What influences child feeding in the Northern Triangle? A mixed‐methods systematic review

Abstract Optimising child feeding behaviours could improve child health in Guatemala, Honduras and El Salvador, where undernutrition rates remain high. However, the design of interventions to improve child feeding behaviours is limited by piecemeal, theoretically underdeveloped evidence on factors that may influence these behaviours. Between July 2018 and January 2020, we systematically searched Cochrane, Medline, EMBASE, Global Health and LILACS databases, grey literature websites and reference lists, for evidence of region‐specific causes of child feeding behaviours and the effectiveness of related interventions and policies. The Behaviour Change Wheel was used as a framework to synthesise and map the resulting literature. We identified 2,905 records and included 68 relevant studies of mixed quality, published between 1964 and 2019. Most (n = 50) were quantitative, 15 were qualitative and three used mixed methods. A total of 39 studies described causes of child feeding behaviour; 29 evaluated interventions or policies. Frequently cited barriers to breastfeeding included mothers' beliefs and perceptions of colostrum and breast milk sufficiency; fears around child illness; and familial and societal pressures, particularly from paternal grandmothers. Child diets were influenced by similar beliefs and mothers' lack of money, time and control over household finances and decisions. Interventions (n = 22) primarily provided foods or supplements with education, resulting in mixed effects on breastfeeding and child diets. Policy evaluations (n = 7) showed positive and null effects on child feeding practices. We conclude that interventions should address context‐specific barriers to optimal feeding behaviours, use behaviour change theory to apply appropriate techniques and evaluate impact using robust research methods.


| INTRODUCTION
Child undernutrition has serious health and economic effects at the individual, household and national levels (Victora et al., 2008). In Northern Triangle countries (Guatemala, Honduras and El Salvador), the cost of undernutrition is around 5 billion US dollars per year (ECLAC, 2008). Around 14% to 47% of children are short for their age in these countries, and, although undernutrition is in decline, the rate of change is slow, uneven and further complicated by rising levels of Improved child feeding practices, including breastfeeding and complementary feeding, improve nutrition outcomes (Victora et al., 2016). However, in the Northern Triangle, current practices are suboptimal. Between one third to one half of children aged under 6 months are exclusively breastfed (PAHO & WHO, 2017), and only half of children aged 6-23 months receive the World Health Organization (WHO)-defined 'minimum acceptable diet' (UNCF, 2018).
Furthermore, it remains unclear how to most effectively improve feeding practices. Globally, interventions have had large but heterogeneous effects on breastfeeding (Sinha et al., 2015) and smaller effects on complementary feeding and nutrition outcomes (Dewey & Adu-Afarwuah, 2008). Within the Northern Triangle, various programmes and policies have been implemented, but little is known about their relative effectiveness.
Previous research on child feeding behaviours in the Northern Triangle has identified determinants such as socioeconomic status (Mock, Franklin, Bertrand, & O'Gara, 1985) and maternal education (Webb, Sellen, Ramakrishnan, & Martorell, 2009). However, this can only guide interventions towards population subgroups, rather than explain the causes of behaviour (O'gara & Kendall, 1985).
Moreover, evidence on determinants, interventions and policies to improve child feeding has typically ignored behaviour change theory (Webb Girard, Waugh, Sawyer, Golding, & Ramakrishnan, 2019).
Theory can help decipher the causes of behaviour and identify appropriate behaviour change techniques (Briscoe & Aboud, 2012).
Theory-based interventions can iteratively inform theory and help to unpack why interventions succeed or fail (Michie, van Stralen, & West, 2011).

