Unintended consequences of programmatic changes to infant and young child feeding practices in Bangladesh

Abstract BRAC, an international development organization, implemented a home‐fortification programme from 2014 to 2018 in Bangladesh. This study aimed to understand the unintended consequences of programmatic changes that occurred during the implementation of the programme on the prevalence of good infant and young child feeding (IYCF) practices and other associated factors. We used pooled data from eight cross‐sectional surveys and data from a series of qualitative investigations carried out as part of a mixed‐methods evaluation approach. A total of 6,479 caregivers of children aged 6 to 23 months participated in the surveys. The prevalence of good IYCF practices increased from baseline (42.1%) to midline (45.3%), but it decreased at the endline survey (31.9%). Qualitative investigations identified several reasons for low IYCF practices at the programme level, such as the withdrawal of community health worker (CHW) incentives for promoting IYCF, providing incentives for the home‐fortification of micronutrient powder (MNP) and changing the focus from IYCF promotion to MNP promotion. A multivariable generalized estimating equation model for pooled data revealed that caregivers were 28% (adjusted risk ratio [ARR]: 0.72, 95% CI [0.67, 0.78]) less likely to maintain good IYCF practices during the period when CHWs were not incentivized to promote IYCF compared to the period when CHWs were incentivized to promote it. The prevalence of good IYCF practices decreased from both baseline and midline to the endline survey due to the unintended consequences of the programmatic changes. An integrated intervention strategy to promote the home‐fortification of MNP and IYCF could be helpful to avoid unintended negative consequences of programmatic changes.


| INTRODUCTION
Suboptimal infant and young child feeding (IYCF) practices are a major contributor to undernutrition and related morbidity and mortality in children under the age of 5 years . Nutritional deficiencies also increase the risk of mortality, morbidity and long-term developmental delays (Black et al., 2013). According to the World Health Organization (WHO), exclusive breastfeeding for the first 6 months of life with early initiation, continuation of breastfeeding for 2 years and nutritionally adequate, safe and age-appropriate complementary feeding starting at 6 months are recommended as appropriate IYCF practices (WHO, 2008). Recent analyses have found that, during the first 2 years of life, faltering growth is more severe; therefore, this time period is considered the critical 'window of opportunity' for improving child growth (Victora, de Onis, Hallal, Blossner, & Shrimpton, 2010).
Globally, approximately 35% of children aged 6 months are exclusively breastfed (Egata, Berhane, & Worku, 2013); 60% of children To improve IYCF practices globally, several interventions, such as integrated child development services provided through community-based workers (Chaturvedi, Nakkeeran, Doshi, Patel, & Bhagwat, 2014), peer counselling by mother support groups (Kushwaha et al., 2014) and social and behavioural change communication (Menon et al., 2016), have been implemented. To improve IYCF practices and achieve the sustainable development goals related to child nutrition which are to obtain 'zero hunger, achieve food security and improved nutrition and promote sustainable agriculture' (goal 2), government and nongovernmental organizations (NGOs) in Bangladesh are providing several nutrition-specific interventions such as promoting, protecting and supporting IYCF practices, controlling micronutrient deficiencies, managing moderate and severe acute malnutrition, providing nutrition services in emergencies and engaging in social and behaviour change communication on nutrition and nutrition-sensitive interventions including food safety programmes, deworming programmes, water, sanitation and hygiene programmes at the community level (Plan & Nutrition, 2017). Intervention activities include counselling for women with children on exclusive breastfeeding, complementary foods and other related messages. To promote good IYCF practices, the National Nutrition Service provides training (through a standard training module based on the National Strategy for IYCF) to relevant service providers (Plan & Nutrition, 2017).

