Effectiveness and cost-effectiveness of home-based postpartum care on neonatal mortality and exclusive breastfeeding practice in low-and-middle-income countries: a systematic review and meta-analysis

Randomized trials and quasi-experimental studies that report effectiveness or cost-effectiveness of postpartum home-based interventions on practice of exclusive breastfeeding and neonatal mortality or economic evaluation studies that compared home-based interventions with routine care were included. Articles written in the English language, with no publication date restriction to get adequate studies and interventions implemented in LMIC were considered for this study.

Citations without abstracts and/or full text, commentaries, letters, duplicate studies and editorials, studies written by different languages other than English were excluded from the review. Additionally, studies were excluded if it, 1) did not report on postpartum home visit, newborn mortality, exclusive breastfeeding, or cost, 2) did not have comparator, and 3) were not done in LMIC setting.

The research team searched peer-reviewed journal articles from electronic databases including PubMed, Popline, ClinicalTrials.​gov, Cochrane Central Register of Controlled Trials and National Health Service (NHS) Economic Evaluation databases. For more articles, reference lists of the initial search were also checked.

First, we identified concepts including 1) effectiveness or cost-effectiveness, 2) postpartum care or neonatal care, and 3) LMICs using the PICO review acronym—Population, Intervention, Comparison, and Outcome. Then for each concept, search terms (including synonyms and MeSH terms) were identified and used in a variety of combinations for these concepts. For database search, combination of the following keywords were used “(postnatal care OR postpartum OR puerper* OR post partum OR post natal OR neonatal OR newborn) AND (home visit OR home care OR facilit* OR hospital care OR health center care OR facility based care OR home care service*) AND (infant mortality OR neonatal mortality) OR (breastfeeding OR exclusive breastfeeding) OR (cost benefit analysis OR cost analysis OR efficien* OR health care cost OR economic evaluation OR cost effectiveness OR impact OR effect*))”. A search strategy for PubMed, Popline, Cochrane Central Register of Controlled Trials and NHS Economic Evaluation databases are included in the additional file (Additional file 1). ClinicalTrials.​gov was searched for recently completed trials.

Interventions—including counseling; examination and management; and provision of services—provided to women and newborns in the first six weeks after birth at home by health providers or community health workers.

Any home visit innovations, initiatives, approaches, or activities carried out during the postnatal period, with the aim of either providing maternal and newborn health service or influencing maternal and newborn care-seeking behavior and practices, including home-based treatment for illness, community mobilization efforts, and any home-based/community-based neonatal interventions, were considered as home-based postnatal care.

Routine postnatal care provided to mothers who delivered in a health facility on discharge or care provided after discharge within six weeks when the women visited the facility. Usually, in LMICs, PNC is provided on discharge after facility delivery within 6 h and when the women visited the facility for family planning or immunization services after discharge within six weeks.

Low-and-Middle-Income Countries are countries with a gross domestic product (GDP) per capita of less than US $4655 as categorized by the World Bank in 2017. The list of countries is included in the additional file (Additional file 1).

The definitions for the outcome variables— neonatal mortality, exclusive breastfeeding, and cost-effectiveness—are given below.

Exclusive breastfeeding: women who exclusively breastfed their child as per the age of the neonate determined by each study at the time of the survey.

The cost results reported in non-US dollars (USD) were converted to USD and inflated to 2016 prices [20, 21]. The GDP per capita was used as a benchmark to consider against the cost-effectiveness of the intervention. For this purpose, the 2016 country’s USD prices were used [21]. World Health Organization (WHO) considers strategies and interventions to be cost-effective if the cost per DALY averted is less than three times the GDP per capita and highly cost-effective if less than the GDP per capita [22].

The search returned 1081 records after removing duplicates. Endnote reference manager was used to upload search results and create library of all search results for the purpose. Two review authors (GT and CB) independently screened the titles and abstracts yielded by the search against the inclusion criteria. We obtained full reports for all titles that appear to meet the inclusion criteria. Discrepancies between reviewers regarding the decision of inclusion were resolved through discussions.

During the title and abstract review stage, publications were excluded based on the exclusion criteria. Additionally, at the full-text review stage, publications were excluded if it did not report on our outcome measures and presented secondary rather than primary analysis (in which case the references were checked for more articles for inclusion).

Accordingly, 42 full articles were identified from screening titles and abstracts. The final synthesis was based on 14 journal articles. The results of the search and the process of screening and selecting studies for inclusion are illustrated using the study flow diagram seen below (Fig. 1).