One model for understanding behaviour is the 'Behaviour Change
Wheel'-the product of a systematic review and synthesis of previous behaviour change frameworks (Figure 1) (Michie et al., 2011).
At the centre of the framework is the capabilities, opportunities, motivations and behaviour ('COM-B') system, which explains why behaviours are adopted: • 'Capabilities' are the psychological and physical ability to behave in a certain way, for example, problems with latching.
• 'Motivations' are the automatic and reflective brain processes that drive behaviours, like children's tastes and caregivers' desires to have healthy children.
• 'Opportunities' are social and physical external factors that influence behaviour, like social support for breastfeeding or food availability (Michie et al., 2011;Michie, Atkins, & West, 2018).
These determinants can act singularly or in combination to influence behaviour. Circling the COM-B core are categories of interventions and policies that influence capabilities, opportunities, motivations and the behaviour of interest (Michie et al., 2018(Michie et al., , 2011. We systematically review evidence on the capabilities, motivations and opportunities that determine infant and young child feeding F I G U R E 1 Behaviour Change Wheel by Michie, Van Stralen, and West. Notes: Recreated by authors, based on the original (Michie et al., 2011) Key messages • We lack high-quality evidence explaining child feeding behaviours in the Northern Triangle, particularly in El Salvador.
• Existing evidence on child feeding behaviours in the Northern Triangle focuses mainly on mothers' beliefs about the health benefits of breast milk and other foods, the influence of husbands and grandmothers and physical constraints such as time and money.
• Future child feeding interventions in the Northern Triangle should explicitly use and build on existing evidence on context-specific drivers of behaviour and use behaviour change theory at the formative design stages.
(IYCF) behaviours, alongside evidence for the effectiveness of interventions and policies that address these behaviours in the Northern Triangle. Using the Behaviour Change Wheel to map the evidence, we identify gaps in the understanding of child feeding behaviours.

| METHODS
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines (Moher et al., 2009). The study protocol was developed in consultation with librarians from the London School of Hygiene & Tropical Medicine and prospectively registered (Deeney & Harris-fry, 2018).

| Search strategy
Between July 2018 and March 2019, we searched five databases, grey literature websites and reference lists. Databases were Cochrane Database of Systematic Reviews (1999 to present), Medline (1950 to present), EMBASE (1980to present), Global Health (1973 and LILACS (1982 to present Our search string (dollar sign, $, indicating a truncation) was as follows: (breastfeed$ or breast-feed or breastfed or breast-fed or "breast fed" or lacta$ or colostrum or "bottle feed" or bottle-feed or bottlefed or wean$ or feeding or nutrition or "solid food" or "semi-solid food" or "soft food" or diet$ or meal or formula) AND (child$ or infant $ or baby or babies) AND (guatemala or honduras or "el salvador" or "northern triangle").
Equivalent Spanish search terms were applied to LILACS database and grey literature websites.

| Inclusion criteria
We included all relevant quantitative (experimental or observational), Literature was excluded if it did not report original evidence, or only characterised feeding behaviours.

| Data screening and extraction
Search results were imported into EndNote reference management software. Two authors independently screened the results, and disagreements were resolved by discussion. We extracted the following prespecified data: author, publication date, study design, location, population, sample size, exposure (source of behaviour, intervention or policy type), outcome measure, analysis method and key findings.

| Quality assessment
We used an adapted version of the Risk of Bias in Non-Randomised Studies of Interventions (ROBINS-I) tool (Sterne et al., 2016)  or low (<5) quality.

| Analysis
Two reviewers independently mapped exposures onto the Behaviour Change Wheel, and disparities in categorisation were resolved through discussion. As we anticipated high heterogeneity in exposureoutcome pairings, we did not plan to conduct a meta-analysis. Children 6-7 years old were not eligible for our review though sample size by age group was not reported.

