interventions during the antenatal, intrapartum and postnatal periods in reducing neonatal mortality, and packages of interventions have been proposed for wide-scale implementation. However, no systematic review of the empirical data on packages of interventions, including consideration of community-based intervention packages, has yet been performed. To address this gap, we reviewed peer-reviewed journals and grey literature to evaluate the content, impact, efficacy (implementation under ideal circumstances), effectiveness (implementation within health systems), type of provider, and cost of packages of interventions reporting neonatal health outcomes. Studies employing more than one biologically plausible neonatal health intervention (i.e. package) and reporting neonatal morbidity or mortality outcomes were included. Studies were ordered by study design and mortality stratum, and their component interventions classified by time period of delivery and service delivery mode.

We found 41 studies that implemented packages of interventions and reported neonatal health outcomes, including 19 randomized controlled trials. True effectiveness trials conducted at scale in health systems were completely lacking. No study targeted women prior to conception, antenatal interventions were largely micronutrient supplementation studies, and intrapartum interventions were limited principally to clean delivery. Few studies approximated complete packages recommended in The Lancet's Neonatal Survival Series. Interventions appeared largely bundled out of convenience or funding requirements, rather than based on anticipated synergistic effects, like service delivery mode or cost-effectiveness. Only two studies reported cost-effectiveness data. The evidence base for the impact of neonatal health intervention packages is a weak foundation for guiding effective implementation of public health programmes addressing neonatal health. Significant investment in effectiveness trials carefully tailored to local health needs and conducted at scale in developing countries is required.

Introduction

Child survival initiatives have made significant strides in reducing under-five child mortality in recent decades. As the burden of post-neonatal deaths declines, neonatal deaths—deaths within the first 28 days of life—comprise an increasing share, currently estimated at 38% of all under-five deaths and totalling 4 million annually (Darmstadt et al. 2005; Lawn et al. 2005). Ninety-nine per cent of neonatal deaths occur in developing countries (Lawn et al. 2005), usually soon after birth, at home (Black et al. 2003) against a backdrop of poverty, suboptimal care-seeking and weak health systems (Lawn et al. 2004; Bhutta et al. 2005; Lawn et al. 2005).

Fostering social development, reducing inequity and promoting economic growth among the poor in developing countries are important objectives addressed by the Millennium Development Goals (Sachs and McArthur 2005). There is an immediate need, however, for wide-scale implementation of evidence-based, cost-effective health programmes and interventions to improve newborn health outcomes. Meeting Millennium Development Goal (MDG) 4, which calls for a two-thirds reduction in under-five mortality by 2015, will require a significant reduction in neonatal deaths, particularly early neonatal deaths. A recent analysis of the evidence base for efficacy (i.e. impact under ideal conditions) and effectiveness (i.e. impact within a health system) of interventions, and their cost-effectiveness, suggested that feasible, cost-effective interventions exist that could prevent roughly two-thirds of all neonatal deaths (Darmstadt et al. 2005). Many of these interventions are relatively simple and highly cost-effective (Adam et al. 2005).

While evidence suggests that most neonatal deaths can be prevented even in settings with high neonatal mortality and weak health systems (Darmstadt et al. 2005), programmes face serious logistical challenges in delivering interventions at scale, particularly where access to health services is poor (Knippenberg et al. 2005). Health systems in much of the world are hobbled by bureaucracy, corruption and lack of financial and human resources (Freedman et al. 2005). Cost-effective interventions rarely reach people who need them most, as programmes delivering these interventions often suffer from intermittent coverage, poor quality, inequitable access and lack of long-term objectives (Victora et al. 2004b).

A recent series in The Lancet addressed deficiencies in health systems, from inequities to logistical and resource constraints, and drew attention to the dearth of health systems effectiveness studies (Haines and Victora 2004). The weak evidence base yields little consensus about how best to strengthen health systems. Travis et al. (2004) note an urgent need to translate the limited existing knowledge about intervention effectiveness into practice. However, strengthening health systems is costly and requires long-term investment. A recent global meeting convened by the World Health Organization (WHO) identified this lack of health systems research as a critical barrier to achieving the MDGs (Task Force on Health Systems Research 2004), and advocated for increased resources to generate the data needed to design equitable, effective, efficient and sustainable health systems. Health systems research must also consider how the quality and accessibility of health systems (Box 1) vary widely by region and even within countries, as do factors that promote programme sustainability and scalability (Bhutta 2005).

The growing body of evidence from clinical and programme research suggests that effective neonatal health interventions could be bundled together to enhance cost-effectiveness and to suit existing health systems (Victora et al. 2004b; Darmstadt et al. 2005). The 2003 series in The Lancet by the Bellagio Child Survival Study Group hypothesized that under conditions of optimal delivery—universal coverage and acceptance—evidence-based child and neonatal health interventions could prevent 63% of under-five child deaths and up to 55% of neonatal deaths (Jones et al. 2003). While this study highlighted the growing fraction of neonatal deaths, its analysis covered the efficacy of a relatively limited range of single interventions that impact neonatal health.

Subsequently, The Lancet's Neonatal Survival Series built upon the concept of organizing and packaging neonatal health interventions for effective delivery by three service delivery modes: family-community, outreach, or facility-based clinical care (World Bank 2004; Bhutta et al. 2005; Darmstadt et al. 2005). Presenting analyses from a health systems perspective, The Lancet's series recommended that interventions be packaged according to target population, time period of implementation and service delivery mode (Bhutta et al. 2005; Knippenberg et al. 2005). As a newborn infant's health is intrinsically linked with its mother's, incorporating newborn care into existing Safe Motherhood as well as Child Survival programmes has also been suggested as a cost-effective way to bundle interventions to simultaneously improve neonatal, maternal and child health outcomes (Bhutta et al. 2005; Knippenberg et al. 2005). However, thus far there has been no published systematic review of the empirical evidence for the impact of packages of interventions on neonatal health outcomes, nor has there been an assessment of the efficacy or effectiveness of such packages.

