Effectiveness of interventions to manage acute malnutrition in children under 5 years of age in low- and middle-income countries: A systematic review

What is this review about?

Malnutrition among children under 5 years old is a major public health concern. This review assesses the evidence for the management of severe (SAM) and moderate acute malnutrition (MAM) according to the current World Health Organization (WHO) protocol, using facility- and community-based approaches. It also assesses the effectiveness of ready-to-use therapeutic food (RUTF), ready-to-use supplementary food (RUSF), prophylactic antibiotic use and vitamin A supplementation.

What is the aim of this review?

This Campbell systematic review summarises findings from 42 studies to inform policy on malnutrition among children under 5 years old.

What studies are included?

A total of 42 studies (48 papers) with 35,017 children were included in this review. All the studies were conducted in either community, hospital, health centre or nutrition rehabilitation centres in developing countries. All the studies targeted children with malnutrition, aged from 6 to 59 months. Thirty-three of the included studies were randomised controlled trials (RCTs). Six studies were quasi-experimental and three studies were cost studies.

Do programmes to combat severe and acute malnutrition work?

The included studies are studies with active controls, which means they compare one treatment with another. A finding of no effect means that the treatment does not work any better than the comparison treatment, and not that it does not work at all. The exception is the study of prophylactic antibiotics which are compared with no treatment.

Overall, the evidence shows that none of the interventions studied has any larger effect than the interventions to which they are compared. Prophylactic antibiotics given to children with SAM without complications can affect mortality.

For the other outcomes—recovery and weight gain—the evidence shows the following: –

Community-based approaches are better than standard care and in-patient management for recovery, but worse for weight gain and show no effect on mortality.

The only comparison showing positive effects on all three outcomes (recovery, weight gain and mortality) is prophylactic antibiotic compared with no antibiotic.

The quality of evidence is low, with high risk of bias, partly because of lack of blinding of the participants, personnel and outcome assessment. There is also high heterogeneity between studies, which is partly explained by imprecision on account of small sample size.

What do the findings of this review mean?

The evidence shows the equivalence of many approaches, so that decisions may be made on cost grounds. Existing limited cost data suggest that community or out-patient management of children with uncomplicated SAM and MAM is the most cost-effective strategy.

The evidence base remains thin and study quality is a concern. Future studies assessing the effectiveness of interventions should report pertinent nutrition-specific outcomes, including stunting, wasting, underweight, infections and potential adverse effects. Further studies should assess the relative cost and cost-effectiveness of various interventions addressing malnutrition in low- and middle-income countries.

How up-to-date is this review?

The review authors searched for studies published up to February 2019.

Types of studies

We included the following study designs:

RCTs, where participants were randomly assigned, individually or in clusters, to intervention and comparison groups. Cross-over designs were also eligible for inclusion.

Types of participants

We included studies targeting children under 5 years of age with MAM and SAM in low- and middle-income countries (LMIC). Studies including both eligible and non-eligible participants were included only if the results for the eligible participant subgroup was separately provided in the study. We used the following definition of MAM and SAM by WHO (WHO. 2013):

SAM: WHZ < −3 SD, WFH < 70% of the median National Center for Health Statistics (NCHS) or WHO reference or MUAC < 115 mm or oedema. Complicated SAM: SAM cases without appetite and/or with medical complications. Uncomplicated SAM:SAM children with successful standard appetite test, without fever, clinical infections or complications.

We excluded studies conducted on HIV populations specifically.

Types of interventions

The following interventions were considered and compared against the suggested comparison groups separately:

Community-based strategies to screen, identify and manage SAM and MAM compared with no community-based strategies [e.g., active community-based surveillance by CHWs vs. no active surveillance; training of CHWs for community-based screening vs. no training; community-based management with RUTF vs. standard care practices].

Types of outcome measures

We did not use the outcomes listed below as criteria for including studies but rather as a list of the outcomes of interest. We used denominators for the outcomes according to the intention to treat analysis to avoid misleading results.

Electronic searches

We searched the following databases till 11 February 2019: Cochrane Database of Systematic Reviews (CDSR) and the Cochrane Central Register of Controlled Trials (CENTRAL) in the Cochrane Library; World Health Organization regional databases; The Campbell Library; MEDLINE (PubMed); EMBASE; CINAHL; Web of Science; POPLINE; CAB abstracts and Global Health; PAHO; IndMED (indmed.nic.in/indmed.html and WHO Global Health Index. We also searched the WHO International Clinical Trials Registry Platform (ICTRP; http://www.who.int/ictrp/en/); ClinicalTrials.gov and Epistemonikos (https://www.epistemonikos.org)/. We did not restrict our searches by date, language or publication status.