| Study characteristics
Database searches returned 1,853 unique results with an additional 15 records from searching websites and reference lists. Of these, 68 were included in our review (Figure 2).
Characteristics of studies on the sources of behaviour are given in Table 1 and on interventions and policies in Table 2.
Studies were published between 1964 and 2019. Most were conducted in Guatemala (68%; n = 46), followed by Honduras (29%; n = 20), with very few in El Salvador (4%; n = 3) (one was in Guatemala and El Salvador). A total of 39 studies explored reasons behind child feeding behaviours; 22 described interventions, of which 10 were randomised trials; and seven were policy evaluations. Sample sizes ranged from three households to 8,332 children. Over half (56%) provided quantitative evidence of associations between at least one exposure and outcome, another 25% included no statistical analysis and 24% were qualitative. There were more studies on breastfeeding (n = 33) than on complementary feeding (n = 23), and 12 studies explored both.

| Quality assessment
Risk of bias in quantitative studies (excluding studies containing only descriptive analyses) is summarised in Table S1. Only four studies were considered to be at low risk of bias, eight were 'moderate' , and 26 were 'serious' or 'critical'. Issues arose predominantly from the risk of confounding, missing data and participants not receiving their assigned intervention.
Quality of qualitative studies is summarised in Table S2. A common issue was a lack of reporting on recruitment strategy, participant-researcher relationship and analysis methods.

| Mapping the evidence
Tables 1 and 2 describe and categorise evidence on the sources (Table 1) and interventions and policies (  Tables S3 (sources of behaviour) and S4 (interventions and policies).

Evidence is also mapped visually against the Behaviour Change
Wheel and coded by data type (Figure 3) and study country (Figure 4).
Many studies identified multiple exposures; each exposure is mapped individually.

| Sources of IYCF behaviours
From 39 studies on the sources of IYCF behaviour, we identified and mapped 98 exposures to all six Behaviour Change Wheel subcategories.

| Reflective and automatic motivations
Reflective motivations were most frequently explored (28 studies).
Studies had wide-ranging publication dates (1979 to 2017) but consistent conclusions.
We found dichotomised views on colostrum-some mothers believed that it 'cleans the stomach', whereas others viewed it as bad or causing illness (Atyeo et al., 2017). Negative beliefs were reported in older and more recent studies (Atyeo et al., 2017;O'gara & Kendall, 1985). Mothers who did not believe that breast milk could transmit negative emotions and illness to children had higher odds of initiating breastfeeding early [odds ratio (OR) = 2.42, 95% confidence interval (CI) = 1.30, 4.57] (Wren et al., 2015).
Studies between 1981 and 2016 consistently linked early termination of breastfeeding to mothers' fears that their milk was insufficient and that children should be trained to accept solid food (Brown et al., 2016;Cohen et al., 1999;Cohen, Haddix, Hurtado, & Dewey, 1995;Nieves et al., 1994;Olney et al., 2012;World Health Organization, 1981). Mothers stopped breastfeeding because of: F I G U R E 3 Evidence on determinants of child feeding behaviours, mapped onto the Behaviour Change Wheel by quality of evidence. Notes: Largest dots = high quality/low risk of bias; medium dots = moderate quality/moderate risk of bias; smallest dots = low quality/serious or critical risk of bias. Green = quantitative evidence (tests associations); purple = quantitative evidence (descriptive only); blue = qualitative evidence beliefs around illness (Olney et al., 2012;Vemury & CARE, 1981), subsequent pregnancies (Brown et al., 2016) and use of contraceptives and medication (O'gara & Kendall, 1985).
Mother's beliefs about food determined the timing and types of foods that they provided. Beliefs around the 'heating' and 'cooling' properties of foods were reported in both older and newer studies.
For example, giving oil or sugar water was believed to 'cool the infant from the birth process' (Izurieta & Larson-Brown, 1995, p. 255) and prepare the child for breast milk (Vemury & CARE, 1981). Liquids such as herbal infusions were used before 24 months to 'heat' the child and protect against illness .
Foods considered cooling (chicken, avocado, fruits, potatoes, beans, eggs and fatty foods) were avoided because they were believed to cause or exacerbate illness (Kincaid et al., 2013;. During illness, some mothers felt that thick foods helped 'plug up the child and stop the diarrhea'; others preferred thin foods 'to replace liquid that is lost' (Parker et al., 1998, pp. 9-10). A belief that malnourished children should be fed less frequently to avoid sickness was also reported (Izurieta & Larson-Brown, 1995). In recent studies, some mothers linked supplements with increased child health; others perceived adverse effects such as diarrhoea (Garcia-Meza, Gonzalez, Tumilowicz, & Solomons, 2017;Gonzalez et al., 2017;Olney et al., 2012).
Beliefs about intrahousehold food distribution, with a pro-male bias, were discussed in two studies in 1979 and 2015. In one study, some mothers felt that high-status items should be reserved for economic earners and greater quantities of these foods were given to boys rather than girls (Pigott & Kolasa, 1979). In another, perceptions that boys were hungrier than girls led to more frequent meals and earlier complementary feeding in boys (Tumilowicz et al., 2015).
Few studies investigated automatic motivations (n = 6), all published in the last decade. Studies mostly described children's tastes for food transfers or supplements. According to some mothers, children's dislike for meat or fortified cereal transfers reduced consumption of those foods (Newman et al., 2014). In one study, consumption of lipid-based nutrient supplements did not differ by flavours provided (Matias et al., 2011); elsewhere, flavour was reported to limit use Gonzalez et al., 2017;Olney et al., 2012).