This review evaluates available data on the implementation of packages of interventions (i.e. two or more biologically plausible neonatal health interventions) intended to reduce neonatal mortality and morbidity, and identifies research gaps in the implementation of packages of neonatal health interventions in health systems. The scope of this review is limited to interventions addressing proximal (direct) determinants of neonatal outcomes rather than distal determinants such as economic development and social or gender inequities, because of the lack of empirical data quantifying how distal determinants impact neonatal outcomes.

Methods

Selection of interventions

We evaluated the available evidence in the global literature for the impact of a wide range of interventions during the antenatal, intrapartum and postnatal periods on perinatal and neonatal health status outcomes. We used results of previously published systematic reviews of the evidence for single interventions that had estimated mortality impacts using peer-reviewed studies and expert opinion (Bhutta et al. 2005; Darmstadt et al. 2005). Selection of individual interventions as package components was based on biological plausibility and potential feasibility of inclusion in maternal, neonatal and/or child health care programmes. Those chosen for review are shown in Box 2.

Only studies from developing countries testing packages of these health interventions (i.e. more than one component intervention) were considered eligible for inclusion. We considered iron/folic acid supplementation during the antenatal period a single intervention for the purposes of this review, as the supplements were usually administered together (only two studies administered iron without folic acid) (Bouvier et al. 1997; Ramakrishnan et al. 2003). Although efforts were made to identify health systems studies measuring the effectiveness of intervention packages, no such studies were found. Two of the studies we reviewed approximated effectiveness trials of a package of interventions; one (Srinivasan et al. 1995) offered a package of interventions but study arms differed only in the degree of provider training, and the other (Yan 1989) lacked a control or comparison group and provided little detail about implementation, making both fall short of the criteria for a true effectiveness trial of a package of interventions.

Literature search and review strategy

The search for evidence explored all available electronic health and social science reference libraries, including PubMed/MEDLINE, POPLINE, LILACS, PAHO, African Index Medicus and EMRO. Manual reviews were conducted of Safe Motherhood and Child Survival monographs, including programme evaluations and technical reports. Evidence from the Cochrane Library and the WHO Reproductive Health Library was reviewed where applicable. Further information and unpublished material was solicited directly from agencies, institutions and leading public health researchers working in developing countries. Papers in languages other than English, provided an abstract was available in English, were reviewed. While we prioritized the identification of randomized controlled trials (RCTs), the scarcity of developing country RCTs comparing packages of interventions and the theoretical limitations of RCTs, including lack of generalizability and inappropriateness in health systems settings (Horwitz 1987; Rabeneck et al. 1992; Concato and Horwitz 2004; Victora et al. 2004a), led us to include other study designs as well, namely prospective cohort, quasi-experimental and retrospective designs.

Priority was given to evidence of impact on perinatal or neonatal mortality, but when data on these primary health status outcomes was nonexistent, we assessed the evidence for impact on determinants of mortality and morbidity, including outcomes such as reductions in prematurity, birth asphyxia incidence, low birth weight (LBW) rates, breastfeeding and infectious disease morbidity (i.e. ‘secondary outcomes’). Studies of interventions such as population-based nutrition education or hand-washing interventions that included but did not quantify impact for neonates (Luby et al. 2004; Sobel et al. 2004; Jones et al. 2005; Luby et al. 2005) were considered beyond the scope of this paper.

Qualitative description and ranking of studies

Several different systems of standard ranking criteria have been devised and applied to international clinical trial data, such as those designed by the National Institute for Clinical Excellence (NICE), the Scottish Intercollegiate Guidelines Network and the Cochrane Collaboration (SIGN 2001; Jackson and Waters 2004; NICE 2004). Their application to non-randomized studies remains of limited use, however, as the criteria assume a high degree of design and analytical robustness rarely found in developing country field trials. We attempted to use these scoring schemas in our review, but the criteria failed to apply in all but a handful of cases due to the small number of studies identified for review and the variability in design and analytical rigour among them.

In lieu of adopting an international standard scoring schema, studies were evaluated for size, design quality and setting (including service delivery mode and WHO neonatal mortality stratum) (Zupan and Aahman 2005). Two principal reviewers independently evaluated all studies, and a common reporting matrix was employed in summarizing findings (Tables 1–4). Studies were classified into three descending tiers of study design quality: (1) RCTs (Tables 1, 2); (2) prospective cohort trials with control/comparison groups (Table 3); and (3) retrospective or quasi-experimental non-controlled trials (Table 4). We found no data on whether appropriate statistical adjustment was made for older cluster-randomized studies (e.g. Kielmann et al. 1978a,1978b), which may weaken the accuracy of these mortality estimates in comparison with newer cluster-randomized RCTs that clearly specified statistical adjustment.

Component interventions in each package were identified and categorized according to time period of delivery (e.g. periconceptual, antenatal, intrapartum or postnatal). The service delivery mode by which each component intervention was implemented was classified as family-community, outreach, or facility-based clinical care. Family-community interventions were defined as those engaging family members or volunteer community members as providers, including adoption of improved care practices and appropriate care-seeking for illness (e.g. behaviour change communications and accompanying community mobilization strategies, education to improve care-seeking, and community-based case management of illnesses by community health workers). Outreach interventions were defined as interventions delivered periodically by trained providers. Facility-based clinical care was delivered within the formal health care system by providers skilled in managing acute clinical problems. In any cases of inter-observer disagreement, which occurred rarely, most commonly in interventions including training of traditional birth attendants (TBAs) via outreach to encourage family-community care, the two reviewers examined the original paper together to mutually agree on a classification. Studies were sorted by design and ordered hierarchically in a matrix to maximize clustering of interventions with like service delivery modes. The empirical body of evidence for the impact of packaged interventions was then compared with recent recommendations and impact modelling in The Lancet's Neonatal Survival Series recommendations and impact modelling (Bhutta et al. 2005; Darmstadt et al. 2005).