The risk of bias of each included study was assessed using the Cochrane Collaboration criteria: sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete data, selective reporting and other biases [23]. Through discussion, researchers evaluated the possible risk of bias of each of the six domains from the extracted information and rated as high risk, low risk or unclear, if there was insufficient detail reported in the study, based on the criteria for judging the risk of bias. Graphic representations of risk of bias within and across studies were computed using Review Manager Software v5.3 (RevMan) software [23, 24].

In cluster-randomized trials, allocation concealment was treated as low risk if trials randomized all clusters at once though it lacked concealment of an allocation sequence [23]. In addition, recruitment bias, contamination bias, baseline imbalance, loss of clusters, and lost to follow-up were considered in the analysis of the risk of bias in cluster-randomized trials.

The quality of the evidence pertaining to the outcomes was graded using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach based on study limitations, consistency of effect, imprecision of effect estimates, indirectness of evidence and potential publication bias [23]. Researchers jointly evaluated the quality of the evidence for each outcome through discussion and a Summary Findings table was produced using GRADEPro software [25].

Standard reporting checklists—Preferred Reporting Items for Systematic and Meta-Analysis (PRISMA) for systematic review and meta-analysis and International Society for Pharmacoeconomics and Outcomes Research (ISPOR) for the trial-based cost-effectiveness analysis— were followed to establish minimum information that should be included when reporting.

The quality of the trial-based cost-effectiveness studies was assessed based on the core items recommended for conducting economic analyses alongside clinical trials [26]. Each study was given a 1-point score for each item fully met, 0.5 for partially met, and 0 for none. Then, trials that scored 75% or more were categorized as high quality, trials scored between 50 and 74% were ranked as medium, and studies scored less than 50% scores were rated as poor.

The first reviewer (GT) completed the data extraction form for all studies using RevMan software [23, 24], and the second reviewer (CB) assessed the accuracy of the extracted data. Differences were resolved through discussions.

For those cluster-randomized trials where the analysis did not properly account for cluster design or did not have sufficient information regarding proper accounting of cluster design, we extracted the data after correcting the analysis to reduce the size of each trial to its effective sample size as described in the Cochrane Handbook [23]. The effective sample size of a single intervention group in a cluster-randomized trial is its original sample size divided by the design effect. The design effect is given by.

where M is the average cluster size and ICC is the intra-cluster correlation coefficient.

We used an estimate of 0.02 ICC derived from the previous trials [27‐32].

We identified a trial [33] which included more than two arms. As such, we combined the groups using the methods described in the Cochrane Handbook [23] to analyze them.

Descriptive information about the eligible studies was summarized using text and Tables. A narrative synthesis was used to analyze and interpret the findings.

Random-effects meta-analysis model [23, 34] was used to pool the estimates of the outcomes, accounting for the variability among studies using Stata v15 [35]. Both single and multiple level meta-regression analysis was done to identify the independent predictor variables for the outcome of interest.

The results were presented as risk ratio (RR) & odds ratio (OR) with 95% confidence intervals and the estimates of Tau² and I² statistic for heterogeneity. We also investigated the presence of publication and other bias in the extracted data using funnel plot and Stata’s “metabias” command [34, 35].

However, meta-analysis was not appropriate for the cost-effectiveness outcomes as the analytic approaches were different among trials which makes it difficult to combine the outcomes together.

The P-value of the Chi-squared test of heterogeneity and the I² and Tau² statistics were examined for heterogeneity between the trials to judge whether there were any apparent differences in the direction or size of the treatment effect between studies. The thresholds for the presence of heterogeneity among the trials were determined if the value of I² was greater than 30%, and the value of Tau² was greater than zero or the P-value of the Chi² test for heterogeneity was greater than 0.1.

We undertook subgroup analysis to examine the effect of home visit on neonatal mortality varied by 1) number or frequency of PNC visits (< 3 vs. > 3 visits), 2) coverage of PNC visits made to the mother and/or newborn that is proportion of neonates and/mothers receiving a postnatal visit in the intervention areas/clusters) (< 70% vs. > 70%), 3) type of care (preventive interventions vs preventive and curative interventions, 4) type of health worker (health professionals vs. community health workers (CHWs)), 5) intervention components (community mobilization efforts and home visits vs. home visits alone), 6) time of home visits (antepartum and postpartum vs. postpartum only), and 7) publication year (< 2008 vs. > 2008). Likewise, subgroup analysis was done for exclusive breastfeeding outcome which varied by, 1) number or frequency of PNC visits (< 3 vs. > 3 visits), 2) coverage of PNC visits (< 70% vs. > 70%), 3) duration of intervention or follow-up (< 4 vs. > 4 weeks), 4) age at exclusive breastfeeding (neonatal period vs. more than neonatal period), 5) type of health worker (health professionals vs. CHWs), and 6) publication year (< 2008 vs. > 2008).

