Importance of coverage and quality for impact of nutrition interventions delivered through an existing health programme in Bangladesh

Abstract Understanding implementation of interventions is critical to illuminate if, how, and why the interventions achieve impact. Alive & Thrive integrated a nutrition intervention into an existing maternal, neonatal, and child health (MNCH) programme in Bangladesh, documenting improvements in women's micronutrient supplement intake and dietary diversity. Here, we examined how well the nutrition intervention was implemented and which elements of implementation explained intervention impact. Survey data were collected in 2015 and 2016 from frontline health workers (FLW) and households in areas randomized to nutrition‐focused MNCH (intensified interpersonal counselling, community mobilization, distribution of free micronutrient supplements, and weight‐gain monitoring) or standard MNCH (antenatal care with standard nutrition counselling). Seven intervention elements were measured: time commitment, training quality, knowledge, coverage, counselling quality, supervision, and incentives. Multiple regression was used to derive difference‐in‐differences (DID) estimates. Using village‐level endline data, path analysis was used to determine which elements most explained intervention impacts. FLWs in both areas were highly committed and well supervised. Coverage was high (>90%) for counselling, supplement provision, and weight‐gain monitoring. Improvements were significantly greater for nutrition‐focused MNCH, versus standard MNCH, for training quality (DID: 2.42 points of 10), knowledge (DID: 1.20 points), delivery coverage (DID: 4.16 points), and counselling quality (DID: 1.60 points). Impact was substantially explained by coverage and delivery quality. In conclusion, integration nutrition intervention into the MNCH programme was feasible and well‐implemented. Although differences in coverage and counselling quality most explained impacts, all intervention elements—particularly FLW training and performance—were likely important to achieving impact.


| INTRODUCTION
Comprehensive programme evaluations should examine not only the impacts produced by an intervention but also the mechanisms and elements of interventions that influence their impacts (Campbell et al., 2000;Damschroder et al., 2009;Fixsen et al., 2005). Understanding the implementation processes of interventions is critical to illuminate if, how, and why the interventions achieve impact. In addition, implementation data are needed to test theories driving the assumptions of the intervention and to accurately interpret findings. Timely use of implementation data can also support informed decision-making for programme adjustments and continuous quality improvements (Duerden & Witt, 2012).
Implementation of interventions consists of several components, among them how interventions are implemented as planned (fidelity or adherence to design), the amount that has been delivered (dosage or quantity), how well different intervention components have been conducted (quality), participation rates (reach), and participant responsiveness (Carroll et al., 2007;Saunders, Evans, & Joshi, 2005). There is strong evidence that effective implementation is associated with better impacts of intervention. A review of more than 500 studies of health programmes reported that high level of implementation, including those components named above, was associated with at least two or three times higher impact in promotion (mental and physical health) as well as prevention programmes (alcohol, tobacco, and substance abuse; Durlak & Dupre, 2008). The review further highlighted the lack of or limited implementation of data in most studies and that the assessment of implementation is a necessity in programme evaluations; implementation data are needed to document what was conducted and how outcome data should be interpreted (Durlak & Dupre, 2008).
There is increasing evidence of the critical role of implementation science in the field of nutrition, particularly on maternal and child health programmes (Leroy & Menon, 2008;Loechl et al., 2009;Menon et al., 2008;Robert et al., 2006;Robert et al., 2007) and on promotion of infant and young child feeding practices (Avula et al., 2013;Kim et al., 2015;Nguyen et al., 2014;Rawat et al., 2013). Despite growing literature, research that links implementation elements and nutrition impact is limited, and more attention to process-oriented research is needed that can shed light on how nutrition interventions can be operationalized effectively to achieve desired outcomes (Leroy & Menon, 2008;Shekar, 2008;Stoltzfus, 2008).

Maternal undernutrition is a significant public health problem in
Bangladesh; more than half of all pregnant women are anaemic (Hyder, Persson, Chowdhury, Lonnerdal, & Ekstrom, 2004), and nearly a quarter of the women of reproductive age are undernourished or underweight (BMI < 18.5 kg/m 2 ; National Institute of Population Research and Training, 2011), mainly due to poor diets and low intake of micronutrient supplements (Arsenault et al., 2013), among other determinants. To address these challenges, the Alive & Thrive initiative, in collaboration with a large national non-governmental organization in Bangladesh (BRAC), integrated a maternal nutrition-focused, multi-component behaviour change intervention into its existing maternal, neonatal, and child health (MNCH) platform. We have previously shown that the intervention successfully improved multiple outcomes such as maternal dietary diversity, micronutrient supplement consumption, and exclusive breastfeeding practice (Nguyen et al., 2017). For this paper, our objectives were to (i) examine how well the integrated nutrition interventions were implemented and (ii) identify which elements of implementation were most important in explaining intervention impacts on three outcomes: consumption of iron and folic acid (IFA) and calcium supplements, and dietary diversity.