Searching other resources

We contacted experts in relevant fields for identification of eligible studies for inclusion. We also went through the references of identified studies and relevant reviews. We also ran citation searches of included studies in Google Scholar and Web of Sciences for other potentially relevant papers.

Selection of studies

Two reviewers independently assessed relevant studies by screening the titles and abstracts for inclusion. The selected studies underwent full-text evaluation and were assessed for eligibility based on predefined eligibility criteria. Disagreements about appropriateness of the inclusion of studies were resolved by discussion between all review authors. Studies that meet the inclusion criteria on full-text screening but upon further investigation became ineligible were added to the “characteristics of excluded studies” table, along with the reasons for their exclusion. We also planned to contact the study authors regarding eligibility for studies where eligibility was unclear.

Data extraction and management

Two review authors independently extracted data on a predefined and pre-tested data extraction sheet. We extracted the following information, where available, from relevant studies and any discrepancies were resolved by group discussion.

Study method:

Study dates

Participants:

Sample size

Intervention:

Micronutrients and vitamin A supplementation (doses and timing)

Comparison group:

No intervention or placebo or standard practice or other treatment.

Outcomes

Primary and secondary outcomes, as outlined in the types of outcome measure section. We used denominators for the outcomes according to the intention to treat analysis to avoid misleading results

Quality assessment

On all Cochrane “Risk of bias” assessment tool indicators

Assessment of risk of bias in included studies

Two review authors independently assessed methodological quality of studies and any disagreements were resolved by discussion among all review authors. The Cochrane “Risk of bias” assessment tool (Higgins & Green, 2011) was used for RCTs and quasi-experimental (natural experiment) studies. We rated each of the following components as either “low risk”, “high risk” or “unclear risk” and provided justifications for the judgements: 1.

Selection bias (due to inadequate generation of a randomised sequence or concealment of allocations prior to assignment)

For CBA studies, we planned to use the Cochrane Effective Practice and Organisation of Care (EPOC) guidelines based on the following criteria (Cochrane Effective Practice and Organisation of Care [EPOC], 2017). We planned to rate each of the following components as either “low risk”, “high risk” or “unclear risk” and provide justifications for the judgements:

Baseline outcome measurements similar

For ITS studies, the following criteria from EPOC was considered (EPOC, 2017). We planned to rate each of the following components as either “low risk”, “high risk” or “unclear risk” and provide justifications for the judgements:

Intervention independent of other changes

Since all the included studies were either RCTs or quasi-experimental studies; we only used the Cochrane “Risk of bias” assessment tool. For future updates, we will use the aforementioned criteria for CBA and ITS studies.

Measures of treatment effect

We separately analysed the dichotomous and continuous outcomes. For dichotomous outcomes, we presented the results as summary risk ratios (RRs) with 95% confidence intervals (CI). We combined incidence data as RRs (events per child) and rate ratios (events per child year) because of their similar interpretation and scale. We presented continuous outcome data as either a mean difference (MD), if outcomes have been measured on the same scale, or a standardised mean difference, if outcomes have been measured on different scales, with 95% CI. If outcomes were reported at multiple time points in the included studies, we reported the outcomes at the last reported time period, unless other time point points were relevant for the subgroup analysis. For studies reporting outcomes at multiple time points, we reported the last outcome reported at last follow-up. We planned to conduct subgroup analysis for outcomes reported at different time periods.

Unit of analysis issues

We conducted separate meta-analysis for different study designs, that is, RCTs (both individual and cluster) and CBA/ITS; and for subcategories of interventions and outcomes. For cluster RCTs, we planned to contact trial authors for an estimate of the intra-cluster correlation coefficient if the clustering effect was not accounted for in the analysis, If we were unable to contact the trial author we planned to calculate an interclass correlation coefficient based on the other studies in the review and use the variance inflation factor to adjust the standard errors appropriately. Subsequently, effect sizes and standard errors were meta-analysed by using the generic inverse method in REVMan (RevMan, 2014). If there were multiple papers that describe the same trial, these were combined and coded as a single study. For trials that included multiple intervention arms, we selected one pair (intervention and control) that satisfied the inclusion criteria of the review and excluded the rest. If >2 intervention groups met the eligibility criteria, then these groups were combined into a single pair-wise comparison group and data were disaggregated into corresponding subgroups, or these arms were separated into different forest plots to ensure that there was no double counting of participants. Multiple outcome estimates within the same study were analysed separately.

