Introduction

High rates of neonatal and early infant deaths from preventable causes are major health concerns, especially in resource-limited countries; lack of awareness, non-identification of the health condition, delay in referral and hesitation in seeking prompt treatment are important contributors. An increasing proportion of child deaths are in sub-Saharan Africa and Southern Asia.1 About 73% of under-five deaths are estimated to be due to acute respiratory infections (mainly pneumonia), diarrheal diseases, prematurity and low birth weight and neonatal infections such as sepsis, birth asphyxia, trauma and malaria. One third of all under-five deaths occur in the neonatal period, the three main causes being prematurity and low birth weight, neonatal infections, mainly sepsis and pneumonia and birth asphyxia and birth trauma.2 A World Health Organization report shows that about 44% of under-five deaths in 2012 occur in the neonatal period.3 Three quarters of newborn deaths occur in the first week of life and up to one half (about 2 million) on the first day itself,4 and therefore this may be the optimal time for home visits by trained health workers.5 The Young Infants Clinical Signs Study Group reported on the utility of various clinical signs in detection of severe illness in young infants (⩽59 days of age).6

Home visits by trained community health workers (CHWs) in the first 2 days of life have been shown to significantly reduce neonatal mortality.7 These deaths could be prevented through increased coverage of hospital-based treatment by enhancing availability and access; and by identifying effective treatment for a group of infants with severe infection/serious illness that could potentially be managed at a first level health facility or at the community level, when families do not accept or cannot access referral.

The majority of neonatal deaths can be prevented with quality care during pregnancy and immediate postnatal care. Similarly, most child deaths due to infectious diseases can be prevented by known, simple, affordable and low-cost interventions such as exclusive breastfeeding up to 6 months of age, immunization, appropriate use of antibiotics, oral rehydration therapy and zinc, insecticide-treated bed nets and anti-malarials.3 Diarrheal diseases and respiratory infections are the two main infective causes leading to mortality in all regions.2 The United Nations in its Millennium Development Goal 4 calls for a two-thirds reduction in the under-five mortality rate by the year 2015.1 Several studies have shown the beneficial role of CHWs in reducing neonatal and young infant mortality and morbidity.8, 9, 10 These CHWs can have a key role in early identification and prompt referral of neonates and young infants with danger signs and encourage the community to use health facilities.

Simple management and referral guidelines for common life-threatening conditions at community and primary health facilities, where referral to hospital is not possible, can potentially save lives. It is unclear whether home visits by trained CHWs improve correct identification of serious illnesses in young infants and their referral to facilities.

The main objective of this review was to evaluate the effect of home visits by trained CHWs to successfully identify newborns and young infants (up to 59 days of age) with serious illness and improve care seeking.

Materials and methods

Criteria for considering studies for this review

Types of studies

Randomized controlled trials (RCTs), randomized at the level of the individual child or cluster (such as village or district), were included for review. To be eligible, such trials had to have a concurrent comparison group (‘no home visits’ as an intervention) and adjustment for baseline characteristics and confounders. We only included RCTs with at least one intervention and one control arm.

Types of participants

Children 59 days of age or less in low- and middle-income countries (as categorized by The World Bank by Gross National Income per capita in US dollars, using the Atlas conversion factor).11

Types of interventions

The intervention was home visits by CHWs trained to identify serious illness in young infants. According to the World Health Organization, ‘CHWs are men and women chosen by the community and trained to deal with the health problems of individuals and the community, and to work in close relationship with the health services. They should have had a level of primary education that enables them to read, write and do simple mathematical calculations’.12 The comparison group had no home visits. Studies providing specific additional interventions in both intervention and comparison areas were eligible for inclusion, as long as these additional interventions were similar.

Types of outcome measures

Primary outcomes were:

  1. 1

    Successful identification of seriously ill young infants

  2. 2

    Improved care seeking from health facilities.

Search methods for identification of studies

We searched the Cochrane Central Register of Controlled Trials (The Cochrane Library), MEDLINE (from 1966 to 21 October 2014), EMBASE (from 1980 to 12 September 2014). The following search terms were used—‘community health worker’ AND (‘neonate’ OR ‘infant’) AND ‘home visits’ (Supplementary Table 1). Abstracts of all articles were read by two authors independently and the relevant articles were selected. There were no language restrictions. Experts and researchers in the field were contacted for information on additional completed or ongoing studies as needed. We also performed an electronic lateral search of the references of the selected articles.