| Social and physical opportunities
Social relationships were reported to influence child feeding in 16 studies between 1979 and 2017. Whilst husbands, relatives, other mothers, neighbours and health care workers were all identified as influencers, paternal grandmothers were consistently the most frequently discussed. Grandmothers were reported to encourage early introduction of other (non-breast milk) fluids into child diets and dominated feeding decisions (Brown et al., 2016;Cohen et al., 1999;Garcia et al., 2012;Olney et al., 2012;.  (Brown et al., 2016;Chary et al., 2011;Garcia-Meza, Montenegro-Bethancourt, et al., 2017). However, less income was not always associated with poorer feeding behaviours; one study reported that poverty increased colostrum feeding because women could not afford alternative foods for the child (Atyeo et al., 2017).
Access to foods, based on cost, accessibility and seasonality, influenced which foods were fed to children in three studies (Olney et al., 2012;Pigott & Kolasa, 1979;Vemury & CARE, 1981). Studies consistently found that mothers' work within or outside the home restricted their time for breastfeeding McKerracher et al., 2016) and their employment decreased exclusiveness and duration of breastfeeding McKerracher et al., 2016).  , 1985), F I G U R E 4 Evidence on determinants of child feeding behaviours, mapped onto the Behaviour Change Wheel by country of study. Notes: Largest dots = high quality/low risk of bias; medium dots = moderate quality/moderate risk of bias; smallest dots = low quality/serious or critical risk of bias. Red = Guatemala; orange = Honduras; yellow = El Salvador breastfeeding recommendations and techniques for ensuring sufficient milk supply (Cerezo & Claros, 1993;Gutiérrez Cabrera & Turcios España, 2004). One study found that mothers who were given information about breastfeeding by a health care professional were more likely to initiate breastfeeding early, but other studies found no association between knowledge sharing and breastfeeding practices (Atyeo et al., 2017;.

| Psychological and physical capabilities
Knowledge of healthy complementary feeding practices was also low, with confusion about how to use micronutrient supplements and a lack of awareness that malnourished children required more food (Engle & Nieves, 1992;. Although one study reported good levels of knowledge, physical constraints and certain beliefs still prevented optimal practices (Olney et al., 2012).
Six studies, published between 1977 and 2017, described child illness as a barrier to optimal feeding. Illness was associated with lower energy and protein intakes (Martorell et al., 1980;Mata et al., 1977) and lower intakes of nutritional supplements (Matias et al., 2011;Olney et al., 2012).