Dealing with missing data

We reported the missing data or dropouts along with the reasons. We planned to contact the study authors if the missing data was not accounted for or the reasons for dropping out were unclear. If authors have accounted for missing data (i.e., multiple imputations), we used the adjusted data within our analysis.

Assessment of heterogeneity

Statistical heterogeneity was assessed using τ ², I ² and significance of the χ ² test; we also assessed heterogeneity visually using forest plots. Based on prior theory and clinical knowledge, we expected clinical and methodological heterogeneity in effect sizes in this literature. Therefore, we attempted to explain any observed statistical heterogeneity using subgroup analysis (see Section 4.3.10).

Assessment of reporting biases

If the number of studies were sufficient (>10), we planned to use a funnel plot to visually inspect for publication bias. In addition, we performed Egger's test to determine funnel plot asymmetry.

Data synthesis

Statistical analysis was carried out separately for each intervention using Review Manager 5.3 (RevMan, 2014). Separate meta-analyses were conducted for each type of intervention and comparison group and study design. Where analysis had not been ideal in the original papers, we attempted to reconstruct if the data presented allowed us to. Considering the expected heterogeneity in interventions, comparisons, outcomes and settings within the included studies, we used random effects meta-analyses. Where meta-analysis was deemed inappropriate due to substantial statistical or clinical heterogeneity between studies, the findings of the included studies were summarised in narrative form. In cases where we included multiple groups from one study, we combined all relevant experimental intervention groups of the study into a single group, and combined all relevant control intervention groups into a single control group or included each pair-wise comparison separately, but with shared intervention groups divided out approximately evenly among the comparisons to avoid double counts (Higgins & Green, 2011). We checked for the accuracy of the numeric data by comparing the magnitude and direction of effects reported by studies with how these were presented in the review,

We set out the main findings of the review for the primary outcomes in “Summary of findings” tables prepared via the GRADE approach (Guyatt et al., 2008) with GRADEpro 2014. We listed the primary outcome for each comparison with estimates of relative effects along with the numbers of participants and studies contributing data for those outcomes. For each primary outcome, we assessed the quality of the evidence using the GRADE approach, which involved consideration of within-study risk of bias (methodological quality), directness of evidence, heterogeneity, precision of effect estimates and risk of publication bias. We rated the quality of the body of evidence for each key outcome as “high”, “moderate”, “low” or “very low”. Randomised trials without important limitations provide high-quality evidence, while observational studies without special strengths or important limitations provide low-quality evidence. Non-randomised experimental trials (quasi-RCTs) without important limitations also provide high-quality evidence, but are automatically downgraded for limitations in design (risk of bias), such as lack of concealment of allocation

There are five criteria that can downgrade evidence for RCTs and quasi-RCTs (Guyatt et al., 2008):

Risk of bias in individual studies

There are three criteria that can upgrade the evidence for quasi-experimental studies with no serious methodological limitations (Guyatt et al., 2008):

Large magnitude of effect

Subgroup analysis and investigation of heterogeneity

Depending on data availability, we planned conduct exploratory subgroup analyses for the following subgroups:

Age (1–6 months, 6–59 months)

We planned to use the χ ² test to assess subgroup differences.

Due to the limited number of studies, we could not conduct the planned subgroup analysis; however, we have separately analysed the various supplementary foods that were compared with standard RUTF and standard RUSF.

Sensitivity analysis

We conducted sensitivity analysis based on the risk of bias of the included studies by removing studies judged to be at high risk of bias for sequence generation, allocation concealment and blinding of participants from the meta-analysis to determine whether the removal of studies with high risk of bias impacts the estimates.

Results of the search

Our searches identified a total of 8,451 potentially relevant titles from the electronic searches and 35 records from searching other sources. After removing duplicates, we screened 7,684 records for eligibility and excluded 7,618 on the basis of title and abstract. We obtained the full-text reports of the remaining 66 records, and of these, excluded 18 studies and included 42 studies (48 papers). Figure 1 depicts the search flow diagram.