Data collection and analysis

Selection of studies

Two authors independently reviewed all titles and abstracts retrieved to assess eligibility according to the inclusion criteria. Full text copies of all potentially eligible papers were obtained. Any disagreements were resolved by mutual discussion, with a third author, if necessary. We selected the studies that fulfilled our criteria.13

As data on successful identification of serious illness in newborns and young infants were not available from the RCTs on home visits, we separately searched for studies reporting on the ability of CHWs to successfully identify serious illness/signs of serious illness in newborns and young infants (Supplementary Table 2). These data are reported with sensitivity and specificity.

Data extraction and management

We developed a structured data extraction form to collect relevant information from the selected papers. Data were extracted on to the pretested form by two authors independently. Differences were resolved by discussion with the third author. We collected information about the randomization procedure of the study; setting; participants (the number of patients enrolled and the number of patients assessed); the demographic profile of the population studied (age range and gender); the diagnostic criteria and interventions. Any co-interventions were documented and the data on the outcomes (primary and secondary) were also recorded.

Assessment of risk of bias in included studies

Two review authors independently assessed the risk of bias for each controlled trial using the criteria outlined in the Cochrane Handbook For Systematic Reviews Of Interventions and those recommended by Effective Practice and Organisation of Care.14 The judgment for each entry involved assessing the risk of bias as ‘low’, ‘high’ or ‘unclear’, with the last category indicating either lack of information or uncertainty over the potential for bias. Plots of ‘risk of bias’ assessments were created in RevMan software.15 Any disagreements were resolved by discussion.

Measures of treatment effect

Risk ratio (RR) estimations with 95% confidence intervals (CIs) were used for dichotomous outcomes.16

Unit of analysis issues

For cluster RCTs, the stated cluster-adjusted RR and 95% CI were used, irrespective of the method employed.16

Dealing with the missing data

Where data were missing, we contacted authors in an attempt to obtain them.

Assessment of heterogeneity

We assessed for variability in the participants, interventions and outcomes studied to identify clinical heterogeneity, and for variability in study design to describe methodological diversity. Statistical heterogeneity was identified and measured as recommended by the Cochrane Handbook For Systematic Reviews Of Interventions.17A rough guide used for interpretation was:

  • 0 to 40%: might not be important

  • 30 to 60%: may represent moderate heterogeneity

  • 50 to 90%: may represent substantial heterogeneity

  • 75 to 100%: considerable heterogeneity.

As the number of trials included was small (only six), a P-value of 0.10 from the χ2-test was used to determine the statistical significance with regard to heterogeneity.

Assessment of reporting biases

Owing to the small number of studies, we could not evaluate the reporting biases by funnel plot18 or conduct formal statistical tests for funnel plot asymmetry, namely the Begg’s and Egger’s methods.19, 20

Data synthesis

We performed statistical analysis using Review Manager software (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark), and conducted the meta-analysis of included RCTs in concordance with current recommendations.14, 15 Pooled estimates (RR with 95% CIs) of the evaluated outcome measures were calculated by the generic inverse variance method. We report the CI together with the exact P-value.

We chose the random effects model in anticipation of expected variations in studies. For studies with more than one intervention arm, we pooled the data from two or more intervention arms for the purpose of final analysis, if the intervention groups had home visits by health workers as one of the components.

Results

Applying our search strategy, we found 423 titles. After checking reference lists, other sources and consulting experts, the full text of 20 articles was retrieved. Thirteen articles were then excluded as they did not meet our review inclusion criteria. Finally, seven full text articles were included in the review (Figure 1).

Figure 1
figure 1

Study flow diagram.

Full size image

Included studies

One of the seven studies included was an individual RCT21 and the other six were cluster RCTs.8, 22, 23, 24, 25, 26 A summary of included studies is shown in Table 1. Two trials8, 21 had two intervention arms; the other studies had one intervention arm with home visits (and other interventions) and the other control arm with no home visits but similar interventions. All the studies were done in middle- and low-income countries (as categorized by The World Bank by Gross National Income per capita in US dollars, using the Atlas conversion factor)11—Bangladesh, Ghana, India (two studies), Pakistan, South Africa and the Syrian Arab Republic. The baseline characteristics of included studies are shown in Supplementary Table 3. The health workers underwent training of some type in all the above-mentioned studies; the duration of training varied from 5 days to 15 months (Supplementary Table 4).