| Evidence on IYCF interventions
We found 22 intervention studies reporting on 46 exposures, addressing all intervention categories in the Behaviour Change Wheel apart from coercion.
Most (n = 10) employed a combination of enablement and education techniques including food transfers, or provision of high-energy or fortified supplements with behaviour change communication or nutrition counselling.
Studies from the 1990s aimed to determine optimal exclusive breastfeeding duration (Cohen et al., 1994), including for lowbirthweight children , and to understand the effect of exclusive breastfeeding on food acceptance after 6 months (Cohen, Rivera, et al., 1995). Promotion of exclusive breastfeeding for the first 4 months, combined with provision of complementary foods after this (compared with encouraging exclusive breastfeeding for 6 months), was associated with the same or lower breastfeeding frequencies and decreased duration of breastfeeding and breast milk intake at 26 weeks (Cohen et al., 1994;Dewey et al., 1999). Exclusive breastfeeding to 6 months did not affect food acceptance at 9 or 12 months .
More recent interventions used food assistance and education to encourage exclusive breastfeeding to 6 months and improve comple-  (Olney et al., 2013). The final evaluation, based on longitudinal data between 2011 and 2015, showed smaller improvements in early initiation of breastfeeding (5 pp higher), increased exclusive breastfeeding at 6 months (11 pp) and some (mixed) improvements to complementary feeding (Heckert et al., 2018). Other combined interventions improved child diets: nutrition education provided with Plumpy'doz (lipid-based nutrient supplement) increased energy intake (Flax et al., 2015) and education with fortified cereal product versus meat increased dietary diversity and the number of main meals but decreased the number of additional meals (Krebs et al., 2012).
Interventions using enablement techniques alone (five studies) primarily did so by providing food transfers or high-energy nutrient supplements. Given to mothers, high-energy (vs. low-energy) supplements increased exclusive breastfeeding at 20 weeks postpartum (96% vs. 84%, p < 0.04) (Gonzalez-Cossio et al., 1998). When given to children, provision of beans and corn , a highprotein energy drink (Islam & Hoddinott, 2009) and high-energy cookies ) each increased energy intake.
Education alone, and combinations of education, training and social support interventions, had mixed effects on breastfeeding.

| Evidence on IYCF policies
We found evidence from seven policy studies, relating to communications/marketing, guidelines, regulation and service provision.
A national programme in Honduras, Project for the Support of Breastfeeding (PROALMA), aimed to improve breastfeeding practices by educating health care professionals and changing hospital policies, such as eliminating infant formula (Popkin et al., 1991). Evaluations found positive effects on prevalence of breastfeeding at 6 months (15 pp higher between programme and control) (American Public Health Association, 1987) and 12 months (35 pp increase between years 1982 and 1985) (Canahuati, 1990). Although initiation of breastfeeding was expected to decline due to secular changes, an increase of 2 pp was observed between 1981 and 1984, possibly as a result of PROALMA (Popkin et al., 1991). Another national programme, Atención Integral a la Niñez en la Comunidad (AIN-C), led by the Honduran Ministry of Health, delivered growth monitoring and promotion activities, including tailored nutrition counselling. Midterm evaluation showed large effects on child feeding, particularly exclusive breastfeeding until 4 and 6 months (<4 months: 56% vs. 24%; <6 months: 46% vs. 19%) (Van Roekel et al., 2002). Subsequent evaluations showed sustained but smaller effects on breastfeeding practices (Schaetzel et al., 2008;Sierra et al., 2019).
As part of the Guatemalan government's Zero Hunger Pact Plan, a communication strategy was launched using sociodramas, community nutrition lotteries, television and radio programmes, to reduce child undernutrition (Grajeda & Campos, 2016). A qualitative evaluation suggested that child feeding knowledge improved more than practices (Grajeda & Campos, 2016).