| Study context and intervention description
This study is part of an evaluation on the feasibility and impact of integrating intensified maternal nutrition interventions into the existing MNCH programme in Bangladesh. Detailed description of the intervention, the study design, data collection, and main results has been reported elsewhere (Nguyen et al., 2017 • This paper provides insights on where and how maternal nutrition interventions can be integrated with and delivered through routine healthcare services. weight and explain optimal weight-gain patterns, (iv) counsel on adequate rest during pregnancy, (v) promote optimal breastfeeding prac-

| Study design and participants
This study used a cluster-randomized impact design in which 20 rural upazilas (subdistricts) were randomly assigned to either nutrition-  The consumption of IFA and calcium supplements was assessed among recently delivery women by asking women to report how many IFA or calcium tablets they consumed during their last pregnancy.
During monthly visits to women's homes, FLWs recorded the number of IFA and calcium tablets consumed in a mother-baby book as part of the MNCH programme; this book was used to assist women in their recall. Maternal dietary diversity was assessed using a 24-hr dietary recall; women were asked to describe all foods and beverages they consumed the preceding 24 hr; these food items were then grouped into 10 food groups categories following the Women Dietary Diversity Score classification (Fao & Fhi360, 2016). The dietary diversity measure is the number of food groups consumed and thus ranged from 0 to 10.
Intervention implementation was measured using seven elements: FLW's time commitment, quality of training, knowledge, coverage of service delivery, quality of counselling service, supervision, and incentives. FLW's time commitment was measured by a set of items related to the numbers of days worked and conducted home visits in a month, the time they spent for each home visit and discussed maternal and child nutrition, and the number of pregnant women and children under 6 months of age in their catchment areas.
The training indicator was constructed based on items related to FLW's participation in full training and refresher training, time from last refresher training, and topics discussed at the last full training. Total quality of training score was created by adding scores from these items.

FLW's knowledge on maternal nutrition was assessed from
responses to items about the benefits of adequate nutrition, types of food that pregnant women should eat every day, duration and benefit of taking IFA or calcium supplements during pregnancy, and knowledge of the foods that interfere with IFA absorption. Each knowledge item was given a score of 1 (correct) or 0 (incorrect), and the sum was used as the knowledge score.
Coverage of service delivery was measured by the percentage of women who received home visits from FLWs, the number of home visits during pregnancy (score 1 if >4 visits), whether women were provided with IFA and calcium supplements or with any dietary advice, and whether women had weight measured. Each service that women received was given a score of 1, and the sum was used as the coverage of service delivery score.
Quality of counselling service was assessed by the percentage of women who received various messages on pregnancy care including dietary quantity and quality, taking IFA and calcium tablets, gaining weight, taking rest, and avoiding heavy workload. Each message women received was given a score of 1, and the sum was used as the quality of counselling service score.
FLW's perceptions about supervision were measured by a set of 15 items related to supportive feedback and exchange, and help to manage stress and workload. FLWs were asked to indicate their degree of agreement with each statement, using responses on a 5-point Likert scale. A summed scale was constructed for the supervision score.
Only heath volunteers received incentives from BRAC. Incentives were assessed by asking health volunteers whether they ever received incentives from BRAC, if they received incentives in the last month, and for which services they received incentives.
The total scores for each element of implementation were rescaled to the theoretical range of 0 to 10 for comparability across elements. Other FLW characteristics that may influence FLW's performance were examined, including age, duration of time worked with BRAC, and education level.

| Data analysis
Descriptive analyses were used to report the characteristics of the study samples. Means and standard deviations were calculated for continuous variables, and proportions were calculated for categorical variables. To address the first study objective, we derived differencein-difference (DID) estimates for each element of intervention implementation (except incentives) using regression models that estimated the difference in changes over time between the two study groups (Gertler, Martinez, Premand, Rawlings, & Vermeersch, 2011). All models accounted for clustering at the district and subdistrict level using a cluster sandwich estimator (Hayes & Moulton, 2009).
To address the second study objective, we aggregated endline FLW and household data to the village level. To evaluate the assumption for path analysis of no interaction between exposure and mediator (Vanderweele, 2015), we tested if there was an interaction between study group and each implementation element separately.
For each implementation element where there was an interaction, we report the results of the regression models with interaction. For each implementation element where there was no interaction, we used a path model (Kline, 2011) to examine how much of the programme differences in intake of IFA supplements, calcium supplements, and dietary diversity was explained through that element; the indirect effect for each outcome indicator was calculated as the product of the unstandardized regression coefficients for the paths from intervention group to element and from element to outcome.