Results

We found a paucity of data from developing countries and poor rates of inclusion of a number of key evidence-based interventions. Forty-one studies from developing countries that included a package of interventions for neonates were identified for in-depth review, including 19 RCTs, 16 prospective cohort studies, and six retrospective/quasi-experimental studies. Sample size ranged from 126 women (Qureshi et al. 1973; Bouvier et al. 1997) to more than 100 000 births (Meegan et al. 2001); however, most packages were implemented with relatively small sample sizes, and quality and publication source varied widely.

Study design

RCTs were subdivided into two types: (1) those involving interventions delivered across varying time periods and service delivery modes (n = 7) (Table 1), and (2) those involving only antenatal interventions (n = 12), most of which were limited exclusively to outreach-based micronutrient supplementation (n = 10) (Table 2). By comparison, prospective studies were heavily oriented towards postnatal interventions (Table 3). All 16 prospective studies had some postnatal component, and seven of these incorporated only postnatal interventions. No clear patterns emerged from the retrospective/quasi-experimental studies, due in part to their small number (n = 6) (Table 4).

Service delivery mode

Studies tended to cluster interventions by service delivery mode; 27 of the 41 studies employed a single service delivery mode (commonly outreach). In cases where more than one service delivery mode was used, most commonly there was a mix of all three modes, but this was observed almost exclusively during the postnatal period.

Across all time periods, interventions in packages were delivered predominantly via the family-community service delivery mode (95/195), especially clean home delivery (Table 5). Clinical (43/195) and outreach (57/195) service delivery modes were also used less commonly, although outreach was the most common among antenatal interventions (Table 5). This was heavily influenced, however, by the 12 antenatal micronutrient supplementation RCTs (Table 2), which accounted for 27 of 39 antenatal outreach interventions. During the antenatal and intrapartum periods, clinical interventions—interventions normally provided at a clinic or hospital facility by a skilled health care provider—were included in only three studies (Yan 1989; Srinivasan et al. 1995; Smith et al. 2000). Many evidence-based outreach (e.g. periconceptual folic acid supplementation, syphilis screening and treatment, intermittent presumptive treatment for malaria, detection and treatment of asymptomatic bacteriuria) and clinical interventions [e.g. antibiotics for preterm premature rupture of membranes (PPROM), corticosteroids for preterm labour] were not included in any packages. Screening for pregnancy-induced hypertension was conducted in only one study (Table 1) (Srinivasan et al. 1995).

Studies primarily using the family-community service delivery mode were more likely than others to provide a continuum of care from pregnancy to childbirth and the postnatal period. An illustrative example is Dutt and Srinivasa's engagement of Anganwadi workers [a type of community health worker (CHW) under the Indian government's Integrated Child Development Services scheme] as community-based providers of antenatal supplement and tetanus toxoid immunization, facilitators of referral and emergency transport for maternal and newborn complications, and educators during postnatal home visits about family planning and newborn care (Dutt and Srinivasa 1997). This approach achieved 100% coverage of antenatal care and TT immunization for all mothers and eliminated neonatal deaths due to tetanus and measles, resulting in a 15% decline in the neonatal mortality rate.

Service delivery mode was not closely linked to study design. Except for antenatal-only RCTs, which only used outreach for delivery, there were examples in all three study design categories of studies (12/41) that mixed family-community, outreach and clinic-based service delivery modes, yet included a limited number of evidence-based interventions.

Time period of interventions

No interventions in any package targeted women of reproductive age periconceptually, the critical period for folic acid supplementation (Bhutta et al. 2005; Darmstadt et al. 2005). Rather, folic acid supplementation was frequently included in antenatal packages (n = 10), in spite of a lack of evidence that it improves neonatal health outcomes. Antenatal interventions were almost exclusively restricted to micronutrient supplementation trials (n = 12), tetanus toxoid immunization (n = 7) and/or birth preparedness (n = 5). Birth preparedness and tetanus toxoid immunization were implemented together in only two studies, using multiple service delivery modes (Tables 1, 2) (Greenwood et al. 1990; Srinivasan et al. 1995).

Intrapartum interventions overwhelmingly involved promotion of clean delivery. Of the 16 studies that incorporated any intrapartum intervention, 14 included clean delivery practices. Among RCTs, most studies limited intrapartum interventions to clean delivery, except for one study that utilized skilled attendants at delivery (Srinivasan et al. 1995). Similarly, among studies of other designs that included intrapartum components, clean delivery was the only intrapartum intervention in six of the eight prospective studies and three of the four retrospective studies. Six studies packaged clean delivery training for birth attendants with tetanus toxoid immunization (Kielmann et al. 1978a; Rahman 1982; Berggren et al. 1983; Greenwood et al. 1990; Kapoor et al. 1991; Srinivasan et al. 1995), and three studies provided labour surveillance in addition to clean delivery (Berggren et al. 1983; Janowitz et al. 1988; Smith et al. 2000), but none of these used a partograph.

Overall, postnatal interventions (n = 123) were more common components of packages than antenatal (n = 52) or intrapartum (n = 20) interventions (Table 5). Common postnatal interventions included TBA/CHW training (n = 21; notably, their activities often spanned the antenatal-intrapartum-postnatal continuum), postnatal visitation, maternal health education, breastfeeding promotion, newborn resuscitation, care of LBW infants, and hypothermia prevention and management. Very few studies included more technically involved but efficacious postnatal interventions such as newborn resuscitation or injectable antibiotics for neonatal infections.