OPEN IN VIEWER

Figure 1

Study flow diagram

Included studies

We included a total of 42 studies (from 48 papers) including 35,017 children (Ackatia-Armah et al., 2015; Ashworth, Huttly, & Khanum, 1994; Bahwere et al., 2014, 2016, 2017; Berkley et al., 2016; Bhandari et al., 2016; Chapko, Prual, Gamatie, & Maazou, 1994; Ciliberto et al., 2005; Diop, Dossou, Ndour, Briend, & Wade, 2003; Donnen et al., 1998; Fabiansen et al., 2017; Hossain et al., 2009; Hsieh et al., 2015; Irena et al., 2015; Isanaka et al., 2016; Jones et al., 2015; Karakochuk et al., 2012; LaGrone et al., 2012; Manary, Maleta, & Trehan, 2012; Manary, Ndkeha, Ashorn, Maleta, & Briend, 2004; Matilsky, Maleta, Castleman, & Manary, 2009; Maust et al., 2015; Medoua et al., 2015; Mishra, Rai, Swain, & Behera, 2019; Nackers et al., 2010; Nikièma et al., 2014; Oakley et al., 2010; Phuka et al., 2009; Puett et al., 2012; Sandige, Ndekha, Briend, Ashorn, & Manary, 2004; Sattar et al., 2012; Scherbaum et al., 2015; Shewade et al., 2013; Sigh et al., 2018; Singh et al., 2010; Stobaugh et al., 2016; Thakur, Singh, & Patel, 2013; Thakwalakwa et al., 2010; Vanelli et al., 2014; Versloot et al., 2017; Wilford, Golden, & Walker, 2011).

Thirty-three of the included studies were RCTs (Ackatia-Armah et al., 2015; Bahwere et al., 2014, 2016, 2017; Berkley et al., 2016; Bhandari et al., 2016; Chapko et al., 1994; Diop, Dossou et al., 2003; Donnen et al., 1998; Fabiansen et al., 2017; Hsieh et al., 2015; Irena et al., 2015; Isanaka et al., 2016; Jones et al., 2015; Karakochuk et al., 2012; LaGrone et al., 2012; Manary et al., 2012; Matilsky et al., 2009; Maust et al., 2015; Medoua et al., 2015; Mishra et al., 2019; Nackers et al., 2010; Nikièma et al., 2014; Oakley et al., 2010; Phuka et al., 2009; Sattar et al., 2012; Shewade et al., 2013; Sigh et al., 2018; Singh et al., 2010; Stobaugh et al., 2016; Thakwalakwa et al., 2010; Vanelli et al., 2014; Versloot et al., 2017); six studies were quasi-experimental studies (Ciliberto et al., 2005; Hossain et al., 2009; Manary et al., 2004; Sandige et al., 2004; Scherbaum et al., 2015; Thakur et al., 2013) while three of the included studies (Ashworth et al., 1994; Puett et al., 2012; Wilford et al., 2011) were cost-effectiveness studies.

Four of the included RCTs were cluster RCTs (cRCTs) (Ackatia-Armah et al., 2015; Irena et al., 2015; Karakochuk et al., 2012; Maust et al., 2015) while others were individually randomised trials. All the cRCTS were adequately adjusted for the effect of clustering.

See “Characteristics of included studies” tables.

Excluded studies

A total of 18 studies were excluded (Agha 2004; Aguayo et al., 2018; Ahmed et al., 1999; Ashworth et al., 1994; Bachou, Tumwine, Mwadime, Ahmed, & Tylleskar, 2008; Badaloo et al., 1999; Baker, Baker, Margo, & Reuter, 1978; Bhandari et al., 2001; Burza et al., 2016; Donnen et al., 2007; Dubray et al., 2008; Javan, Kooshki, Afzalaghaee, Aldaghi, & Yousefi, 2017; Linneman et al., 2007; Nagar, Nagar, & Gupta, 2016; Roy et al., 2005; Simpore et al., 2006; Zongo, Zoungrana, Savadogo, & Traoré, 2013). The major reasons for exclusion included study design not being appropriate and intervention not of interest. Please see “Characteristics of excluded studies” section.