| Time commitment
Health workers worked full-time for BRAC (25 days per month) and conducted home visits nearly every day (Table 1). Compared with health workers in the standard MNCH areas, those in the nutritionfocused MNCH areas spent more time discussing maternal and child nutrition in each home visit.

| Exposure to and quality of training
The proportions of health workers that received basic training was uniformly high across groups and exceeded 80% (

| Health workers' knowledge on maternal nutrition
Health workers in both areas had good knowledge of maternal nutrition (Table 3). Health workers' knowledge improved from baseline in both areas, but the improvements were significantly greater among health workers in nutrition-focused MNCH areas for several indicators, including the importance of adequate nutrition of pregnant women, knowledge on why women should take IFA and calcium tablets, health risks for pregnant women if lacking iron in the diet, and beverages that decrease iron absorption.

| Coverage of service delivery
The percentage of households visited by health workers was higher in nutrition-focused MNCH than that in standard MNCH areas at endline (97% vs. 88%; Figure 2). Women in the nutrition-focused MNCH areas were also visited at home more frequently during pregnancy (6.2 vs. 4.2 times). At endline, nearly all women in nutritionfocused MNCH areas received free IFA and calcium tablets, which was much higher than those in standard MNCH areas (at 25% for IFA and 30% for calcium). Nearly all women in nutrition-focused MNCH areas were weighed during their last pregnancy and also had weight measured earlier and more frequently compared with those in standard MNCH areas. Note. CI = confidence interval; DID = difference-in-difference; MNCH = maternal, neonatal, and child health. a Difference-in-difference impact estimate between baseline and endline adjusted for clustering effect at district and subdistrict levels.
b Incentive for health volunteer.

| Quality of counselling service
During the household visits, a higher proportion of women in the nutrition-focused MNCH areas received messages on maternal nutrition from health workers, including messages related to dietary quantity and quality, taking IFA and calcium daily, gaining weight, taking rest, and avoiding heavy workload during pregnancy (Table 4). In addition, a higher proportion of husbands in the nutrition-focused MNCH areas received messages on how to support and ensure that their wives get adequate nutrition and enough rest during pregnancy and postpartum (results not shown).

| Perceptions about supervision
Overall perceptions about supervision were high in both groups at both baseline and endline (Table S1). A majority of the health workers felt that they received supportive feedback and enough guidance and support from their supervisors for their daily work. Health workers also reported that their supervisors often or always took their concerns into account when planning activities and brought them to the higher management level if needed, praised them when something was done well, and used times when mistakes were made as opportunities to help improve skills. No differences were observed between the nutrition-focused MNCH and standard MNCH areas at both baseline and endline.

| Incentives for health volunteers
At endline, about half of the health volunteers in the nutrition-focused MNCH areas had ever received incentives from BRAC, and about one-third received incentives in the last month for different activities, which was higher compared with those in the standard MNCH areas (at 10%). There was a positive association between incentives and coverage (β = 1.43 points) and quality of counselling service (β = 1.12 points; results not shown).