Packaging patterns versus existing recommendations

Commonly bundled interventions included TBA/CHW training and postpartum visitation, implemented together in 12 studies; eight of these also included maternal health education (Tables 1, 3). Training CHWs was the most common strategy to provide neonatal resuscitation (n = 5), diagnose and refer or treat pneumonia cases (n = 6) and provide breastfeeding advice (n = 5). During the antenatal period, interventions appeared bundled out of convenience of administration, and no clear patterns defined the overall packaging of supplements or immunizations.

Only one study (Bang et al. 2005) implemented the complete family-community package of evidence-based essential newborn care interventions described in The Lancet's Neonatal Survival Series (Darmstadt et al. 2005). Bang et al. reported a 70% decline in the neonatal mortality rate (NMR), a 56% decline in the perinatal mortality rate (PMR), and a 49% decline in the stillbirth rate. A community-based study by Manandhar et al. (2004) did not implement a designated package of interventions but instead educated women's groups about proven interventions using pictorial cards, allowing each group to select the package of interventions they wanted to implement. Although only a small fraction of the population was involved in women's groups, the intervention led to a 30% decline in the NMR, suggesting successful community mobilization and communication of messages outside of the group meetings. These studies incorporating the full range of family-community interventions were among those with the greatest impact on primary outcomes (Table 6). Five others approached the family-community package ideal, implementing packages missing just one or two of the recommended interventions (Kielmann et al. 1978a; Rahman 1982; Daga et al. 1996; Arif and Arif 1999; Kumar et al. 2004).

Only one study approximated The Lancet-recommended antenatal package of outreach-based interventions (Srinivasan et al. 1995), though it lacked breastfeeding outreach. Most studies lacked intrapartum interventions other than clean delivery. However, several studies provided packages of interventions closely corresponding to recommendations for care of LBW infants (Rahman 1982; Pratinidhi et al. 1986; Datta et al. 1987; Bhakoo et al. 1989; Daga et al. 1996; Arif and Arif 1999; Bang et al. 1999; Bose et al. 1999; Smith et al. 2000; Kumar et al. 2004); case management of pneumonia (Kielmann et al. 1978a; Rahman 1982; Bartlett et al. 1991; Arif and Arif 1999; Bang et al. 1999); and emergency neonatal care (Pratinidhi et al. 1986; Borulkar et al. 1998; Bang et al. 1999; Bose et al. 1999; Smith et al. 2000; Manandhar et al. 2004). Notably, emergency care in these studies rarely conformed to the recommended model of prompt referral of outborn infants with complications to facilities; in many settings, this model was infeasible because of economic or geographic barriers to access care, lack of quality care in local facilities, and/or cultural traditions to protect vulnerable newborns by keeping them at home for some time after birth. No study offered any of the situational or additional interventions recommended by The Lancet's series, including periconceptual folic acid supplementation, intermittent presumptive treatment for malaria, detection and treatment of asymptomatic bacteriuria, and corticosteroids for preterm labour. Several intervention packages recommended by The Lancet's series were also missing: none of the studies included the emergency obstetric care package, the skilled maternal and immediate newborn care package, or the antenatal care III package, which includes tetanus toxoid as well as syphilis screening and treatment in addition to taking a pregnant woman's medical history and providing a complete physical examination to screen for complications.

Community mobilization and participatory methods

More than a quarter (11/41) of the reviewed studies included community mobilization components (Web Table 1). Among studies without community mobilization components, there were examples of significant formative research with community members (Meegan et al. 2001) and community members as intervention implementers (Dutt and Srinivasa 1997). Specific activities constituting ‘community mobilization’ varied considerably, ranging from community-wide sensitization meetings (Pratinidhi et al. 1986; Greenwood et al. 1990; Kumar et al. 2004; Bang et al. 2005) and health education sessions (Datta et al. 1987; Bang et al. 1994; Bang et al. 1999; Kumar et al. 2004) to weekly community women's groups run by trained local facilitators (O’Rourke et al. 1998; Manandhar et al. 2004). Studies that had community mobilization components generally fell into two categories: (1) older studies offering health education to improve uptake of externally generated interventions (Kielmann et al. 1978a; Pratinidhi et al. 1986; Datta et al. 1987; Greenwood et al. 1990), and (2) newer studies adopting participatory research approaches (O’Rourke et al. 1998; Bang et al. 1999; Smith et al. 2000; Manandhar et al. 2004). Most studies targeted mothers and fathers, and occasionally neighbours and other community members.

Community mobilization activities were commonly delivered by community members or CHWs (7/11 studies). A family-community service delivery mode was prominent in 10 of these 11 studies, but three also used outreach or clinical-based interventions (Datta et al. 1987; Greenwood et al. 1990; Manandhar et al. 2004). All 11 studies with community mobilization components reported primary outcomes, compared with just 12 of 30 studies that did not employ community mobilization strategies. Several studies with community mobilization activities reported the highest declines in PMR (56–63%) (O’Rourke et al. 1998; Bang et al. 2005) and NMR (70–84%) (Meegan et al. 2001; Bang et al. 2005) of all the studies we reviewed. One study lacking a community mobilization component asserted that intervention effectiveness might have been improved by community mobilization activities, particularly maternal education (Rahman 1982).

Care-seeking components and behaviours

Care-seeking components were defined as any behaviour or service that enabled or enhanced the ability of mothers or their neonates to receive skilled medical care during delivery or for obstetric or neonatal complications; 16 of the 41 studies reviewed contained a care-seeking component (Web Table 2). Studies incorporating community mobilization activities were more than twice as likely as other studies to encourage care-seeking activities (6/11 studies with community mobilization activities, versus only 9/30 studies without community mobilization activities). Most studies that included care-seeking components were prospective cohort studies (8/16); antenatal RCTs, predominantly based on outreach activities, uniformly did not include promotion of care-seeking. In two studies, mothers were trained to provide care for their neonates and seek help from medical professionals in case of problems while within a hospital environment (Daga et al. 1992; Arif and Arif 1999).