Allocation (selection bias)

Twenty-eight studies were judged to be at low risk of bias for sequence generation since adequate methods were used to generate random sequence (Bahwere et al., 2014, 2016, 2017; Berkley et al., 2016; Bhandari et al., 2016; Dossou et al., 2003; Fabiansen et al., 2017; Hsieh et al., 2015; Irena et al., 2015; Isanaka et al., 2016; Jones et al., 2015; Karakochuk et al., 2012; LaGrone et al., 2012; Manary et al., 2012; Matilsky et al., 2009; Maust et al., 2015; Medoua et al., 2015; Nackers et al., 2010; Nikièma et al., 2014; Oakley et al., 2010; Phuka et al., 2009; Sattar et al., 2012; Shewade et al., 2013; Sigh et al., 2018; Singh et al., 2010; Stobaugh et al., 2016; Thakwalakwa et al., 2010; Versloot et al., 2017). Six studies were judged to be at unclear risk for sequence generation since they did not provide sufficient information regarding sequence generation methods (Ackatia-Armah et al., 2015; Ashworth et al., 1994; Chapko et al., 1994; Donnen et al., 1998; Mishra et al., 2019; Vanelli et al., 2014). Six studies were judged to be at high risk of sequence generation; all these studies were quasi-experimental studies (Ciliberto et al., 2005; Hossain et al., 2009; Manary et al., 2004; Sandige et al., 2004; Scherbaum et al., 2015; Thakur et al., 2013).

Twenty four studies were judged to be at low risk of bias for allocation concealment and used adequate methods to conceal the random assignment (Ashworth et al., 1994; Bahwere et al., 2014, 2016, 2017; Berkley et al., 2016; Bhandari et al., 2016; Fabiansen et al., 2017; Hsieh et al., 2015; Isanaka et al., 2016; Jones et al., 2015; Karakochuk et al., 2012; LaGrone et al., 2012; Manary et al., 2012; Matilsky et al., 2009; Medoua et al., 2015; Nackers et al., 2010; Oakley et al., 2010; Phuka et al., 2009; Sattar et al., 2012; Shewade et al., 2013; Sigh et al., 2018; Stobaugh et al., 2016; Thakwalakwa et al., 2010; Versloot et al., 2017). Ten studies did not provide sufficient information regarding methods used to conceal allocation and were judged to be at unclear risk of bias (Ackatia-Armah et al., 2015; Chapko et al., 1994; Donnen et al., 1998; Dossou et al., 2003; Irena et al., 2015; Manary et al., 2004; Maust et al., 2015; Mishra et al., 2019; Nikièma et al., 2014; Vanelli et al., 2014). Six studies were judged to be at high risk of bias for allocation concealment due to inadequate methods used to conceal the random allocation (Ciliberto et al., 2005; Hossain et al., 2009; Sandige et al., 2004; Scherbaum et al., 2015; Singh et al., 2010; Thakur et al., 2013).

Blinding (performance bias and detection bias)

Thirteen studies were judged to be at low risk of bias for blinding of participants and personnel (Bahwere et al., 2014; Berkley et al., 2016; Donnen et al., 1998; Hsieh et al., 2015; Isanaka et al., 2016; Jones et al., 2015; Manary et al., 2012; Matilsky et al., 2009; Medoua et al., 2015; Oakley et al., 2010; Sattar et al., 2012; Stobaugh et al., 2016; Versloot et al., 2017). Mishra et al. (2019) provided insufficient information regarding blinding of participants and personnel and was judged to be at unclear risk of bias. Majority of the studies (twenty-six studies: Ackatia-Armah et al., 2015; Ashworth et al., 1994; Bahwere et al., 2016, 2017; Bhandari et al., 2016; Chapko et al., 1994; Ciliberto et al., 2005; Dossou et al., 2003; Fabiansen et al., 2017; Hossain et al., 2009; Irena et al., 2015; Karakochuk et al., 2012; LaGrone et al., 2012; Manary et al., 2004; Maust et al., 2015; Nackers et al., 2010; Nikièma et al., 2014; Phuka et al., 2009; Sandige et al., 2004; Scherbaum et al., 2015; Shewade et al., 2013; Sigh et al., 2018; Singh et al., 2010; Thakur et al., 2013; Thakwalakwa et al., 2010; Vanelli et al., 2014) were judged to be at high risk of bias for inadequate blinding of the participants and personnel mainly due to the nature of intervention.