| Effect of each implementation element on intake of IFA, calcium, and dietary diversity
All villages in the nutrition-focused MNCH areas showed a large shift towards higher training, knowledge, coverage, and quality of counselling service (Figure 3). Health workers' time commitment and their perceptions about supervision were high with no difference between study groups and, therefore, were not included in the path model.
Incentives were only available for health volunteer data and, thus, were not included.
For IFA consumption, intervention area interacted with the training and knowledge elements; in nutrition-focused MNCH areas, villages with better training and knowledge scores had higher IFA consumption (β = 3.75 and 6.49, respectively); in contrast, no association was observed in standard MNCH areas (Figure 3). Path analysis was conducted for coverage of service delivery and counselling quality because there was no interaction between the intervention area and these elements. The indirect paths, estimated as the products of the regression coefficients for each path, for coverage of service delivery and quality of counselling service explained 52.9% and 44.1%, respectively, of the difference in IFA supplement consumption (Figure 4).
Similar results were found for calcium supplement consumption. Note. CI = confidence interval; DID = difference-in-difference; IFA = iron and folic acid; MNCH = maternal, neonatal, and child health. a Difference-in-difference impact estimate between baseline and endline adjusted for clustering effect at district and subdistrict levels.
For dietary diversity, intervention area interacted with coverage of service delivery, such that no association was seen between coverage score and number of food groups consumed in the standard MNCH areas, but a one-point difference in coverage score was asso-  Note. CI = confidence interval; DID = difference-in-difference; IFA = iron and folic acid; MNCH = maternal, neonatal, and child health; PW = pregnant women. a Difference-in-difference impact estimate between baseline and endline adjusted for clustering effect at district and subdistrict levels. Training is an important element to developing and maintaining health worker competencies for delivering quality services (World Health Organization, 2015). In our study, we observed an association between higher training quality or knowledge scores and higher consumption of IFA and calcium supplements in nutrition-focused MNCH areas, but no association was observed in standard MNCH areas. It is possible that the more intensive and in-depth training contents with interactive training methods in nutrition-focused MNCH areas contributed to these differences. Going beyond general training, in nutrition-focused MNCH areas, trainings have focused on clear and relevant instructions on counselling approach for FLWs, together with budget discussion on practical and feasible family foods, as well as counting and recording micronutrient tablets. All these careful trainings on micro level task has helped FLWs' success in delivering counselling services to mothers and family members and likely helped achieve high consumption among women.   (Schuler, 2015). Providing adequate advice and support to address these constraints and promote consumption of varied diets requires not only good counselling skills but also clear instruction and practical discussion with family members will be useful to explore the factors affecting the delivery of behaviour change messages intended to address maternal undernutrition.

FIGURE 4
Path models for consumption of iron and folic acid supplements for coverage of service delivery and quality of counselling services (Values are coefficient; results are from separate path analytic models using health worker data; indirect effects through coverage of service delivery: 52.9% of total effect; through quality of counselling service: 44.1% of total effect). *p < .05, **p < .01, ***p < .001 A limitation of this paper was the lack of direct observations of FLWs' activities, thus limiting our full understanding of the quality of service delivery during home visits. Also, perceptions of supervision were high overall, which may be due to courtesy bias where employees tend to avoid expressing negative opinions about their supervisors, resulting in overly favourable responses. Because this study was carried out in the context of a well-functioning and robust MNCH platform, the generalizability of intervention effects should be interpreted with caution in other contexts, particularly with weaker health systems.
Our study has several strengths. The rich data at both FLW and household levels enabled a theory-driven approach to study implementation across multiple elements. By analysing the differences between the nutrition-focused MNCH and standard MNCH areas within the context of a cluster-randomized programme evaluation, this paper strengthens the plausibility that improvements in various implementation elements are attributable to the intervention. Additionally, the path analysis allows us to quantify the contribution of different implementation elements in explaining intervention impacts. Finally, the study contributes to the literature on implementation science in nutrition, providing insights on where and how maternal nutrition interventions can be improved. Lesson learned from this intervention, together with training manuals, job aids, and other materials, are being shared with the Bangladesh government and elsewhere in the efforts to integrate and reinforce nutrition interventions delivered through routine healthcare services (Billah et al., 2017).

| CONCLUSIONS
The integration of the maternal nutrition-focused intervention into an existing MNCH programme was feasible and well-implemented.
Although differences in coverage and counselling quality most explained impacts, all intervention elements-particularly FLW training and performance-were likely important to achieving impact.

ACKNOWLEDGMENTS
We gratefully acknowledge data collection by Data Analysis and Technical Assistance, Ltd, Dhaka. We would also like to thank Sadia Shabnam and Nazia Islam for substantial contributions to the programme implementation.

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

CONTRIBUTIONS
PHN contributed in the study design, coordinating data collection, developing research questions and conducting the statistical analysis of data, drafting and revising the manuscript. EAF contributed to the study design, provided guidance to the statistical analysis and inputs for the manuscript, and critically reviewed and revised the manuscript.
TS contributed in developing research questions and interpreting results and provided inputs to the sections on intervention design and implementation. SSK reviewed and provided inputs for data interpretation and participated in drafting and revising the manuscript.
SA contributed in developing research questions, data interpretation, and comments for the manuscript. LMT conducted the statistical analysis of data and prepared tables and figures for the manuscript. ZM and BA provided inputs to the sections on intervention design and implementation and provided comments on the manuscript. PM contributed in the study design, contributed in developing research questions, and critically reviewed and revised the manuscript. All authors read and approved the final submitted manuscript.