Referral procedures were the most commonly promoted form of care-seeking (13 of the 16 studies with care-seeking components included referrals) (Kielmann et al. 1978a; Pratinidhi et al. 1986; Datta et al. 1987; Greenwood et al. 1990; Bartlett et al. 1991; Srinivasan et al. 1995; Bouvier et al. 1997; Dutt and Srinivasa 1997; Bang et al. 1999; Bose et al. 1999; Smith et al. 2000; Manandhar et al. 2004; Jokhio et al. 2005), but were often limited to instructing TBAs, midwives or CHWs about the indications for referral (Berggren et al. 1983; Datta et al. 1987; Srinivasan et al. 1995; Bouvier et al. 1997; Bang et al. 1999; Bose et al. 1999; Smith et al. 2000; Jokhio et al. 2005). Enhanced follow-through of referral through supportive intervention components was uncommon (n = 5 studies), although approaches included creating demand among parents through education (Bang et al. 1990; Bartlett et al. 1991), providing accompanied referral (Greenwood et al. 1990; Bartlett et al. 1991), developing a cadre of personnel to link TBAs to the health system (Jokhio et al. 2005), ensuring that referred cases were within the capacity of referral centres to handle (Bartlett et al. 1991), or strengthening the capacity of health services to handle referred cases (Manandhar et al. 2004).

Problems with promotion of care-seeking were noted by several studies. Five studies found parents uncooperative with or unable to pursue recommended referrals (Pratinidhi et al. 1986; Datta et al. 1987; Srinivasan et al. 1995; Bang et al. 1999; Bose et al. 1999). Pratinidhi et al. (1986) noted parents had problems obtaining transportation, while Smith et al. (2000) found that TBA care was associated with prolonged labour, suggesting a possible reluctance to refer by TBAs, although this was not confirmed. A meta-analysis of the effectiveness of TBA training on pregnancy outcomes has shown more promising results (Buekens 2003; Sibley and Sipe 2004), but identifying appropriate roles for TBAs during the antenatal, intrapartum and postnatal periods remains a priority research area (Bhutta et al. 2005). Datta et al. (1987) found that CHWs/TBAs were uncomfortable or inaccurate in their assessments of who needed referral, and Yan (1989) noted possible inability of referral centres to provide quality care, evidenced by high PMRs.

A few studies that did not include care-seeking components remarked that their interventions might have been improved through including education about danger-sign recognition (Kapoor et al. 1991; Srinivasan et al. 1995), encouraging care-seeking by parents (Rahman 1982), and/or altering typical referral arrangements such as substituting care by nurses (Borulkar et al. 1998) or mothers (Bhutta et al. 2005) for typical care by paediatricians at referral centres due to scarcity of trained staff, although such approaches to increase care-seeking for childhood illnesses have met with variable success (Kidane and Morrow 2000; Mohan et al. 2004).

Health systems and effectiveness trials

Although several studies involved the local health system, none was considered a true effectiveness trial implemented at scale in a health systems setting (Web Table 3). Less than one-quarter of all studies reviewed engaged any part of the health system and even fewer considered sustainability issues; no study was fully integrated into the health system. Ten studies utilized one or more components of the existing health system to deliver some of the interventions in the package, (Arif and Arif 1999; Bang et al. 1999; Bartlett et al. 1991; Borulkar et al. 1998; Deorari et al. 2000; Rahman 1982; Smith et al. 2000; Srinivasan et al. 1995; van der Mei et al. 1994; Yan 1989), including involving local staff and/or facilities or collaborating with the health system and international non-governmental organizations.

Ten additional studies involved health systems to a lesser degree. Two prospective studies (Bhakoo et al. 1989; Dutt and Srinivasa 1997) used hospital records for baseline and retrospective analyses. Six studies used the existing health system only to identify study participants (Mardones-Santander et al. 1988; Fawzi et al. 1998; Caulfield et al. 1999; Ndyomugyenyi and Magnussen 2000; Merialdi 2001; Kumwenda et al. 2002), and two studies (Berggren et al. 1983; Bang et al. 1993) attempted to involve referrals to the existing health system but were either unsuccessful or had unintended consequences. Bang et al. (1993) found that few patients referred for severe pneumonia sought follow-up care, while Berggren et al. (1983) observed that having midwives refer patients for obstetric complications and cord care overburdened health facilities.

Public versus private providers in intervention delivery

Of the 41 studies, private providers alone were most commonly used to deliver interventions (16/41) (Web Table 4), but public providers (11/41), as well as a combination of public and private providers (14/41), were also common. Providers at the community level, most commonly TBAs and/or CHWs, were instrumental in delivering interventions in 22 studies; TBAs/CHWs were involved in 17/40 studies and were the principal intervention implementers in half of these studies. These TBAs/CHWs were predominantly private providers (and most commonly, independent community members rather than trained NGO staff); TBAs/CHWs were clearly affiliated with government programmes in only three studies (Greenwood et al. 1990; Srinivasan et al. 1995; Dutt and Srinivasa 1997), and in one of these cases, most TBA trainees had practiced privately before the programme began (Greenwood et al. 1990). TBAs/CHWs were almost universally involved in RCTs that included postnatal interventions (5/6 studies) (Kielmann et al. 1978b; Rahman 1982; Srinivasan et al. 1995; Kumar et al. 2004; Manandhar et al. 2004; Jokhio et al. 2005). By contrast, TBAs were rarely used in antenatal RCTs to deliver interventions, but one study trained private village health workers (CHWs) to distribute supplements (Dijkhuizen and Wieringa 2001). Antenatal micronutrient supplementation RCTs in rural community settings, often characterized by poorly functioning government health infrastructure, were conducted using predominantly private providers, i.e. study staff (Qureshi et al. 1973; Ross et al. 1985; Bouvier et al. 1997; Dijkhuizen and Wieringa 2001; Ramakrishnan et al. 2003); in areas with a more developed health infrastructure, studies tended to enlist the help of public providers (hospital or antenatal clinic staff) to recruit participants or to distribute supplements (Mardones-Santander et al. 1988; Caulfield et al. 1999; Ndyomugyenyi and Magnussen 2000; Merialdi 2001; Kumwenda et al. 2002). Studies with a prospective design were most likely to use a combination of public and private providers, accounting for 8 of 14 such studies.