| Country-level analysis
Overall, studies from Guatemala were more recently conducted, greater in number and of higher quality than those from Honduras and El Salvador. Themes we identified as sources of behaviour did not differ substantially by country or publication date. We did not conduct further analysis of between-country differences due to the paucity of information on the same exposure-outcome pairings.

| DISCUSSION
Our review maps evidence on the drivers of child feeding behaviours in the Northern Triangle and the effectiveness of related interventions and polices. We find a predominance of studies on mothers' beliefs and perceptions about food and breast milk, the influence of relatives (especially paternal grandmothers) and mothers' lack of resources on breastfeeding and child feeding practices. Interventions mostly provided food rations and high-energy supplements nutritional education.
We found only three evaluated policies. Evidence mapped onto the Behaviour Change Wheel reveals that, although there are pockets of well-explored influences on child feeding, evidence is lacking on automatic reflections and capabilities. Evidence from El Salvador and in urban areas is also scarce. Intervention and policy evaluations were lacking in number, were not linked to the behavioural theory and many were not robustly evaluated.

| Drivers of child feeding behaviours
The number of studies across the COM-B categories appears to reflect the state of evidence in other countries, with comparatively more studies on reflective motivations and opportunities and fewer studies on automatic motivations and physical capabilities. This trend may mirror the relative importance of these categories in driving behaviours or may represent research gaps.
Themes within categories are also common to studies from other countries. For example, perceived milk insufficiency, lack of decisionmaking power and gender bias have been reported to influence feeding practices in other countries (Basu, Aundhakar, & Galgali, 2014;Burns et al., 2016). Similarly, the importance of child willingness to be breastfed, illness and lactation problems has been highlighted elsewhere (Balogun, Dagvadorj, Anigo, Ota, & Sasaki, 2015).
These barriers reveal an important particularity of child feeding behaviours: to feed optimally, two sets of behavioural drivers must converge so that mother and child are willing and able to partake in child feeding (Ventura & Worobey, 2013). Furthermore, the motivations and capabilities of mothers and children may interact. For example, breastfeeding can strengthen the bond between mother and child, in turn acting as a predictor of continued feeding and milk supply (Agunbiade & Ogunleye, 2012).

| Intervention and policies
Dietary behaviour change interventions in the Northern Triangle have typically justified their methods by citing evidence for their effectiveness in other settings, but rarely consider contextual or theoretical reasons for why the desired behaviour is not being carried out or how it is hypothesised to improve. These are common criticisms of behaviour change interventions and are important because they can limit intervention success (Aboud & Singla, 2012). Exceptionally, the formative research conducted for PROCOMIDA (Olney et al., 2012) identified the motivations of mothers, fathers and grandmothers, which may have contributed to its success.
Interventions providing resources may be more likely to improve child diets than breastfeeding. We found that poverty and food insecurity had more negative effects on child diets than breastfeeding, and the interventions in our review that gave mothers food or supplements tended to improve dietary diversity more than breastfeeding.
Consistent with this, international reviews also find that food or cash transfers improve dietary diversity (Bastagli et al., 2016;Hidrobo, Hoddinott, Peterman, Margolies, & Moreira, 2014), but effects on breastfeeding indicators are more mixed (Bassani et al., 2013). This may be because the link between resources and breastfeeding is more complex, as poverty is associated with prolonged feeding (Chary et al., 2011) and the costs of breastfeeding are borne out more in terms of mothers' time and energy than the household food budget.
Particularly for breastfeeding, but also for complementary feeding, the provision of resources may be ineffective or insufficient (Bassani et al., 2013). Our review found that mothers' nutrition knowledge and beliefs guide child feeding behaviours, indicating that education and training are relevant techniques. Although education and counselling interventions in our review had mixed effects, international reviews find that they are highly effective at improving breastfeeding practices (Sinha et al., 2015) and have (smaller) effects on child diets (Webb Girard & Olude, 2012 (Kim et al., 2016). In India, Participatory Learning and Action community groups and home counselling, which engage the wider community, improved child minimum dietary diversity (OR = 1.47) (Nair et al., 2017). Interventions to address these social barriers could be tested alongside techniques to increase resources, knowledge and motivations.
A systematic search of policies addressing malnutrition across Latin America showed that all three Northern Triangle countries have implemented WHO-recommended policies, including growth monitoring and promotion, regulation of the marketing of breast milk substitutes (part of the Baby-Friendly Hospital Initiative) and paid maternity leave. Apart from the AIN-C evaluations, our review demonstrates a lack of evidence on the impacts of these policies despite evidence that they have improved breastfeeding rates in other countries (Mirkovic, Perrine, & Scanlon, 2016;Sinha et al., 2015). None of the Northern Triangle countries have specific policies to address micronutrient deficiencies and there is insufficient resourcing of existing nutrition policies (Tirado et al., 2016), which may also explain the shortage of evaluations found in our review. The lack of monitoring and evaluation has been described as a barrier to progress in the region (Martorell, 2012) and must be a priority moving forwards.