Key messages
• The prevalence of good IYCF practices increased from baseline (42.1%) to midline (45.3%), but it decreased in the endline survey (31.9%) due to the unintended consequences of programmatic changes such as withdrawal of community health worker (CHW) incentives for promoting IYCF practices.
• Household visits by CHWs and effective coverage of MNP had a positive association with good IYCF practices.
• When designing and implementing a programme, careful measures should be taken to avoid unintended negative consequences, and the integration of MNP interventions and CHW home visits could achieve the optimal IYCF practices.
implementation of a programme. Findings from 11 MNP programmes undertaken in different geographic settings showed unintended negative consequences for IYCF practices, such as introducing complementary feeding before the child reaches 6 months of age, not purchasing or giving foods of animal sources to children when introducing MNP (the perception that MNP contains foods from animal sources) and the forced feeding of children (Siekmans, Bégin, Situma, & Kupka, 2017).
It is important to understand the unintended consequences of programmatic changes on IYCF practices and the factors associated with good IYCF practices so that the programme can be informed about the measures that should be taken to avoid negative unintended consequences. Previous literature suggests that there might be some unintended consequences (either positive or negative) during the implementation of a programme (Harris & Ogbonna, 2002;Rottenberg, 1991;Uddin, Sarma, Bari, & Koehlmoos, 2013) that should be taken into consideration. In the Maternal, Infant and Young Child Nutrition (MIYCN) programme, we did not suspect negative consequences, but after the endline survey, we found a significant decrease in good IYCF practices. Therefore, we aimed to explore the unintended consequences of the programmatic changes on good IYCF practices among caregivers of 6-to 23-month-old children and the factors associated with good IYCF practices.

| Study design and setting
The study used a multimethod evaluation with both qualitative and quantitative approaches. In quantitative analyses, we used pooled data from eight cross-sectional surveys conducted at the household level, and these surveys were conducted according to a pre-post design as part of the evaluation of BRAC's MIYCN Phase 2 programme in Bangladesh. The surveys were conducted at different time points in the MIYCN programme areas of BRAC during the implementation period ( Figure S1). Because the intervention of the MIYCN pro- The evaluation team had to forgo the midline survey for Phase 3 in consensus with the implementers and collaborators because, after 2 years of implementation, the programme was considered mature, and therefore, an endline survey of the last phase (Phase 3) was done in place of the midline survey. The surveys were conducted in three phases and, to limit the seasonal effect, surveys were conducted at the same time of the year in which the previous surveys had been done in each phase. Moreover, to understand the unintended consequences and other factors associated with good IYCF practices among the study population in a more holistic way, we analysed qualitative findings using the concurrent assessments, process evaluation and operations research conducted throughout the evaluation period.
The concurrent qualitative assessment was conducted from October 2016 to January 2017; operations research was conducted from October 2015 to February 2016, and process evaluation was conducted from March 2016 to June 2017. Process evaluation and operations research were also conducted as part of the main evaluation of the MIYCN programme. For this study, we also used qualitative data from process evaluation and operations research. The evaluation method has been discussed in detail in other papers (Sarma, Uddin, Harbour, & Ahmed, 2016;Sarma et al., 2020a).

| Sample size
We applied two sample-size calculation formulas for the eight cross-sectional surveys. Initially, during programme implementation periods, a total of five cross-sectional surveys (three baseline and two midline surveys) were conducted to ensure the coverage of the programme. For these five surveys, we considered a 50% prevalence of MNP coverage, a precision of ±10% and a design effect of 2. Our estimated minimum sample size was 192 households per district for caregivers of 6-to 59-month-old children for the first five cross-sectional surveys. At the end of the programme, three endline surveys were conducted at three different times (in alignment with the previous surveys) in the programme areas.
Based on the baseline information regarding outcome indicators (including anaemia and good IYCF practices) obtained from the initial five surveys, we estimated sample size considering a 10% effect size of anaemia, 20% effect size of good IYCF practices, 90% power and a design effect of 2. We chose the prevalence of anaemia for the final sample-size calculation since the sample size based on the prevalence of anaemia was the highest. For this article, we used data collected from caregivers of children aged 6 to 23 months who comprised 38.3% (n = 6,479) of the estimated total sample of 6-to 59-month-old children and their caregivers (n = 16,936).