Table 1 Characteristics of included studies
Full size table

Excluded studies

Thirteen articles with full text were excluded from the review as they did not meet criteria. One study was a prospective follow-up study, not an RCT.27 In the rest of the studies, identification and referral of severely ill children was not reported as a study outcome.9, 10, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37 A summary of the excluded studies is shown in Table 2.

Table 2 Characteristics of excluded studies
Full size table

Risk of bias in included studies

The risk of bias summary for the included studies is shown in Figure 2.

Figure 2
figure 2

Risk of bias summary: review authors’ judgments about each risk of bias item for each included study.

Full size image

Allocation (selection bias)

Randomization was appropriately done in all included studies and rated as ‘low risk’ for random sequence generation (selection bias). For allocation concealment, only one study21 (which was an RCT) was rated as ‘low risk’. In the remaining studies the method of allocation concealment was rated as ‘high risk’.

Blinding (performance bias and detection bias)

Blinding was not achievable in any of the included studies as the two groups were easily identifiable from interviews. We rated studies as ‘high risk’ for this bias.

Incomplete outcome data (attrition bias)

There was ‘low risk’ of incomplete outcome data in any of the included studies.

Selective reporting (reporting bias)

There was ‘low risk’ of selective outcome reporting in the included studies.

Other potential sources of bias

There was ‘low risk’ of other possible sources of bias in the included studies.

Effects of interventions

The included studies evaluated the effect of home visits by CHWs on successful identification of serious illness in neonates and young infants and improved care-seeking at an appropriate health facility (Supplementary Table 5). For studies with more than one intervention arm8, 21 the data from the intervention arms were pooled for the purpose of final analysis as all the intervention groups had home visits as one of the components of the intervention. Of the seven selected studies, data from one study were not pooled for meta-analysis;26 this study was carried out in a high HIV prevalence settingin South Africa. Although some data on identification of illness and care seeking were given in the results, owing to the specific clinical setting and selective reporting of data, we found it inappropriate to pool the data of this study with other studies, so we provided a narrative summary of the results separately.

Studies comparing effect on successful identification of seriously ill young infants

None of the seven studies compared the diagnosis of illness in newborns and young infants made by the CHWs to a ‘gold standard’ diagnosis. The number of ill children identified in the intervention and control arms was available in six studies.8, 21, 22, 23, 24, 25 The way the studies mentioned the identification of illness in neonates and young infants varied, with three studies mentioning identification of complications or danger signs;8, 23, 25 while the other three did not specifically state the process of identification of illness.21, 22, 24

As the studies did not provide the data for the identification of sick young infants by CHWs, we were unable to synthesize the results. The number of sick young infants reported (a combination of identification by parents, CHWs and other health workers) is likely to be influenced by the effect of the intervention package on the reduction in morbidity in young infants.

In view of the lack of data from the intervention studies evaluating the impact of home visits, we carried out secondary analysis to determine whether trained CHWs were able to successfully identify sick newborns and young infants. A total of eight studies were found using our search strategy, and the full text of five of these studies was assessed for eligibility. Of these, we identified three evaluating the ability of CHWs to correctly identify serious illness in newborns and young infants.38, 39, 40 The sensitivity to identify serious illness in these studies ranged from 33.3 to 90.5% and specificity from 75.61 to 98.4% (Supplementary Table 5). In a study carried out in rural Nepal to identify optimal sign-based algorithms to define omphalitis in the community and to evaluate the reliability and validity of cord assessments by non-specialist health workers for clinical signs of omphalitis, the sensitivity and specificity of worker evaluations were high for pus (90% and 96%, respectively) and moderate for redness (57% and 95%, respectively). The authors concluded that ‘a composite definition for omphalitis that combined pus and redness without regard to swelling was the most sensitive and specific’.41 Two studies were excluded (Supplementary Table 6) as the data for calculating sensitivity and specificity were not available in one42 and in the other, the diagnosis made by the female community health volunteer was compared with that made by the facility-based CHW and not a physician.43

Studies comparing the effect on improved care-seeking at a health facility

Data on the effect of home visits by CHWs on improved care-seeking at a health facility were available in six studies.8, 21, 22, 23, 24, 25 Data were included for analysis purposes where care-seeking was from an appropriate provider, that is, doctor, nurse or trained paramedical worker. After pooling the data from five studies, care-seeking in the intervention group, that is, the home visit group, was better than in the control arm (RR=1.35; 95% CI=1.15 to 1.58) (Figure 3).