| Strengths and limitations
Our review benefits from a systematic search and duplicate assessment of study inclusion, quality and categorisation of studies onto the Behaviour Change Wheel. By focusing our review on the Northern Triangle, we restrict the generalisability of our findings, but this approach has enabled an in-depth, context-specific analysis of child feeding behaviours and demonstrates a mapping approach that could be applied in other contexts.
By using the COM-B model, we found that categories overlapped and were difficult to disentangle-a challenge that would arise from applying any framework to a complex reality. For example, we found evidence that colostrum is withheld from children because of mothers' beliefs that it may cause illness. This belief could be categorised as mothers' reflections (reflective motivations), cultural norms (social opportunities) and/or a lack of knowledge (psychological capabilities). No studies related their findings to the COM-B model, so our subjective categorisation of evidence is a potential source of bias. To ensure that evidence was mapped consistently, two reviewers independently categorised them, and we assigned multiple categories where necessary.
The choice of the Behaviour Change Wheel over another framework is another potential source of bias. The Behaviour Change Wheel places less emphasis on the stages of behavioural change than other frameworks, such as the transtheoretical model (Prochaska & Di Clemente, 1982). However, no studies in our review described these cognitive stages. Other nutrition-specific frameworks unpack the direct, proximate and distal determinants of undernutrition (Black et al., 2008). However, they do not apply the behavioural theory or enable us to unpack the processes by which these determinants affect individual behaviours. We have confidence in our selection of the COM-B framework because it is based on a systematic review of the behaviour change theory and is therefore very broad (Michie et al., 2011).
Because we were interested in understanding the capabilities, opportunities and motivations that explain behaviours, we excluded evidence on more distal determinants, which would require further assumptions to be made about the specific pathways to influencing individual behaviours. For example, differences in socio-economic status may reflect differences in employment and therefore a mother's time to feed or financial resources to buy appropriate foods. Although distal factors can direct research to population groups, they do not explain the root causes of behaviour.
Given the range of study dates, designs and quality, it is difficult to determine to what extent our results reflect the current drivers of feeding behaviours and their relative importance across population subgroups. However, we do find consistency in results over time, and we highlight evidence gaps where we find them.
Finally, publication bias and reporting bias are possible limitations and are difficult to assess quantitatively due to the lack of studies on each exposure-outcome pairing.
To comprehensively understand the capabilities, opportunities and motivations of mothers and children to change feeding behaviours, we need recent high-quality, mixed-methods research in varied contexts within the Northern Triangle. This, along with evidence reviewed here, should be used with the behavioural theory to inform future interventions. These interventions require robust impact and process evaluations so effective approaches can be scaled up.

ACKNOWLEDGMENTS
Self-funded and author HHF was funded by a Sir Henry Wellcome Fellowship from the Wellcome Trust. Grant number: 210894/Z/18/Z.

CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest.