Predictors of micronutrient powder (MNP) knowledge, coverage, and consumption during the scale‐up of an integrated infant and young child feeding (IYCF‐MNP) programme in Nepal

Abstract Large‐scale programmes using micronutrient powders (MNPs) may not achieve maximum impact due to limited/inappropriate MNP coverage, consumption, and use. We identify predictors of MNP coverage, maternal knowledge of appropriate use, and child MNP consumption in Nepal. A cross‐sectional survey was conducted in 2,578 mother–child pairs representative of children 6–23 months in two districts that were part of the post‐pilot, scale‐up of an integrated infant and young child feeding‐MNP (IYCF‐MNP) programme. Children aged 6–23 months were expected to receive 60 MNP sachets every 6 months from a female community health volunteer (FCHV) or health centre. Outcomes of interest were MNP coverage (ever received), maternal knowledge of appropriate use (correct response to seven questions), repeat coverage (receipt ≥ twice; among children 12–23 months who had received MNP at least once, n = 1342), and high intake (child consumed ≥75% of last distribution, excluding those with recent receipt/insufficient time to use 75% at recommended one‐sachet‐per‐day dose, n = 1422). Multivariable log‐binomial regression models were used to identify predictors of the four outcomes. Coverage, knowledge of appropriate use, and repeat coverage were 61.3%, 33.5%, and 45.9%, respectively. Among MNP receivers, 97.9% consumed MNP at least once and 38.9% of eligible children consumed ≥75% of last distribution. FCHV IYCF‐MNP counselling was positively associated with knowledge, coverage, repeat coverage, and high intake; health worker counselling with knowledge and coverage indicators; and radio messages with coverage indicators only. FCHV counselling had the strongest association with knowledge, coverage, and high intake. Community‐based counselling may play a vital role in improving coverage and intake in MNP programmes.


| INTRODUCTION
Undernutrition during the first 1,000 days-from conception to 2 years -increases the risk of illness, death, and developmental delays in childhood (Black et al., 2013). Micronutrient deficiencies are particularly prevalent-globally, 43% of children aged 6-59 months in low-and middle-income countries suffer from anaemia, approximately a quarter of which is due to iron deficiency (Petry et al., 2016). The World Health Organization (WHO) recommends the introduction of diverse, solid and semi-solid, and complementary foods at age 6 months because breastmilk is no longer sufficient to meet children's nutritional needs during this critical time period (Brown, Dewey, & Allen, 1998;WHO, 2003). In resource-limited settings, however, micronutrient-rich foods such as animal-source foods are often inaccessible for families, and when they are available, they often are not provided in sufficient quantities to young children (Dewey & Adu-Afarwuah, 2008;Murphy & Allen, 2003). Currently, only one in six children aged 6-23 months in low-and middle-income countries are fed the minimum acceptable diet (defined as minimum meal frequency of solid or semi-solid foods and minimum dietary diversity; UNICEF, 2016).
Many countries are implementing programmes using MNP in communities where complementary feeding practices are suboptimal and micronutrient deficiencies are common. MNPs are often used instead of iron drops/syrups due to research showing MNPs have higher acceptability, fewer side effects, and similar efficacy (Dewey, Yang, & Boy, 2009). From 2011 to 2015, the number of countries implementing programmes with MNP tripled from 22 to 65, with over 10 million children aged 6-59 months receiving MNP in 2015 (UNICEF, 2017). Of the 65 countries implementing programmes with MNP in 2015, 25 had initiated national or subnational programmes (UNICEF, 2017). Most programmes distribute MNP free of charge through the health sector, usually as part of ongoing infant and young child feeding (IYCF) programmes . The free health sector MNP programmes, which usually distribute MNP through routine health facility visits, health facility visits and community outreach, or biannual child health days, have coverage rates ranging from 32% to 83% based on each programme's definition of coverage (Jefferds et al., 2015;Korenromp et al., 2016;Reerink et al., 2017), and intake adherence (defined by each programme) of 35-88% (Jefferds et al., 2015;Korenromp et al., 2016;Reerink et al., 2017;World Vision, 2005).