Outcome reporting

Twenty-eight studies reported primary outcomes and 23 reported secondary outcomes; 7 and 14 RCTs reported primary and secondary outcomes, respectively (Table 1). In some cases, study sample size limited researchers’ ability to report statistically significant findings. While many studies reported declines in perinatal and neonatal mortality, the association between particular interventions or sub-packages of interventions and primary outcomes was impossible to ascertain from available data. In almost all studies where primary outcomes were reported, these figures reflected the aggregate impact of the packaged interventions; measured or estimated effects attributable to individual component interventions or sub-packages were not reported. Only one study, by Rahman (1982), presented disaggregated results for TBA training and tetanus toxoid immunization, but its phased study design precluded computation of the impact of the packaged interventions.

Of the 23 studies reporting primary outcomes, 20 found moderate to sizeable declines in mortality rates (15–84% reduction) and three reported no change in mortality (Deorari et al. 2000; Smith et al. 2000). Eight studies reported declines in neonatal mortality that exceeded 50% (Rahman 1982; Berggren et al. 1983; Bhakoo et al. 1989; Bartlett et al. 1991; Kapoor et al. 1991; Bang et al. 1999; Meegan et al. 2001; Kumar et al. 2004".). In almost all of these studies, dramatic decreases in tetanus cases appeared to contribute substantially to the reduction in mortality, as three-quarters included tetanus toxoid immunization and/or clean delivery, and all but one were conducted in countries with a mortality stratum of 3 (NMR = 30–45 per 1000 live births) or 4 (NMR >45 per 1000), characterized by incomplete tetanus toxoid immunization and a significant proportion of deaths from tetanus neonatorum. When multiple service delivery modes were used, particularly if family-community interventions were delivered, the mortality impact appeared marginally higher than with single service delivery modes (Table 6).

Studies taking a more holistic approach to packaging interventions (>5 interventions, spanning multiple time periods; n = 11) were more likely to report mortality rates (9 of 11 studies) than more targeted studies (≤5 interventions, generally in a single time period; n = 30) (12 of 30). However, the range of mortality reduction was comparable; PMR reductions of holistic studies ranged from 29–63% (mean: 45%, median: 41%) compared with 25–58% (mean: 42%, median: 40%) among targeted studies, and NMR reductions ranged from 15–72% (mean: 38%, median: 30%) among holistic studies compared with 0–85% (mean: 54%, median: 48%) among targeted studies. Outcome ranges for holistic studies were also similar in magnitude to studies that employed community mobilization activities, which reported declines in PMR of 0–63% (mean: 35%, median: 41%), and declines in NMR of 27–84% (mean: 31%, median: 42%). This is in large part because studies including community mobilization activities were more likely to be holistic (5/9) than studies without such activities (6/32). Unfortunately, the number of studies was too small, and the array of interventions each offered too varied, to perform appropriate statistical comparisons of holistic and non-holistic studies.

Cost-effectiveness

While some recent studies have modelled cost-effectiveness data for neonatal health interventions (Adam et al. 2005; Darmstadt et al. 2005), empirical cost-effectiveness data are scarce. Cost-effectiveness data were reported by only two studies implementing neonatal intervention packages (Bang et al. 1999; Manandhar et al. 2004). In rural India, a complete package of home-based antenatal, intrapartum and postnatal care (Bang et al. 1999), including home visits to provide health education, manage birth asphyxia, care for LBW/preterm babies, prevent and treat hypothermia, and diagnose and treat sepsis, cost US$5.30 per newborn, including $1.30 in non-recurring costs and $3.80 in recurring costs, much lower than the cost of hospital care for newborns in India (Modi and Kirubakaran 1995; Shanmugasandaram et al. 1998) and equally effective, at a cost of $95 per life saved. An RCT investigating the impact of facilitating women's groups to implement essential newborn care practices on neonatal health outcomes in rural Nepal (Manandhar et al. 2004) reported that the cost per newborn life saved was US$3442 ($4397 including costs of health-service strengthening activities). The intervention cost $111 per year of life saved ($142 including costs of health-service strengthening activities).

Discussion

The empirical evidence evaluating packaged neonatal interventions is most substantial for antenatal outreach (primarily micronutrient supplementation) and postnatal family-community interventions, and weakest for intrapartum family-community and intrapartum and postnatal clinical interventions. Care across the continuum of time periods and across multiple service modes was conspicuously absent in most studies, as were linkages between community and facility-based care and effectiveness trials within health systems. Integration of care among private and public providers was also lacking. The small number of package studies and their failure to incorporate key evidence-based interventions and packages supports assertions from prior reviews that data on packaged interventions remains a significant research gap (Bhutta et al. 2005; Darmstadt et al. 2005). Thus, key questions remain regarding the feasibility of delivering packages of neonatal interventions, particularly at the community level, as TBAs and CHWs may be limited in their ability to provide multiple interventions, and delivering packages of interventions can require, or be perceived to require, intensified logistical coordination and resource inputs that may be challenging in some low-resource settings.