| Sampling procedure
A two-stage sampling procedure was applied for selecting study participants at the household level from the selected districts. In the first stage, systematic random sampling was used to select communities or primary sampling units (PSUs), with an equal selection probability for each community or PSU. Systematic samples of 16 PSUs were drawn from a complete list of the targeted communities or PSUs, which were sorted by district and by subdistrict within the programme districts (the programme districts for each phase were fixed and we conducted baseline, midline and endline surveys in the same districts of a phase) to reach the minimum estimated sample size. This ensured that all the target communities had an equal chance of being selected for the sample. In the second stage, a physical map-segment sample approach was exercised to segment the selected communities or PSUs. At the final stage, the Expanded Programme on Immunization method was applied by spinning a bottle/pen placed in the centre of the segment, counting the households along that route and picking the fifth household. The selection of households depended on the eligibility criteria (caregivers of 6-to 23-month-old children, mothers/caregivers having resided there for at least 12 months before the day of the interview, the child not being physically challenged or ill). An Android-based tablet (smartphone device) using the Open Data Kit software was used for collecting data.

| Outcome measures
The outcome variable in this study was IYCF practices (categorized as good IYCF practices and poor IYCF practices) among caregivers of 6to 23-month-old children. Information on IYCF practices was obtained using a 24-h recall questionnaire, and IYCF practices were assessed for children aged 6 to 23 months. A modified Infant and Child Feeding Index (ICFI) score of 6 was considered to indicate 'good IYCF practices', and any score less than 6 was rated as 'poor IYCF practices'. The ICFI score was estimated based on age-specific breastfeeding, ageappropriate dietary diversity and age-appropriate meal frequency (Guevarra et al., 2014). Details about the ICFI score calculation are presented in Table S1. We also performed a sensitivity analysis using different ICFI score cut-offs (Table S2).

| Covariate measures
The covariates were identified from the review of relevant literature (Agho et al., 2016;Epstein et al., 2019;Gautam et al., 2016;Issaka, Agho, Burns, Page, & Dibley, 2015;Kabir et al., 2012;Kassa et al., 2016;Khan et al., 2017;Khanal et al., 2013;Locks et al., 2017;Locks et al., 2018;Mihrshahi et al., 2010) and included the known ones that might have a potential effect on the outcome. The covariates for this study were household size (two categories: <5 and ≥5 members), child's sex (categorized as male and female), child's age (categorized as 6 to 11 months and 12 to 23 months), caregiver's age (categorized as <25 years and ≥25 years), caregiver's education (categorized as <5 years and ≥5 years), father's age (categorized as <30 years and ≥30 years), father's education (categorized as <5 years and ≥5 years), child morbidity status (any complications, such as diarrhoeal diseases, fever, difficulties or fast breathing with cough, breathing difficulties or fast breathing with blocked or running nose, breathing difficulties or fast breathing with cough and blocked or running nose in the last 14 days categorized as 'yes' and no such complications categorized as 'no'), household visit from a CHW within the last 12 months (categorized as 'yes' and 'no'), caregiver's religion (categorized as 'Other' , including Hindu, Buddhist, Christian and 'Muslim'), effective coverage of MNP (child fed three or more sachets of MNP in the 7 days before the day of interview: categorized as 'yes' if got effective coverage, otherwise 'no') and wealth index (categorized as poor, middle and rich). The wealth index was calculated by using household materials (e.g., materials used for the floor, roof and wall of the house) and household assets (including type of latrine used and sources of drinking water) by performing a principal component analysis (National Institute of Population Research and  Training, ICF International,,, & Mitra and Associates, 2014). Additionally, we considered incentives for SSs (categorized as 'yes' if the SS received incentives for IYCF promotion and 'no' if the SS did not receive incentives). The household size, caregiver's age and father's age were categorized based on the median value (one category for values below the median and another for values at and above the median). We developed a conceptual framework to present the hypothesized association of good IYCF practices with other factors (Figure 1). Among the factors in the conceptual framework, child's sex, caregiver's age, caregiver's religion, father's age and incentivize to SS are directly linked to IYCF practices (Issaka et al., 2015;Kabir et al., 2012;Mukta & Ahmed, 2012). Before being linked to IYCF practices, wealth index and effective coverage of MNP depend on both father's and caregiver's education (Liu et al., 2018;Sarma et al., 2020a). Parental education has a potential role to improve wealth status and educated parents are more likely to feed MNP regularly to their children (Liu et al., 2018;Sarma et al., 2020a). The effective coverage of MNP is also linked to the household size (Sarma et al., 2020b). Household visit from a CHW and morbidity status of children depend on child's age because, usually, a CHW visits the households of younger children and the younger children are more likely to become ill compared to their older counterpart (Kamal, Hasan, & Davey, 2015;Sarma et al., 2020b). Thus, households visit from a CHW and morbidity status of the children are linked and further, both the CHW visit and morbidity status are linked with IYCF practices (Mukta & Ahmed, 2012;Sheikh et al., 2020).

| Data analysis
All statistical analyses were performed using the statistical software package Stata (Version 13). Quantitative data analyses included descriptions of the study population and estimates of different indicators related to IYCF practices and are presented in terms of frequency and percentage with 95% confidence intervals (CIs). We performed a chi-square test to determine the differences in sociodemographic and background characteristics at the baseline, midline and endline surveys. We performed an adjusted Wald test to determine the differences in good IYCF practices at different time points of the survey. A simple generalized estimating equation (GEE) model was applied to determine the statistical associations between good IYCF practices and sociodemographic or other covariates by taking into account the adjustment of the clustering effect. The results are presented as risk ratios (RRs) with 95% CIs. Covariates that were significantly associated with good IYCF practices in the simple GEE model and also conceptually linked with good IYCF practices (even if they were not statistically significant in the simple GEE model) were included in the final multivariable GEE model. Finally, a multivariable GEE model was carried out to explore the factors associated with good IYCF practices.
Cluster-adjusted and weighted analyses were performed by using the 'svy' command. We checked the collinearity among the covariates and found that the variance inflation factor was below 1.50. A p value of <0.05 was considered as statistically significant.

| Data analysis
Data analysis was started at the very beginning of data collection and was performed in two phases: (i) during data collection and (ii) after data collection. All the data collectors and investigators were involved in the data analysis process. We used thematic analysis to triangulate

| Ethical considerations
The full study protocol (including survey questionnaire and consent form) was reviewed by the Institutional Review Board of icddr,b, which consisted of two committees: the Research Review Committee and the Ethical Review Committee. We obtained written informed consent from caregivers before conducting interviews. The interviewers read out the consent form to the respondents and answered any questions that arose, prior to obtaining written consent.

| Sociodemographic and background characteristics
In this study, we included 2,633 child-caregiver dyads at baseline, 1,762 at midline and 2,084 at endline. We selected children aged 6 to 23 months and their caregivers for our quantitative analyses. The sociodemographic and background characteristics of the caregivers, fathers and children aged 6 to 23 months were similar across the baseline, midline and endline surveys. We found that 78.0% of caregivers at baseline, 81.1% of caregivers at midline and 82.3% of caregivers at endline survey had completed 5 or more years of schooling.
Approximately 62% of children at baseline, 64.9% of children at midline and 54.8% of children at endline experienced morbidity within the 14 days before the day of the interview (Table 1). For the qualitative assessment, we conducted 27 IDIs, 10 FGDs with BRAC CHWs and 18 KIIs with BRAC programme personnel. We also performed 38 IDIs with caregivers.