Notably, the majority of evaluations reporting coverage and intake data have come from pilot interventions in small areas with more resources than would be available during national or subnational implementation (Jefferds et al., 2015;Korenromp et al., 2016;Reerink et al., 2017;World Vision, 2005). A recent MNP expert consultative group concluded that implementation research from programmes operating at scale (defined as reaching a large population with long-term delivery infrastructure) is urgently needed to inform MNP programmes globally .
In this paper, we analyse household survey data representative of children aged 6-23 years and their mothers in two districts that were part of the post-pilot, scale-up of the integrated IYCF-MNP programme in Nepal. We identify demographic characteristics, indicators of programme exposure (such as exposure to mass media, health worker, and/or community health worker counselling), and maternal and child experiences with MNP that predict key MNP programme indicators of coverage, repeat coverage, knowledge of appropriate use, and adherence.
2 | METHODS 2.1 | Study population and data collection MNP has been locally branded in Nepal as Baal Vita (translated as "Vitamins for Children"); each sachet contains 15 micronutrients including

Key messages
• Large-scale programmes distributing MNP may have less impact in reducing anaemia and iron deficiency than efficacy trials due to reduced coverage and adherence; however, few large-scale programmes have representative indicators of programme implementation.
• Collecting quantitative programme implementation data in a sample that is representative of a large programme area can allow programme implementers and evaluators to assess which programme indicators (i.e., coverage, knowledge, and adherence) are successfully being achieved and also enable them to identify sociodemographic and programme-related variables that can predict these outcomes. In Nepal, hearing MNP radio messages and health worker and FCHV counselling were independently associated with MNP coverage; however, FCHV counselling had the strongest association with coverage, maternal knowledge of appropriate use, and high child intake of MNP.
• Community-based counselling may play a key role in improving coverage and intake in MNP programmes globally; future research should address how to utilize community-health workers without over-burdening their workload. Details on programme implementation from the Nepal scale-up of the integrated IYCF-MNP programme have been published (Locks et al., 2018). In brief, a cascade training approach using the UNICEF community-based IYCF training tools (UNICEF, 2012), with additional modules on MNP, adapted for the Nepali context. Health workers and FCHVs were trained on key components of IYCF such as the promotion of exclusive breastfeeding for the first 6 months, timely introduction of complementary foods, and the importance of diverse and frequent complementary feeding, as well as on the appropriate distribution, storage, and use of MNP. Both health workers and FCHVs were expected to counsel caregivers on IYCF and MNP in order to utilize multiple contact points with consistent messages (our study found that over three quarters of mothers had interacted with both their FCHV and health worker about the health of her child). Health workers were expected to counsel mothers during routine health visits, and FCHVs were to hold monthly mothers group meetings, where they would counsel mothers using the IYCF-MNP counselling cards and flipcharts. In accordance with the national FCHV programme, FCHVs, who volunteer their time, were also encouraged to conduct home visits or individual counselling if/when they had the time to do so. The current WHO guidelines recommend three boxes (90 sachets) of MNP over a 6-month period (WHO, 2016); however, the Nepal IYCF-MNP programme, which was rolled out before the newest WHO guidelines, recommended that mothers receive 60 sachets of MNP every 6 months from an FCHV or health worker (whomever she preferred/interacted with first); mothers were then instructed to feed their child one sachet/day, with an expected 4-month gap after finishing the 60 sachets before the next distribution.
As part of the programme scale-up, MNP were first distributed by FCHVs and health workers in Achham and Kapilvastu in March 2013; however, subsequent distributions were interrupted due to issues with the international supplier which resulted in a temporary national stockout. Achham and Kapilvastu did not have the expected second distribution of 60 sachets in 2013. A partial distribution using MNP from the emergency stock (noncustomized general brand) occurred in January 2014 with a Baal Vita sticker (similar to that of customized brand) on the box; the IYCF components of the programme continued during MNP stock disruptions. New customized Baal Vita arrived in country in March 2014, and refresher trainings and a relaunch of the integrated IYCF-MNP intervention were carried out. Also in 2014, another project started also supporting general IYCF practices in Kapilvastu and Achham, but the programme did not distribute, provide training on, or promote MNP (Cunningham et al., 2017). Full MNP distributions relaunched in Kapilvastu in May 2014 and Achham in June 2014, providing sufficient time for the oldest children in the survey (aged 18-23 months) to complete all three expected distribution cycles (60 sachets every 6 months) before the end line survey.
The district-representative survey was conducted among children aged 6-23 months and their mothers in Kapilvastu and Achham in January-February 2016. Population proportional to size sampling was used to select 40 clusters from each district. A census of young children in selected clusters was conducted to identify all children aged 6-23 months; random sampling was used to select 34 children from each cluster in Kapilvastu and 33 in Achham. There was no replacement for refusals or clusters with less than the needed number of children.
Trained interviewers used a structured questionnaire, the majority of questions had pretested during the pilot intervention. Mothers were asked about demographic characteristics and experiences with the integrated IYCF-MNP programme. Specifically, mothers were asked how often they sought help/counselling for their child from an FCHV or health worker, whether they discussed child feeding with either provider, the amount of time spent travelling to each (by foot, bike, motorcycle, car, or bus) and satisfaction level with the services (very satisfied, satisfied, not satisfied, or very unsatisfied). If the mother reported discussing child feeding with her FCHV, she was asked whether this occurred individually, in a group, or both. Mothers were also asked if they had heard of Baal Vita (MNP), and if so, where did they heard about it. They were also asked seven knowledge questions on appropri- should be consumed. Mothers were also asked whether they had ever received MNP for the child, and if so, how many times; when, where, and how many sachets they last received at their last distribution; and whether they received a reminder to pick up MNP before their last distribution (and if so from whom). All questions on MNP receipt and intake were based on maternal recall, though notably, a study from the Nepal pilot IYCF-MNP programme found that maternal recall on the number of sachets consumed and the number of sachets observed were highly correlated (Ng'eno et al., 2017). Mothers who had never received MNP were asked why not; mothers who had received MNP were asked if they had ever fed their child MNP. Mothers who had fed their child MNP were asked whether they identified changes in the colour, taste, or smell of the food, and if so, whether those changes bothered the child. They were also asked how many sachets the child had consumed from the last batch, whether the child liked food with MNP, and whether they noticed any positive or negative changes in their child since initiating MNP.

| Ethics
Ethical approval was obtained from the Nepal Health Research Council (NHRC). Survey enumerators described the purpose, procedures, risks, and benefits of the study and allowed mothers to ask questions before inviting them to participate in the survey. All mothers provided written informed consent to enrol themselves and their children in the survey; if mothers were illiterate, a witness signature was obtained.

| Data analysis
Descriptive statistics are presented as frequencies and percentages; the average for the two districts was weighted based on district population.
Binary variables were created for four outcomes of interest. Our selection of programmatic outcomes of interest was guided by the WHO/CDC logic model for micronutrient interventions in public health programmes (De-Regil, Pena-Rosas, Flores-Ayala, & del Socorro Jefferds, 2014). At the household level, improving micronutrient status (in this case through the consumption of MNP) is dependent on the target population using the intervention appropriately, which has two proximate inputs: (a) intervention coverage and (b) a target population that knows, demands, accepts, and has the ability to appropriately use the intervention (De-Regil et al., 2014). A prerequisite for MNP intake is thus access to or presence of MNP sachet stock in the community via FCHV or from health facilities. It has also been proposed that caregiver adherence with MNP recommendations can be conceptualized as containing three key elements: initiation, appropriate use, and continued use of MNP (Tumilowicz, Schnefke, Neufeld, & Pelto, 2017). Our four outcomes of interest were thus MNP coverage (ever received MNP), repeat MNP coverage (receipt of MNP ≥2 times among mothers of children aged 12-23 months who had received MNP at least once), maternal knowledge of appropriate use (based on seven knowledge questions), and high child intake adherence (consumption of ≥75% of last MNP distribution). There is no consensus on how to define intake adherence for MNP programmes . In Nepal, mothers were expected to receive two boxes (60 sachets) of MNP during each distribution; however, many reported receiving only one box (30 sachets) at the last distribution. We defined "high intake" as consumption of ≥75% of the last batch of MNP received (among mothers who received MNP) accounting for when MNP was received; mothers who did not have sufficient time to use ≥75% of their sachets at the recommended dose of one sachet per day were removed from the analyses (i.e., mothers who received 30 sachets within the last 23 days or 60 sachets in the last 45 days). In order to confirm that it was appropriate to collapse mothers who received 60 and 30 sachets into a single group with a cut-off of 75% for high intake, we used a chi-square test to confirm there was not a significant difference in the proportion of mothers who fed their child ≥75% of the MNP received based on whether the mother received one or two boxes at the last distribution. Because almost all mothers in the survey who received MNP initiated use (fed the child MNP at least once), this is not included as an independent indicator.
Chi-squared tests were used to assess whether demographic characteristics were associated with each of the four outcomes of interest. Demographic characteristics considered were district, child's sex, child's age, ethnicity, maternal education, source of household income, household food insecurity based on the Household Food Insecurity Access Scale (Coates, Swindale, & Bilinsky, 2007), and household asset tertile (developed from a principle component analysis based on household ownership of electricity, radio, television, mobile, refrigerator, table, chair, bed, sofa, watch, computer, fan, traditional grain miller, and bicycle). All demographic characteristics associated (p < 0.20) with each outcome in chi-squared tests were included in all multivariable models for that outcome. Multivariable generalized estimating equations (GEE) using a log link, binomial distribution, and exchangeable correlation structure to account for clustering were used to estimate prevalence ratios for each outcome. This model was selected over logistic regression because logistic regression odds ratios would have overestimated the prevalence ratio for the association between exposures and outcomes given the high prevalence of many of our outcomes of interest (Spiegelman & Hertzmark, 2005).
When the log-binomial model did not converge (as indicated in the footnotes in Table 4), a Poisson distribution was used because it gives a consistent estimate of the relative risk (Zou, 2004).
Multivariable log-binomial and log-Poisson models with exchangeable correlation structures were also used to estimate prevalence ratios for each outcome of interest with programme exposure indicators as exposures. Programme exposure indicators included amount of time mother spent travelling to her health centre and FCHV, exposure to sources of MNP information (radio messages, health worker counselling, and FCHV counselling), the type of FCHV IYCF-MNP counselling (individual, group, both, or neither), whether the mother received a reminder to pick-up the last distribution from an FCHV or health worker, and specific FCHV and health centre indicators such as frequency of interactions and maternal satisfaction with the services provided (due to the low prevalence of mothers who were unsatisfied with their FCHV, we compared "very satisfied" versus "satisfied, unsatisfied, or very unsatisfied"). We also assessed whether maternal experiences with and perceptions of MNP were predictors of mothers receiving MNP a second time (repeat coverage) and high child intake of MNP.
For models with programme exposure indicators or maternal perceptions as exposures, the same demographic characteristics associated with each outcome (p < 0.20) in chi-squared tests were included in each model. Determination of which additional programme indicators were included as covariates in the multivariable models were determined a priori based on directed acyclic graphs of how programme indicators were hypothesized to influence each other (see Table S1). All analyses were conducted in SAS 9.4 (Cary, NC).

| RESULTS
A total of 2,578 mother-child pairs participated in the survey, representing 96% of those invited. Approximately half of the children were male, and one third fell in each age category (6-11, 12-17, and 18-23 months; Table 1). Three quarters of mothers had sought help/counselling for the child aged 6-23 months from both a health centre and an FCHV, and only 2% of mothers had never interacted with either.
Approximately half of mothers of children 6-23 months in Kapilvastu and three-quarters in Achham had received MNP at least once; however, only one-quarter and one-half of mothers of children aged 12-23 months in each district respectively had received MNP    When asked what MNP are, mothers were considered correct if they indicated that it was a sachet of vitamins and minerals or that it was something that should be added to a child's food.
c Identified benefits of MNP include improvements in appetite, energy, mental development, growth, immunity, health ,or strength.   Adjusted prevalence ratios (Adj. PR) and corresponding 95% confidence intervals (95% CI) and p values were obtained from generalized estimating equations with the log link and binomial distribution accounting for correlated errors using an exchangeable correlation structure. In order to develop parsimonious models, only variables that were significant at the p > 0.2 level in the chi-squared tests were included in the multivariable models.