Sixteen studies were judged to be at low risk of bias for blinding of outcome assessment (Bahwere et al., 2014; Berkley et al., 2016; Donnen et al., 1998; Hsieh et al., 2015; Isanaka et al., 2016; Jones et al., 2015; LaGrone et al., 2012; Manary et al., 2012; Matilsky et al., 2009; Medoua et al., 2015; Oakley et al., 2010; Phuka et al., 2009; Sattar et al., 2012; Sigh et al., 2018; Stobaugh et al., 2016; Thakwalakwa et al., 2010). Mishra et al. 2019; and Versloot et al. 2017 provided insufficient information regarding blinding of outcome assessment and were judged to be at unclear risk of bias. Majority of the studies (twenty-two studies: Ackatia-Armah et al., 2015; Ashworth et al., 1994; Bahwere et al., 2016, 2017; Bhandari et al., 2016; Chapko et al., 1994; Ciliberto et al., 2005; Dossou et al., 2003; Fabiansen et al., 2017; Hossain et al., 2009; Irena et al., 2015; Karakochuk et al., 2012; Manary et al., 2004; Maust et al., 2015; Nackers et al., 2010; Nikièma et al., 2014; Sandige et al., 2004; Scherbaum et al., 2015; Shewade et al., 2013; Singh et al., 2010; Thakur et al., 2013; Vanelli et al., 2014) were judged to be at high risk of bias for inadequate blinding of the outcome assessment mainly due to the nature of intervention.

Incomplete outcome data (attrition bias)

Thiry five studies were judged to be at low risk of bias for attrition (Ackatia-Armah et al., 2015; Bahwere et al., 2014, 2016, 2017; Berkley et al., 2016; Bhandari et al., 2016; Chapko et al., 1994; Ciliberto et al., 2005; Donnen et al., 1998; Dossou et al., 2003; Fabiansen et al., 2017; Hossain et al., 2009; Hsieh et al., 2015; Isanaka et al., 2016; Karakochuk et al., 2012; LaGrone et al., 2012; Manary et al., 2004, 2012; Matilsky et al., 2009; Maust et al., 2015; Medoua et al., 2015; Mishra et al., 2019; Nackers et al., 2010; Nikièma et al., 2014; Oakley et al., 2010; Phuka et al., 2009; Sandige et al., 2004; Sattar et al., 2012; Shewade et al., 2013; Singh et al., 2010; Stobaugh et al., 2016; Thakur et al., 2013; Thakwalakwa et al., 2010; Vanelli et al., 2014; Versloot et al., 2017); while five studies (Ashworth et al., 1994; Irena et al., 2015; Jones et al., 2015; Scherbaum et al., 2015; Sigh et al., 2018) were judged to be at high risk of bias since >30% of participants were lost to follow-up.

Selective reporting (reporting bias)

Fourteen studies (Ashworth et al., 1994; Bahwere et al., 2014; Chapko et al., 1994; Ciliberto et al., 2005; Donnen et al., 1998; Dossou et al., 2003; Hossain et al., 2009; Irena et al., 2015; Manary et al., 2004; Mishra et al., 2019; Nackers et al., 2010; Scherbaum et al., 2015; Thakur et al., 2013; Vanelli et al., 2014) were judged to be at unclear risk of selective reporting since there was no information on trial registration or published protocols while all other studies were judged to be at low risk of bias for selective reporting.

Other potential sources of bias

We did not find any other potential sources of bias in any of the included studies.

Comparison 1: Community-based strategies to screen, identify and manage SAM and MAM compared with no community-based strategies

Two studies (Maust et al., 2015; Wilford et al., 2011) assessed community-based strategies. Maust et al. (2015) compared an integrated community-based protocol to manage MAM and SAM with no community-based management which comprised of no community-based surveillance, while Wilford et al. (2011) compared the cost-effectiveness of existing health services with CMAM to the existing health services without CMAM. We could not conduct meta-analysis for this comparison.

Comparison 2: Facility-based strategies to screen and manage uncomplicated SAM according to the WHO protocol compared with other standards of care

Three studies (Ashworth et al., 1994; Chapko et al., 1994; Puett et al., 2012) assessed facility-based strategies compared with other standards of care which was out-patient and community-based management for uncomplicated SAM while one study Hossain et al. (2009) compared the facility-based management with a locally adapted Institute of Child and Mother Health (ICMH) protocol. We could not pool findings from Hossain et al. 2009 since the comparison group was different from other studies; however, we have reported the findings from this study under each outcome reported.

Findings from Ashworth et al. (1994) and Chapko et al. (1994) should be interpreted with a consideration that these studies were conducted before the current differentiation of complicated and uncomplicated SAM.

Comparison 3: Facility-based strategies to screen and manage uncomplicated SAM according to the WHO protocol compared with other standards of care (in-patient treatment with RUTF compared with F100)

Three studies (Mishra et al., 2019; Thakur et al., 2013; Versloot et al., 2017) assessed in-patient management of SAM with RUTF compared with F100.