| Inconsistent trend in the prevalence of good IYCF practices
Although the programme was targeted to promote IYCF, the prevalence of good IYCF practices was 42.1% at baseline and increased to 45.3% at midline but decreased significantly (p < 0.001) to 31.9% at endline ( Figure 2). Figure 3 illustrates the prevalence of good IYCF practices before and after the cessation of incentives for promoting IYCF. The incentives for the BRAC SSs were specifically aimed at promoting IYCF and were removed in early 2016, and we found that before they ended, the prevalence of good IYCF practices was 43.3%, it decreased to 31.9% (p < 0.001) when incentives to SSs for IYCF promotion ended. Figure 4 displays  Although CHW visit and effective coverage of MNP increased from baseline to midline and endline, the good IYCF practices decreased from baseline and midline to endline because the programme changed its focus from promoting IYCF to promoting MNP use. Therefore, CHWs increased their visit frequency but concentrated on MNP promotion instead of IYCF promotion. The phasewise prevalence of good IYCF practices is presented in Figure S2. We also present the prevalence of the components of IYCF indicators including continued breastfeeding, age-appropriate dietary diversity and age-appropriate meal frequency in Figure S3. Adjusted risk ratios, 95% confidence intervals and p values were obtained from log-binomial regression model using generalized estimating equation (simple GEE model [TableS3]). When the log-binomial model would not converge, a log-Poisson model was used (multivariable GEE model). At baseline and endline surveys, respectively, effective coverage and CHW's visit within 12 months were insignificant in simple GEE model, but we included in the multivariable GEE model because these were confounders. that were visited by BRAC CHWs were 13% (ARR: 1.13, 95% CI [1.05, 1.21]) more likely to adopt good IYCF practices than households that were not visited by CHWs. However, we found a significant association between good IYCF practices and CHW visits in Model 1 (Baseline) and Model 2 (Midline), and this association was not significant in Model 3 (Endline), indicating that visits had no significant effect on IYCF at endline because, during this period, SSs were not incentivized to promote IYCF. Caregivers of children who received effective coverage of MNP were 21% (ARR: 1.21; 95% CI [1.10, 1.33]) more likely to adopt good IYCF practices than caregivers of children who did not receive effective coverage of MNP; however, caregivers of morbid children were 10% (ARR: 0.90; 95% CI [0.84, 0.95]) less likely to adopt good IYCF practices.

| Factors associated with good IYCF practices
Our qualitative findings also revealed that some unintended consequences (an inconsistent trend in the prevalence of good IYCF prac- Model 4 also revealed that caregivers who had 12-to 23-monthold children were 43% (ARR: 0.57; 95% [CI 0.53, 0.60]) less likely to maintain good IYCF practices than caregivers of 6-to 11-month-old children. Caregivers who were Muslim were 21% (ARR: 0.79; 95% CI [0.73, 0.87]) less likely to maintain good IYCF practices than caregivers of other religions. Although the caregivers in Muslim families adopted fewer good IYCF practices than families of Hindus or another religion, the qualitative investigation revealed that the Hindu communities generally offered the first meal to girls and boys respectively at 7 and 8 months of age in accordance with a Hindu ritual called Onnoprashon/Annaprashana, which marks an infant's first intake of food other than milk. This is evidence of a barrier of good IYCF practices, which is compromised by this age-specific dietary diversity indicator. We further found from Model 4 that fathers and caregivers with five or more years of schooling were more likely to adopt good IYCF practices (ARR: 1.19 for father and ARR: 1.10 for caregivers). We also present the factors associated with the IYCF indicators including, continued breastfeeding, age-appropriate dietary diversity and age-appropriate meal frequency in Table S4.
Apart from these factors (child's sex, child's age, caregiver's religion, caregiver's education, father's education, wealth index, morbidity status, CHW visits within the last 12 months and effective coverage of MNP) associated with good IYCF practices, the qualitative findings revealed that the caregiver's attitude influenced complementary feeding practices; economic constraints and lack of appetite were often more important factors in these practices than age-appropriate dietary diversity or age-appropriate meal frequency. Caregivers considered giving foods other than breast milk to their children after a certain age; this practice varied from one child to another. Qualitative findings also revealed that older members (e.g., grandmothers) of the family usually believed complementary foods should be started when mothers were unable to produce sufficient breast milk. The caregivers often did not take into account which foods provided proper nutrition or which foods were appropriate for a minimum acceptable diet, and many of them had little knowledge of what age-appropriate dietary diversity meant. They believed that the children should be offered food that sated their hunger or according to their taste. The findings also revealed how caregiver's perception influenced good IYCF practices, despite the presence of an intervention in a particular programme area. A caregiver (mother) of a 14-month-old child said: I feed my child according to his taste. Now I give hotchpotch, as he likes it. That practice continued for 2-4 months. I fed him rice powder until he was aged 8-9 months. I have been feeding him hotchpotch and, sometimes, rice since the age of nine months.