In subgroup analysis, we presented both random-effects and fixed-effects model estimates to visualize the presence of small-study effects; however, due to substantial heterogeneity, we preferred to discuss the results acquired from random-effects model.

We conducted a sensitivity analysis excluding trials that evaluated the effect of home-based follow-up of neonates born at and recruited from hospitals, to determine the pooled estimate of trials with home-based neonatal care which had poor access to health facilities.

The characteristics of included studies are given in Table 1 below. Fourteen articles (2 randomized controlled trials (RCT), 11 cluster-randomized trials, and 1 quasi-experimental studies) were included. All included trials were published from 1999 to 2017.

Most of the trials were conducted in predominantly rural districts that had poor infrastructure (poor roads, communications, education, and health services). However, three studies were conducted in urban [33] or peri-urban [36, 37] settings which were within a short distance to the countries’ capital cities (Table 1).

Twelve trials recruited CHWs and three studies employed health professionals including midwives and nurses. Health workers were trained for about a week to make home visits to offer preventive and promotive newborn care services. Table 2 presents the intervention packages implemented in each trial. All trials conducted postpartum home visits to promote newborn care practices; 11 trials did both antepartum and postpartum home visits. Moreover, nine trials conducted community activities [28, 29, 38‐43] in addition to home visits to promote newborn care practices. Five of the studies trained providers/CHWs to identify sick newborns, refer them to facilities, and if referral was not possible to treat sick newborns in addition to preventive and promotive newborn care services [38, 40, 42, 44, 45] (Table 2).

Regarding the implementation strength of the studies, defined here as frequency of home visits and coverage of visits, each trial had an average of 4 postpartum home visits and an average of 1 antepartum visit (usually at third trimester). Likewise, coverage of home visits, proportion of mothers and/or their newborns received postpartum home visit, ranged from 30 to 100% (median 79%).

Studies’ follow-up period ranged from 1 to 24 weeks postpartum and more than half of the studies (53%, n = 8) followed-up mothers and their newborns for one week after delivery. The exclusive breastfeeding outcome definition varied across studies; three trials [29, 37, 44] determined exclusive breastfeeding at neonatal age while others determined at 3 months [36], 4 months [33], or 6 months [46]. Three studies observed exclusive breastfeeding in the previous 24 h recall at 12 weeks of age [37], 24 weeks of age [36], and age between days 26 to 32 after birth [29] (Table 2).

The risk of bias of included trials is presented in Fig. 2.

Most trials (n = 11; % = 79) included had low risk of random sequence generation bias [28, 29, 36, 37, 39, 40, 42‐45, 47]; three studies had high risk of random sequence generation bias [33, 38, 41] and one trial had unclear risk of random sequence generation bias [47]. Except for one [41] all other studies had low risk of allocation concealment bias.

Blinding of participant and personnel was adequate in 11 of the studies included [27, 28, 33, 36, 38, 41‐43, 45, 47, 48]. Four studies had high risk, either performance or detection bias [33, 37, 41, 42] and six and seven studies had an unclear risk of performance bias [29, 39‐41, 43, 44] and detection bias [29, 38‐40, 43, 44, 47], respectively.

One trial had high risk of reporting bias [33]. Most studies had low risk of attrition; however, one trial [37] had unclear risk of reporting bias.

Other potential sources of bias including baseline imbalances were identified in two studies [33, 44]. Moreover, possible cluster-level time-varying unmeasured confounders including regional pregnancy leave and employment policy might bias the exclusive breastfeeding outcome. However, there is no compelling reason to believe that these systematically influenced the intervention areas but not in the comparison areas unless otherwise, it happened after allocation of the treatment.

The Summary of Findings table gives estimates of the effects of home-based PNC on neonatal health outcomes as well as grading the level of evidence (Table 3).

Nine studies [28, 29, 33, 38, 39, 41, 42, 44, 45] involving 93,083 participants reported the effect of home visits on neonatal mortality as compared with the routine postpartum care. As presented in Fig. 3 below, the pooled analysis showed that home-based PNC reduced neonatal mortality by 24% (RR 0.76, 95% CI 0.62–0.92) with substantial heterogeneity between studies (I² = 69.0%; p = < 0.01; Tau² = 0.0393).