Very few RCTs examined the impact of intervention packages on neonatal health outcomes (Tables 1, 2). Similarly, differences in study design, varying combinations of interventions and lack of cause-specific mortality data precluded a meta-analysis to project the impact of specific packages on neonatal mortality. Reasons for the dearth of RCTs and rigorous prospective studies on the impact of intervention packages likely include lack of donor support, resource and technical limitations, ethical considerations and, for RCTs, the logistical, financial and analytical challenges associated with factorial designs.

Recommendations and reality

Notably, all of the reviewed studies neglected to include key interventions known to be efficacious and/or cost-effective, including periconceptual folic acid supplementation, syphilis screening and treatment, particularly in sub-Saharan Africa, intermittent presumptive treatment for malaria, antibiotics for asymptomatic bacteriuria, corticosteroids for preterm labour and antibiotics for PPROM (older studies conducted prior to the development of an efficacy and cost-effectiveness evidence base are exempt from this criticism). This may be due, at least in part, to the fact that a number of these interventions require advanced health systems infrastructure (i.e. ‘additional’ interventions such as detection and treatment of bacteriuria, corticosteroids for preterm labour and antibiotics for PPROM) or strong outreach services (e.g. periconceptual folic acid supplementation, intermittent presumptive treatment for malaria). Only two packages attempted to ensure skilled attendance at birth, despite Safe Motherhood guidelines (Srinivasan et al. 1995; O’Rourke et al. 1998). Only one study implemented the full family-community package of evidence-based essential newborn care (Bang et al. 2005), and only one offered a near-complete antenatal outreach package (Srinivasan et al. 1995). While many studies conformed to recommended postnatal packages, there were no examples of complete (or near-complete) intrapartum packages, a significant gap in the evidence base and a missed programmatic opportunity to save maternal and newborn lives. Few studies demonstrated continuity of care through the antenatal, intrapartum and postnatal periods (Darmstadt et al. 2005).

Conversely, some studies employed at least one intervention for which evidence of maternal or neonatal benefit is insufficient (Bhutta et al. 2005). For example, three incorporated hyperbilirubinemia screening, although the incidence of kernicterus in these settings is not unusually high (Bang et al. 1990; van der Mei 1994; Bose et al. 1999), suggesting a lack of focus on life-saving interventions.

Potential impact on neonatal mortality

The Lancet's Neonatal Survival Series projected that reductions in neonatal mortality achievable through outreach and family-community services alone range from 18–37% (Darmstadt et al. 2005). Although a meta-analysis was not possible, empirical evidence suggests that this estimate may be conservative, at least under conditions testing efficacy. Particularly in countries with high incidence of tetanus neonatorum, the empirical evidence suggests that including tetanus toxoid immunization and clean delivery in packages can dramatically reduce neonatal deaths from tetanus and sepsis (Rahman 1982; Berggren et al. 1983; Kapoor et al. 1991; O’Rourke et al. 1998; Bang et al. 1999; Meegan et al. 2001), lending support to The Lancet's recommendation that outreach services during initial programme implementation in areas with high tetanus incidence and weak health systems may be strategic (Figure 1).

Intentional packaging

The concept of evidence-based neonatal care packages is relatively new, and thus many studies qualifying as packages of interventions here may not have been originally conceived of as packages. Many interventions appear to have been bundled primarily out of logistical convenience, donor directives, organizational expertise or specific lines of scientific inquiry rather than consideration of delivery mode, biological or behavioural synergy, or cost-effectiveness (Victora et al. 2004b). Still, opportunistic implementation of interventions may be an effective initial approach to achieving impact while building a functional health system. Overall, this review suggests that the impact of packages of interventions on health outcomes could likely be improved through a more rational approach to the implementation of cost-effectively and synergistically bundled evidence-based interventions.

Tailoring packages to health systems and regions

Significant geographic and cultural diversity between regions and countries, as well as differences in health system capacity (Box 1, Figure 1), must be considered in intervention design and scaling-up to the national level. Local patterns of health behaviours, prevalence of diseases and outcomes, and service availability should determine components of intervention packages; impact of a package in one setting will not necessarily translate to another. Also, a synergistic benefit of packaged interventions is not assured, as benefit can be diminished by compromised delivery performance from poor training or lack of supervision, deficiencies in supply systems, and selection of interventions poorly matched to the public health needs of a population.

From a health systems perspective, the reviewed studies showed few linkages between family-community interventions and the existing health care infrastructure, although a few studies effectively spanned these service delivery modes as well as outreach (Rahman 1982; Srinivasan et al. 1995; Manandhar et al. 2004) and created some linkages between communities, TBAs and facilities (Bartlett et al. 1991; Srinivasan et al. 1995; Dutt and Srinivasa 1997; Manandhar et al. 2004). However, as noted in the analysis of care-seeking components, few studies employed comprehensive efforts to not only identify and refer obstetric and/or neonatal complications, but also to minimize parental refusal, overcome barriers to transferring patients, and ensure facilities had the capacity to handle these referrals. More such studies that develop new ways to build, strengthen and evaluate these linkages can generate data useful for health systems strengthening.

Designing interventions tailored to specific contexts is critical for scalability, sustainability and equity (Victora et al. 2004b). Regional differences in health systems and general infrastructure, differences in prevalence of syphilis, malaria and tetanus, and differential rates of perinatal and maternal outcomes (e.g. obstructed labour, LBW and mortality) suggest that packages of evidence-based interventions must be tailored to specific contexts to maximize impact. For example, of 15 studies conducted where malaria and/or syphilis were endemic, only three provided malaria chemoprophylaxis (Greenwood et al. 1990; Bouvier et al. 1997; Ndyomugyenyi and Magnussen 2000) and none provided syphilis screening and treatment. The capacity of the existing health system must also be considered in selecting intervention components for inclusion in packages; some interventions—particularly clinical interventions requiring advanced diagnostic or treatment strategies, such as diagnosis and treatment of asymptomatic bacteriuria, or use of antenatal antibiotics or intrapartum corticosteroids—could be infeasible in the absence of a well-functioning health system (Figure 1). We reviewed only packages delivered in developing countries, so few of the interventions in these packages were delivered by trained doctors, illustrating and partially explaining the scarcity of clinical interventions in these packages. Lack of integration of community-based care with facility-based care is also apparent.