Comparison 4: Community-based management of children with uncomplicated SAM as outpatients with RUTF compared with standard diet, FBFs or other locally produced foods

Fourteen studies (Bahwere et al., 2014, 2016, 2017; Bhandari et al., 2016; Ciliberto et al., 2005; Dossou et al., 2003; Hsieh et al., 2015; Irena et al., 2015; Jones et al., 2015; Manary et al., 2004; Oakley et al., 2010; Sandige et al., 2004; Shewade et al., 2013; Sigh et al., 2018) compared community-based management of children with uncomplicated SAM with RUTF versus other foods. Standard milk/peanut butter-based RUTF was compared with non-milk/peanut butter-based RUTF, reduced milk/peanut butter RUTF, F100, energy dense home made food, high oleic RUTF elevated n3 PUFA RUTF.

Primary outcomes

Among primary outcome, included studies reported recovery, weight gain and mortality. None of the included studies reported any of the other primary outcomes including relapse and case fatality rates.

Recovery rate: Pooled study results

Ten studies (Bahwere et al., 2014, 2016, 2017; Bhandari et al., 2016; Hsieh et al., 2015; Irena et al., 2015; Manary et al., 2004; Oakley et al., 2010; Sandige et al., 2004; Shewade et al., 2013) reported recovery rate at 8–16 weeks. There was no evidence of difference on recovery rate when standard RUTF was compared with non-milk/peanut butter-based RUTF (RR: 1.03; 95% CI: 0.99 to 1.08; five studies; 5,743 participants; I ² 50%; moderate-quality outcome; Analysis 4.1); energy dense home prepared food (RR: 1.14; 95% CI 0.95 to 1.36; four studies; 959 participants; I ² 75%; low-quality outcome; Analysis 4.1) or high oleic RUTF (RR: 1.06; 95% CI: 0.85 to 1.31; one study; 141 participants; moderate-quality outcome; Analysis 4.1).

We conducted sensitivity analysis by removing one study (Sandige et al., 2004) judged to be at high risk for sequence generation, allocation concealment and blinding. Findings from sensitivity analysis suggest no evidence of effect on recovery rate when standard RUTF was compared with energy dense home prepared food (RR: 1.23; 95% CI: 0.99, 1.52; three studies; 777 participants; I ² 51%; low-quality outcome; Analysis 4.2)

Weight gain: Pooled study results

Eight studies (Bhandari et al., 2016; Ciliberto et al., 2005; Dossou et al., 2003; Hsieh et al., 2015; Irena et al., 2015; Oakley et al., 2010; Sandige et al., 2004; Sigh et al., 2018) reported weight gain at 8–16 weeks (Figure 4). Standard RUTF may improve weight gain by 0.5 g·kg⁻¹·day⁻¹ when compared with non-milk/peanut butter-based RUTF (MD: 0.5 g·kg⁻¹·day⁻¹; 95% CI: 0.02 to 0.99; three studies; 3069 participants; I ² 80%; low-quality outcome; Analysis 4.3) and by 5.5 g·kg⁻¹·day⁻¹ when compared with F100 (MD: 5.50 g·kg⁻¹·day⁻¹; 95% CI: 2.92 to 8.08; one study; 70 participants; low-quality outcome; Analysis 4.3). There was no evidence of effect on weight gain when standard RUTF was compared with energy dense home prepared food (MD: −0.35 g·kg⁻¹·day⁻¹; 95% CI: −1.52 to 0.82; three studies; 1,925 participants; I² 81%; low-quality outcome; Analysis 4.3) and high oleic RUTF (MD: −0.8 g·kg⁻¹·day⁻¹; 95% CI: −1.74 to 0.14; one study; 141 participants; moderate-quality outcome; Analysis 4.3).

OPEN IN VIEWER

Figure 4

(Analysis 4.3) Forest plot of comparison 4: Community-based management with standard RUTF compared with other foods, outcome: 4.2 Weight gain. CI, confidence interval; RUTF, ready-to-use therapeutic food; SE, standard error

We conducted sensitivity analysis by removing two studies (Ciliberto et al., 2005; Sandige et al., 2004) judged to be at high risk of bias for sequence, generation, allocation concealment and blinding. Findings from sensitivity analysis suggest that there was no evidence of effect on weight gain when standard RUTF was compared with energy dense home prepared food (MD: 0.41 g·kg⁻¹·day⁻¹; 95% CI: −0.16, 0.98; one study; 565 participants; low-quality outcome; Analysis 4.4).