| DISCUSSION
Household data from different representative parts of Bangladesh where the MIYCN programme of BRAC was implemented were analysed in this study to understand the unintended consequences of programmatic changes in IYCF practices. We found that the prevalence of good IYCF practices at the endline survey was approximately 32%, whereas the nationally representative data for Bangladesh (BDHS 2014) showed that 23% of caregivers maintained IYCF practices, which is still suboptimal. The results of this study indicate a significant decrease in good IYCF practices from baseline-midline to endline. Unintended consequences of programmatic changes were the major factors for this decreasing pattern. The reason might stem from the sudden decrease in incentives for CHWs after the withdrawal of funding under the A&T programme, which was specifically intended to improve IYCF practices. The A&T programme was the intervention programme implemented in Bangladesh from 2009 to 2014 to improve IYCF practices among the caregivers of children under 5 years of age (Menon et al., 2016;Sanghvi et al., 2016).
Performance-based monetary incentives awarded to SSs changed their performance on IYCF messaging and improved IYCF practices (Mukta & Ahmed, 2012). Furthermore, most of the interventions were designed considering the sale and promotion of MNP rather than the improvement of IYCF practices. However, greater emphasis on increasing MNP coverage by SSs may result in a lack of attention to IYCF counselling.
Our pooled data findings show that CHW visits within the last 12 months were a potential positive confounder of good IYCF practices among caregivers. We also found a significant association of CHW visits with good IYCF practices at baseline and midline but a nonsignificant association with the results of the endline survey. This might be because the baseline and midline surveys were carried out when CHWs were incentivized to promote IYCF, but the endline survey was conducted during the period when the incentive was removed. This demotivated the CHWs to provide counselling on and demonstrations of IYCF practices; as a result, optimal IYCF practices were decreased at the endline survey even though the number of CHW visits increased. A study conducted in Bangladesh showed that IYCF counselling by frontline health workers had a significant effect on improving exclusive breastfeeding rates and minimum dietary diversity (Epstein et al., 2019). Another study, in Madagascar, showed that mothers who attended a meeting at which a CHW talked about IYCF were one and half times as likely to feed MNP to their children (Locks et al., 2017), which partially supports the role of CHWs in improving IYCF practices. However, a sudden decrease in incentives for the CHWs after the withdrawal of funding during the A&T Phase 1 programme had a potential influence on CHW visits for IYCF counselling in the programme areas, a practice that was specifically intended to improve IYCF practices.
In the MIYCN programme of BRAC, the SSs emphasized the sale of MNP more than counselling on IYCF because they received incentives for selling MNP, and not for providing counselling on IYCF. It might be that the CHWs prioritized MNP promotion over IYCF promotion based on incentives. It has been reported that when incentives are provided for a particular practice, they can lead to low performance in routine health services among health workers (Griffiths et al., 2011). An intervention that integrates MNP promotion with an increased number of household visits could play a significant role in improving IYCF practices (Locks et al., 2017).
We found that caregivers and fathers with 5 or more years of schooling had a higher probability of adopting good IYCF practices than did those who were illiterate or had less than 5 years of schooling. Similar evidence was found in a study conducted in Nepal (Gautam et al., 2016). Some other studies have revealed that the low level of maternal knowledge in South Asia is associated with inappropriate IYCF practices (Senarath, Agho, et al., 2012;Victor, Baines, Agho, & Dibley, 2014). This may be why educated parents have both adequate knowledge of and greater awareness regarding feeding practices. Our qualitative findings indicate that caregiver's perceptions of complementary feeding practices could also vary among different areas and could thus contribute to inconsistent patterns of IYCF practices across programme areas. A study recently conducted in Bangladesh illustrates that complementary feeding practices are related to mothers' perceptions (Naila et al., 2018).
Poverty is a curse for all development processes, including the early physical, mental and cognitive development of children (Engle & Black, 2008;Park, Fertig, & Allison, 2011). The association between poor IYCF practices and the wealth index is well-established (Khan et al., 2017). In this study, we also found that in terms of wealth index, a rich group of caregivers was more likely to adopt good IYCF practices compared with the poor caregivers. This finding has also been reported in other studies across different countries (Kabir et al., 2012;Victor et al., 2014;Zongrone, Winskell, & Menon, 2012).
This study found that caregivers practicing Hinduism and other religions paid greater attention to their children's feeding practices than did those who were Muslims. We did not find any literature in the Bangladesh context that corroborates this difference. However, there was evidence in Ghana that Muslim and non-Christian mothers had a higher risk of not introducing complementary foods than Christian mothers did (Issaka et al., 2015). Caregivers who had given at least three MNP sachets (effective coverage of MNP) to their children in the week prior to the interview day had a greater likelihood of adopting good IYCF practices. This could be because caregivers who regularly gave MNP to their children were more aware of IYCF practices. Some studies have found that caregivers who fed MNP to their children had improved IYCF practices for their children (Locks et al., 2018;Mirkovic et al., 2016;Siekmans et al., 2017), which supports our findings.
This study has some limitations. First, the selected sample size was a subset of the total estimated sample-size, which reduced our estimation power. Second, we assessed IYCF practices based on recall data, which may lead to recall bias. Last, this study was carried out in only some areas of Bangladesh; therefore, the findings should not be generalized to the whole country. Our qualitative findings also have some limitations. Some of the thematic findings were generated from a single data collection method, and we were unable to triangulate the results using data from other methods. Each of the studies had its own research objective; therefore, extracting IYCF practice-related data was challenging when home-fortification with MNP was the main area of focus. Moreover, some data were collected years after the caregivers actually engaged in these IYCF practices, which is likely to add some respondent bias. In addition, we were unable to explore the service providers experience with the day-to-day challenges of ensuring that caregivers engaged in good IYCF practices; therefore, we missed the opportunity for potential triangulation with their insights.