Sensitivity analysis was conducted excluding trials that evaluated the effect of home-based follow-up of neonates born in and recruited from hospitals [33]. Following removal of this trial, the overall pooled estimate increased by only 1% (RR 0.75, 95% CI 0.62–0.91) with a slight increase in heterogeneity between studies (I² = 70.9%; p = .001) indicating no or little difference in the outcome and results were not driven by a trial that recruited newborns from hospitals.

The subgroup analysis revealed that more than three PNC home visits contributed to reduction in neonatal mortality (RR: 0.70; 95% CI: 0.53–0.91) than trials with less than three PNC home visits (RR: 0.77; 95% CI: 0.61–0.98; heterogeneity p = .043). Home visits by community health workers were associated with better survival of neonates (RR: 0.69; 95% CI: 0.55–0.87) than visits by health professionals (RR: 1.26; 95% CI: 0.37–4.30; heterogeneity P = .001). Regarding intervention characteristics, community mobilization efforts with home visits to promote newborn care practices helped reduced neonatal mortality (RR: 0.69; 95% CI: 0.54–0.88) than home visits alone (RR: 0.97; 95% CI: 0.90–1.05; heterogeneity P = .001). Moreover, trials with publication year 2008 and before revealed greater reduction in neonatal mortality (RR: 0.58; 95% CI: 0.40–0.85) than publication year after 2008 (RR: 0.94; 95% CI: 0.88–1.01; heterogeneity P < .01).

Trials with coverage of home visits of 70% and above, showed non-statistically significant greater reduction in neonatal mortality (RR: 0.70; 95% CI: 0.50–0.99) than trials with less than 70% coverage of home visits (RR: 0.87; 95% CI: 0.76–1.00). Likewise, statistically non-significant trend towards a greater effect on mortality was observed with curative (inject-able antibiotics) and preventive interventions (RR: 0.82; 95% CI: 0.63–1.05), as compared to only preventive intervention (RR: 0.70; 95% CI: 0.48–1.03; heterogeneity P = .016) (Table 4).

The funnel plot (Fig. 4) appeared symmetrical, which suggests no evidence of small-study effects. The Egger’s test also indicated low possibility of publication bias (Coef. = − 1.263; p = .130).

Six trials [29, 33, 36, 37, 44, 47] involving 20,624 mothers reported on the exclusive breastfeeding outcome. The pooled analysis showed (Fig. 5) that the odds of exclusive breastfeeding practice of mothers in the home visit group were about three times higher than the routine care group [OR: 2.88; 95% CI:1.57–5.29) with substantial heterogeneity between studies (I² = 98.2%; p = <.01; Tau² = 0.5517).

This outcome had substantial heterogeneity; as such, subgroup analysis was carried out. The subgroup analysis showed that groups of mothers who had more than 70% home visits (OR: 4.06; 95% CI: 2.60–6.36) were exclusively breastfeeding significantly more than their counterparts (OR: 1.51; 95% CI: 1.30–1.76). Likewise, groups of mothers who had more than four weeks of follow-up (OR: 5.69; 95% CI: 5.29–6.12) were exclusively breastfeeding significantly more than their counterparts (OR: 2.94; 95% CI: 0.91–9.50).

We found no significant difference among those mothers who had more than three visits (OR: 3.54; 95% CI: 1.64–7.65) and those who had less or equal to three home visits (OR: 2.36; 95% CI: 0.83–6.71). Though statistically non-significant, home visits by health professionals were associated with better exclusive breastfeeding practice (OR: 0.69; 95% CI: 2.60–8.69) than visits by community health workers (OR: 2.25; 95% CI: 1.01–5.29; heterogeneity P < .01). Trials with publication year 2008 and before did not show significant difference in exclusive breastfeeding practice (OR: 2.94; 95% CI: 0.91–9.50) than publication year after 2008 (OR: 2.86; 95% CI: 1.36–6.04; heterogeneity P < .01) (Table 5).

The cost-effectiveness results are presented in Table 6 below. Three trials reported the incremental cost for neonatal mortality outcomes. Different cost-effectiveness and cost-utility measures were used. All three trials reported cost per neonatal death averted [40, 38, 43] or newborn life-year saved [40, 43]. One trial used cost-utility measures (i.e., cost per DALYs gained) [40].

According to World Health Organization’s Choosing Interventions that are Cost-Effective project (WHO-CHOICE) recommendations [49], considering cost per neonatal death averted, newborn life-year saved, and DALY averted measures, home-based neonatal care strategies were found to be cost-effective.