Implications of packages data for health systems research

Among the reviewed studies, the lack of rigorous effectiveness data—data from trials making wide use of existing human and material resources in health systems and documenting external inputs including costs—is a key research gap. The majority of the RCTs we identified were antenatal, outreach-based micronutrient supplementation trials conducted in carefully selected, controlled environments and heavily dependent on external inputs of financial, human and material resources, i.e. they were efficacy trials. While some studies we reviewed were conducted in more realistic settings, none had sufficient rigour to extract accurate effectiveness data.

Victora et al. (2004b) distinguished between intervention efficacy and service delivery mode, arguing that efficacy trials use methods of service delivery that maximize the likelihood of a positive outcome. As none of the studies we identified were true effectiveness trials, Victora's distinction suggests that conducting such trials will require consideration of existing resources, use of multiple service delivery modes, and greater collaboration with health systems to avoid biasing effectiveness measurements.

Packaging of behavioural and technical interventions during the intrapartum and postnatal periods was lacking. To reduce neonatal mortality, a more holistic approach is needed that packages technical interventions within a broader context of behaviour change communications and community mobilization activities, where possible.

Community mobilization as enhancer of effectiveness

This review illustrates the value of community mobilization and empowerment in creating a context in which multiple other health interventions can be effectively provided (Kumar et al. 2004; Manandhar et al. 2004). Those studies that were concerned with community mobilization and empowerment appeared to provide fertile ground for facilitating the uptake of other interventions, even more technically oriented interventions. Frequent deployment of private community members as intervention providers in the absence of community mobilization activities (Rahman 1982; Berggren et al. 1983; Pratinidhi et al. 1986; Janowitz et al. 1988; Bhakoo et al. 1989; Bartlett et al. 1991; Kapoor et al. 1991; Daga et al. 1992; Daga et al. 1993; Daga et al. 1996; Bouvier et al. 1997; Bose et al. 1999; Smith et al. 2000) appears to reflect missed opportunities to enhance the effectiveness of interventions. As studies with community mobilization components had some of the highest documented impacts on NMR and PMR, community mobilization activities themselves may function to enhance the level of effectiveness of component interventions, a contention that requires further evaluation.

When implemented, documentation of community mobilization activities is typically poor, and standardized approaches to monitoring and measuring coverage and impact, and to describing the implementation and evaluation process, are not yet available. Because community mobilization, particularly participatory approaches, can enhance uptake, effectiveness and sustainability of a broad range of interventions (Rifkin 1990; Rifkin 1996; Morgan 2001), innovative strategies to quantify its impact on primary and secondary outcomes are needed.

Cost-effectiveness

Cost-effectiveness data are crucial to select and bundle interventions suitable for scaling up and to tailor interventions to available health systems resources, but empirical cost-effectiveness data are scarce. The two studies that did include cost-effectiveness analyses were both conducted in rural areas of South Asia, and their costing methods and estimates varied widely (Box 2). While cost-effectiveness analyses are rare throughout the public health literature, especially among developing country studies, such data are necessary to test the implicit assumption that bundling interventions by like service delivery mode is cost-effective. Cost-effectiveness of packaged interventions is a priority research area; researchers could facilitate cost-effectiveness meta-analyses by collecting cost-effectiveness data in a uniform manner and including the data in study results.

Conclusions

At present, the empirical evidence for the impact of neonatal health care packages is a weak base on which to build effective programmes (Box 3). An evidence-based approach to packaging interventions is clearly needed but requires substantial investment in high-quality research and programme learning. Intervention packages should employ multiple service delivery modes along the continuum of care spanning the periconceptual to the postnatal periods, and facilitate linkages between communities and available health care facilities. While 14 of the 41 studies in this review incorporated at least one intervention component to encourage these linkages and many encouraged referrals, few studies made the investments necessary to enhance demand creation for care, to improve the likelihood that parents would appropriately recognize complications or accept referrals, to assist with transportation to a facility, or to improve the capacity of health facilities to provide quality care and handle increased demand.

Finally, significant research is needed regarding scaling up the delivery of interventions to large populations (i.e. the national level, or the state level in large countries with diverse populations) and measuring their impact. Care should be taken that efficacy and effectiveness trials are conducted in settings, and using delivery strategies and resources that render scale-up feasible. Effectiveness interventions should foster synergies between programmes, health systems, private providers and community-based workers. Effectiveness trial data can inform new strategies for strengthening health systems and effectively delivering interventions at scale. Another approach to gauge effectiveness, though sometimes logistically and ethically complicated, but particularly advantageous in situations where a non-intervened comparison group is not possible, is the use of step-wedge introduction of packages in cluster-randomized communities to measure impact, which, in a sufficiently large population, could also be used to test different combinations of interventions at the population level.

An analysis of the impact of successful neonatal health interventions at scale, including careful documentation of processes (e.g. project management), is the next logical step in developing evidence-based packages of interventions to save newborn lives.

Acknowledgements

Funding for this study was provided by the Saving Newborn Lives Initiative of Save the Children-US through a grant from The Bill & Melinda Gates Foundation, and by the Office of Health, Infectious Diseases and Nutrition, Global Health Bureau, United States Agency of International Development, Washington, DC, USA, award GHS-A-00–03–00019–00 given to the Department of International Health at The Johns Hopkins Bloomberg School of Public Health through the Global Research Activity Cooperative Agreement. The funding sources had no role in determining the content of the paper. The opinions expressed herein are those of the authors and are not necessarily the views of any of the agencies.

References