| CONCLUSION
This study reported that the prevalence of good IYCF practices among caregivers decreased from both baseline and midline to the endline survey due to the unintended consequences of the programmatic changes. When designing and implementing a programme, careful measures should be taken to avoid unintended negative consequences, to ensure the effective implementation of the programme.
Aside from the unintended consequences, the positive association of effective coverage of MNP with good IYCF practices suggests that the integration of MNP into IYCF interventions can achieve an optimal level of good IYCF practices. Moreover, attention should be paid to providing household visits by CHWs when designing such an integrated programme.

ACKNOWLEDGMENTS
We are thankful to all the caregivers and children for participating as well as data collectors, field supervisors for their tireless support, Zillur Rahman for administrative support and Mohammad Ashraful Islam for the managerial role. We would like to acknowledge GAIN for technical support and BRAC for implementing the programme. We would also like to acknowledge Dr. Catherine Harbour for her support in accomplishing this evaluation. icddr,b is grateful to the Governments of Bangladesh, Canada, Sweden and the United Kingdom for providing core/unrestricted support. This research was funded by the Children's Investment Fund Foundation (CIFF; UK), grant number GR-01136. The views, opinions, assumptions or any other information set out in this article are solely those of the authors and should not be attributed to CIFF or any persons connected with CIFF.

CONFLICTS OF INTEREST
All authors declared that they have no conflict of interest.

AUTHOR CONTRIBUTIONS
MT and HS conceptualized the paper. HS is the principal investigator of the evaluation and supervised MT for developing this manuscript.
MT and GK analysed the quantitative data, SKL and MR analysed and interpreted the qualitative results, MT prepared the first draft and TA, HS, MR, SKL and GK critically reviewed and commented on the subsequent and final version of the manuscript.