g Household food insecurity levels defined based on Household Food Insecurity Access Scale (Coates et al., 2007   Adjusted prevalence ratios (Adj PR) and corresponding 95% confidence intervals (95% CI) and p values were obtained from generalized estimating equations with the log link and binomial distribution accounting for correlated errors using an exchangeable correlation structure. When the log-binomial model did not converge, a Poisson distribution, which gives a consistent but less efficient estimate (Zou, 2004) was used. Logpoisson models are indicated with an asterisk (*) after the reference term. Multivariable models adjust for all sociodemographic covariates that were significant at the p < 0.2 level in chi-square tests for the outcome of interest. For coverage, this includes district, child's age, ethnicity, maternal education, source of household income, and household food insecurity level. For knowledge: district, child's age, ethnicity, maternal education, source of household income, household food insecurity level, and household asset tertile. For repeat coverage: district, child's age, ethnicity, maternal education and household asset tertile. For high intake: district, child's age, ethnicity, maternal education and household food security level.
f Model also adjusts for the amount of time it takes the mother to travel to her health centre. g Mothers missing data on travel time to their FCHV include: mothers who have never interacted with their FCHV (n = 228), mothers who receive home visits only (n = 449) and other (n = 2) in full sample of children 6-23 months.
h Multivariable models with MNP radio message exposure as a predictor also adjust for whether the mother ever received IYCF-MNP counselling from a FCHV and/or health worker.
i Multivariable models with health worker indicators as predictors also adjust for whether the mother received IYCF-MNP counselling from a FCHV and whether she ever heard an MNP radio message.
j Multivariable models with FCHV indicators as predictors also adjust for whether the mother received IYCF-MNP counselling from a health worker and whether she ever heard an MNP radio message.
k Model for receipt of a reminder from a health worker or FCHV also adjusts for whether the mother received counselling from a health worker or FCHV, and whether she heard an MNP radio message. l Only among mothers who had received MNP counselling from an FCHV (n = 1,312 in full sample of children 6-23 months). m Only among mothers who had received MNP counselling from a HW (n = 913 in full sample of children 6-23 months). were also significantly more likely to receive MNP repeatedly (when age appropriate) and to feed their children ≥75% of the MNP compared with mothers who did not receive IYCF-MNP counselling from an FCHV; however, we did not find a significant difference in prevalence of repeat coverage or high intake when comparing mothers who received individual counselling only and mothers who did not receive FCHV IYCF-MNP counselling. Notably, the reference group (mothers who did not receive IYCF-MNP counselling from an FCHV) changed substantially for the repeat coverage (n = 264) and high intake (n = 279) analyses compared with the knowledge and coverage analyses (n = 1264), because the majority of mothers who did not receive FCHV counselling also did not receive MNP and were thus excluded from these analyses. Among mothers who received IYCF-MNP counselling from their FCHV, we did not find that the usual frequency of mother-FCHV interactions for the child was associated with knowledge, coverage, repeat coverage, or high intake; being "very satisfied" with her FCHV was, however, positively associated with maternal knowledge of appropriate use. We did not find an association between health centre visit frequency or maternal satisfaction with any of the MNP indicators.
Among mothers who had tried feeding their child MNP, their perceptions of MNP were strongly associated with repeat coverage and high intake (

| DISCUSSION
In this survey, that was representative of children aged 6-23 months in two districts in Nepal that were part of a post-pilot, scale-up of an integrated IYCF-MNP programme, we found that approximately two thirds of targeted mothers had received MNP for their child, but only one third of mothers of children aged 12-23 months (who were old enough to receive a second distribution) had received MNP sachets at least two times. The coverage indicators from this survey are lower than those from the Nepal IYCF-MNP pilot programme (Jefferds et al., 2015), potentially reflecting the reduction in resources per district as the programme scaled and implementation was fully integrated into the national health system without additional partner supports. A recent review found that countries implementing free health sector MNP distribution programmes have reported coverage (defined by each intervention) ranging from 32% to 83% (Jefferds et al., 2015;Korenromp et al., 2016;Reerink et al., 2017) It has been proposed that >70% should be considered the coverage target for MNP programmes . This threshold is lower than the 90% coverage target for biannual vitamin A supplementation due to the greater behaviour change required by regular MNP consumption but also estimated to be sufficient to achieve nutritional impact . Our findings emphasize the need to strengthen programme coverage in Nepal, and likely in other countries, from both the supply and demand sides. Notably, we observed important disparities in our programme indicators by maternal education, ethnicity, household food insecurity, and distance mothers lived from their FCHV. These findings highlight the importance of investing in programme equity, particularly as IYCF-MNP programmes scale and become part of national and subnational programmes, to ensure that all mothers, including those of lower education levels and those from food insecure households have access to MNP information, reminders, and counselling. Our findings also highlight that simplified, targeted messages for vulnerable community members may also contribute to greater MNP knowledge and appropriate use (Monterrosa et al., 2013;Sanghvi, Jimerson, Hajeebhoy, Zewale, & Nguyen, 2013). Notably, among the mothers who received MNP, over 80% had received it in the last 6 months (as recommended); however, the majority received only one box instead of the recommended two. In addition, among mothers who reported never getting MNP for their child, one in five said it was due to stock out, which further limits MNP coverage and may also undermine MNP demand. Thus, investments in IYCF-MNP behaviour change communication should be complemented with investments in the supply chain to ensure adequate MNP stock at the facility and community level, as well as capacity building for frontline workers to ensure they understand the recommended distribution quantity and how to trouble-shoot supply chain challenges.