Mortality: Pooled study results

Nine studies (Bahwere et al., 2014, 2016, 2017; Bhandari et al., 2016; Ciliberto et al., 2005; Hsieh et al., 2015; Irena et al., 2015; Jones et al., 2015; Oakley et al., 2010) reported mortality at 8–16 weeks suggesting no evidence of effect on mortality when standard RUTF was compared with non-milk/peanut butter-based RUTF (RR: 0.90; 95% CI: 0.72 to 1.12; five studies; 5,743 participants; I ² 3%; moderate-quality outcome; Analysis 4.5); energy dense home prepared food (RR: 1.87; 95% CI: 0.95 to 3.7; two studies; 1,743 participants; I ² 0%; moderate-quality outcome; Analysis 4.5); high oleic RUTF (RR: 5.07; 95% CI: 0.61 to 42.31; one study; 141 participants; low-quality outcome; Analysis 4.5) and elevated n3 PUFA RUTF (RR: 0.33; 95% CI: 0.04 to 2.94; one study; 40 participants; low-quality outcome; Analysis 4.5).

We conducted sensitivity analysis by removing one study (Ciliberto et al., 2005) judged to be at high risk of bias for sequence generation, allocation concealment and blinding. There was no evidence of effect in mortality when standard RUTF was compared with energy dense home prepared food (RR: 5.09; 95% CI: 0.25, 105.53; one study; 565 participants; Analysis 4.6).

Comparison 5: RUSF for MAM compared with standard diet, or FBF or other locally produced foods

Fourteen studies (Ackatia-Armah et al., 2015; Fabiansen et al., 2017; Karakochuk et al., 2012; LaGrone et al., 2012; Matilsky et al., 2009; Medoua et al., 2015; Nackers et al., 2010; Nikièma et al., 2014; Phuka et al., 2009; Scherbaum et al., 2015; Singh et al., 2010; Stobaugh et al., 2016; Thakwalakwa et al., 2010; Vanelli et al., 2014) compared RUSF for MAM with other foods. Other foods included whey RUSF, energy dense home prepared food, CSB and food supplement.

Comparison 6: Prophylactic use of antibiotics in children with uncomplicated SAM compared with no antibiotics

Three studies (Berkley et al., 2016; Isanaka et al., 2016; Manary et al., 2012) compared prophylactic use of antibiotics in children with uncomplicated SAM with no antibiotics. The antibiotics used for prophylaxis included co-trimoxazole (Berkley et al., 2016), amoxicillin (Isanaka et al., 2016; Manary et al., 2012) and cefdinir (Manary et al., 2012).

Primary outcomes

Among the primary outcomes, included studies reported recovery, weight gain and mortality. None of the included studies reported relapse and case fatality rates.

Recovery rate: Pooled study results

Two studies reported recovery rate at 12 weeks suggesting that the recovery in the antibiotic group was 6% better than no antibiotic group (RR: 1.06; 95% CI: 1.03, 1.08; two studies; 5,166 participants; high-quality outcome; I ² = 0%; Analysis 6.1; Figure 5).

OPEN IN VIEWER

Figure 5

(Analysis 6.1) Forest plot of comparison 6: Prophylactic antibiotics versus no antibiotics, outcome: 6.1 Recovery

Weight gain: Pooled study results

Two studies reported weight gain at 12 weeks suggesting that prophylactic antibiotic probably improves weight gain compared with no antibiotic by 0.67 g·kg⁻¹·day⁻¹ (MD: 0.67; 95% CI: 0.28, 1.08; two studies; 5,052 participants; I ² = 49%; moderate-quality outcome; Analysis 6.2).

Mortality: Pooled study results

Three studies reported mortality at 12 weeks suggesting that prophylactic antibiotic administration probably reduces mortality by 26% compared to the no antibiotics group (RR: 0.74; 95% CI: 0.55, 0.98; three studies; 6944 participants; I ² = 52%; moderate quality outcome; Analysis 6.3; Figure 6).

OPEN IN VIEWER

Figure 6

(Analysis 6.3) Forest plot of comparison 6: Prophylactic antibiotics versus no antibiotics, outcome: 6.3 Mortality

Comparison 7: Vitamin A supplementation in the management of SAM and MAM with various doses and frequency of administration

Two studies (Donnen et al., 1998; Sattar et al., 2012) compared high-dose vitamin A supplement with low-dose vitamin A supplement.