Figure 3
figure 3

Forest plot of comparison: (1) effect of ‘home visits versus no home visits’ by community health workers, outcome: (1.1) care-seeking from appropriate health care provider.

Full size image

One study reported a similar hospitalization rate for diarrhea of 5.7% (19/334) in the intervention arm and 4.5% (17/378) in the control arm (RR 1.28;95% CI 0.75 to 2.19).26 Among the infants of HIV-infected mothers, similar proportions were tested for HIV infection at 6 weeks (73.6% (420/571) in the intervention arm as compared with 66.6% (465/698) in the control arm; RR 1.1; 95% CI 0.97 to 1.25).

Discussion

Summary of main results

Seven RCTs fulfilled the inclusion criteria for this review. None of the RCTs compared the identification of seriously ill young infants by trained CHWs with a ‘gold standard’ diagnosis. As the data from the RCTs on the effect of home visits on outcomes in newborn and young infants did not allow us to evaluate the effect of the intervention on successful identification of seriously ill infants, we reviewed the studies that evaluated the ability of CHWs to identify these infants. These studies show that trained health workers can identify seriously ill children with sensitivity ranging from 33.3 to 90.5% and specificity from 75.61 to 98.4%. Data for improved care seeking from a health facility were available from six RCTs. Improved care seeking from an appropriate health provider was significantly higher in the intervention group than in the control group (RR= 1.35; 95% CI 1.15 to 1.58; Figure 3) but with high heterogeneity (I2= 94%; P<0.00001).

One study provided information on the hospitalization rate for diarrhea and the proportion of infants of HIV-infected mothers tested for HIV at six weeks; the outcomes were similar in the intervention and control arms.26

Overall completeness and applicability of evidence

The findings of our review are consistent with earlier publications that showed a beneficial role of CHWs in promoting maternal and child health and in reducing mother and child mortality rates.5, 9, 10, 44, 45 There was an evidence of improved care-seeking from an appropriate provider among ill neonates and young infants in the intervention group. We identified seven studies evaluating improved care-seeking from a health facility. There was substantial variation in clinical settings and relatively high heterogeneity (I2>50%) on pooling the results (Figure 3).

The data available from the included studies did not allow us to evaluate the successful identification of serious illness in newborns and young infants as the studies did not have a ‘gold standard’ for diagnosis to determine sensitivity and specificity. In a few studies, the information about sick children was obtained by interviewing the mother. Reporting of illness in young infants is likely to be influenced by various factors, such as the effect of the home visits program on morbidity and on parents’ awareness about illnesses in young infants, and also on the ability of CHWs to successfully identify seriously ill young infants. An optimal study design for assessing the effect of the home visits program should include verification of the diagnosis by a physician or other reliable health worker.

These results were obtained with trained CHWs in a research setting rather than in an implementation program.

Quality of the evidence

There was a high risk of bias as there was no blinding. We rated the outcomes of improved care seeking as ‘moderate quality’ (Table 3).

Table 3 Summary of findings
Full size table

Potential biases in the review process

We assessed all cluster RCTs as ‘high risk’ for blinding; it was easily identifiable by interviews whether participants belonged to a home visit group or to a non-home visit group. Similarly, allocation sequence concealment was not practical in such studies, giving ‘high risk' for this bias (Figure 2).

Agreements and disagreements with other studies or reviews

A systematic review in 2010 of controlled trials with five trials satisfying the inclusion criteria concluded that home visits for neonatal care by CHWs were associated with reduced neonatal mortality in resource-limited settings with poorly accessible health facility-based care when conducted along with other community mobilization activities.45 In another review in 2013, the authors found moderate to strong quality evidence for the use of CHWs in promoting essential newborn care strategies.44 CHWs were effective in increasing appropriate care-seeking behaviors for newborns by educating mothers on the importance of qualified care.

Conclusions

There was moderate quality evidence that home visits by trained CHWs are associated with improved care-seeking for ill young infants to health facilities in resource-limited settings. This evidence provides support for implementation of home visits by CHWs for improving outcomes of sick newborns and young infants in these areas. Appropriateness of referral may be an issue. Some referrals may be unnecessary (for example, low specificity but high sensitivity), but such a situation may be appropriate for low-resource settings, if the referrals are not so many as to overwhelm the system. It may be better to have high sensitivity in order not to miss any potentially life-threatening illnesses.

Further well-designed studies evaluating the effect of home visits by CHWs on successful identification of seriously ill newborns and young infants verified by a ‘gold standard’ should be carried out.