Once mothers receive MNP, intake adherence can be defined as the initiation, appropriate use, and continued use of MNP . In the Nepal IYCF-MNP programme, 97.9% of mothers who received MNP initiated use and the majority of all mothers were able to correctly answer each MNP appropriate use question. However, only one third of mothers were able to answer all seven appropriateuse questions correctly, and only 38.9% of children whose mothers received MNP were classified as having "high intake." Interestingly, we found that FCHV group counselling or group-plus-individual counselling were both significantly associated with repeat coverage and high intake of MNP, but individual FCHV counselling was not.
We also did not find an association between hearing the MNP radio messages or receiving health worker IYCF counselling and high intake.
The Stages of Change theory (Prochaska et al., 1994) highlights that mass media may be important in the early phases of behaviour change; however skills building, social support, and continued positive reinforcement are essential for maintaining changes in health behaviour.
In Nepal, hearing the MNP radio messages likely served as a cue to action to remind mothers to pick-up MNP, as demonstrated by the association between hearing the radio messages and coverage indicators; however, the brevity of the radio messages was likely insufficient to support mothers in establishing and maintaining a daily routine for feeding their child MNP. A recent systematic review of nutrition Repeat MNP coverage defined as mothers of children 12-23 months who received MNP at least two times (among mothers who had received MNP at least once). Mothers of infants 6-11 months were removed from analyses because they were only eligible to receive MNP once. Analyses are conducted only among mothers who had tried feeding their child MNP.
b High intake of MNP is defined as consumption of ≥75% of sachets received from last distribution. Analysis excludes mothers who did not receive MNP and also mothers who received MNP recently and did not have sufficient time to feed the child one sachet per day and still use 75% of sachets (i.e., those who received 30 sachets within the last 23 days and those who received 60 sachets within the last 45 days). Analysis is conducted only among mothers who had tried feeding their child MNP. c Adjusted prevalence ratios (Adj PR) and corresponding 95% confidence intervals (95% CI) and p values were obtained from generalized estimating equations with the log link and binomial distribution accounting for correlated errors using an exchangeable correlation structure. Multivariable models adjust for all sociodemographic covariates that were significant at the p < 0.2 level in chi-square tests for the outcome of interest. For repeat coverage, this includes district, child's age, ethnicity, maternal education and household asset tertile; for high intake: district, child's age, ethnicity, maternal education and household food security level. All models also adjust for whether the mother ever heard an MNP radio message and whether she received IYCF-MNP counselling from an FCHV or health worker. d Positive effects of MNP include improvements in health, growth, immunity, appetite, energy/activity, or mental development.
e Negative effects of MNP include black or loose stool, constipation, vomiting, or nausea. education and mass media interventions found that of the 18 studies identified, only three directly assessed the impact of mass media alone, and none showed significant changes in behaviour; however, when mass media was combined with individual or group education, several interventions were able to improve IYCF practices (Graziose, Downs, O'Brien, & Fanzo, 2018). Other studies have also documented the importance of group counselling for improving IYCF practices generally (Flax et al., 2014;Nguyen et al., 2014;White et al., 2016) (Locks et al., 2017;Olney et al., 2011).
The reiteration of key messages through facility-based health workers can complement community-based counselling; however, our findings highlight that community health worker counselling may be especially important for supporting a deep understanding of how to use MNP and for supporting sustained use.
If MNPs are well-received by the community, they could also contribute to community recognition of the FCHV and ultimately improve motivation; a recent systematic review of CHW performance found that CHWs who provided curative services or physical goods reported greater recognition from their community and ultimately more motivation (Kok et al., 2015). In the Nepal pilot programme, however, 15 months into programme implementation, 57% and 44% of FCHVs in the FCHV-delivery districts and health facility-delivery districts, respectively, reported needing more support or disliking the added work of MNP (Jefferds et al., 2015). We did not assess FCHV motivation or capacity in this evaluation, but in a separate analysis from this survey population, we did find that mothers who received MNP from their FCHV were more likely to report being "very satisfied" with the performance of their FCHV (Locks et al., 2018), potentially due to improved motivation and quality of counselling. Future research and monitoring at the CHW level are essential to understand how the introduction of MNP affects CHW motivation and performance, as well as how this changes with the provision of adequate supervision and support, and also over time as programmes mature (Vossenaar et al., 2017).
In this analysis, we found that almost half of mothers who fed their child MNP reported that their child disliked the food with MNP due to changes in the taste, smell, or colour of food. We also found that maternal perceptions of a lack of organoleptic (taste, smell, and colour) changes of food when MNP was added, child's acceptance of foods with MNP, and perceived positive changes in the child due to MNP were strong predictors of high MNP intake. The frequency of maternal report of organoleptic changes may be a sign of inappropriate preparation of foods with MNP by mothers (such as adding MNP to hot foods, soups or other liquids, and/or very small portions of food; Tumilowicz et al., 2017); however, we did not find in this analysis that knowledge of appropriate use was associated with maternal report of organoleptic changes. Other MNP interventions have encountered product quality issues that resulted in organoleptic changes to food (Locks et al., 2017;Schauer et al., 2017).
Taken together, these findings highlight the importance of not only ensuring a high quality MNP product (Schauer et al., 2017) but also incorporating behaviour change strategies that support mothers as they try different strategies to encourage the child to eat foods with MNP, such as mixing with highly flavourful foods such as bananas or curry, making sure the amount of food served to the child is sufficient to thoroughly mix in the MNP, or adding MNP to the child's food without their knowledge (Jefferds et al., 2010). Behaviour change interventions should also reiterate to mothers throughout the programme cycle which negative changes mothers might expect to see in their children due to MNP (such as changes in consistency and colour of stool) but also consistently emphasize the positive changes mothers might see as well, such as the increased strength and appetite reported by many Nepali mothers.
This analysis has several limitations. The cross-sectional design prevents analysis of temporal trends and the observational study design prevents us from ruling out associations are due to residual confounding. However, notably, the analyses adjust for several important sociodemographic characteristics such as maternal education and household food security level. The analysis is also limited to surveys with mothers only and is unable to elucidate FCHV or health worker barriers and opportunities in programme implementation. We are also dependent on maternal recall for questions on coverage and intake, though a study from the Nepal pilot IYCF-MNP programme found that maternal recall on the number of sachets consumed and the number of sachets observed were highly correlated (Ng'eno et al., 2017). This analysis also has several strengths. The survey was conducted in a sample that was representative of all children 6-23 months in Achham and Kapilvastu districts, thus allowing conclusions that are generalizable to a large programme area. In addition, the large sample provided ample power for multivariable models adjusting for several key covariates, and the use of log-binomial as opposed to logistic regression models prevents the overestimation of the prevalence ratios given the relatively common outcomes (Spiegelman & Hertzmark, 2005).

| CONCLUSIONS
Continued resources and support for MNP programmes as they scale to the national and subnational level during post-pilot phase are essential to support programme coverage and the adoption and maintenance of positive behaviour change. In this study, we found that counselling by health workers and FCHVs, and hearing MNP radio messages, were all independently associated with MNP coverage and repeat coverage; however, counselling from an FCHV had the strongest association with maternal MNP knowledge, coverage, and high intake. Counselling by community health workers may play a vital role in improving the coverage of and adherence to MNP programmes in similar settings globally, though future research should address how to best utilize volunteer workers without over-burdening their workload.