The cost of improving nutritional outcomes through food-assisted maternal and child health and nutrition programs in Burundi and Guatemala.

Abstract Evidence on the cost‐effectiveness of multisectoral maternal and child health and nutrition programmes is scarce. We conducted a prospective costing study of two food‐assisted maternal and child health and nutrition programmes targeted to pregnant women and children during the first 1,000 days (pregnancy to 2 years). Each was paired with a cluster‐randomized controlled trial to evaluate impact and compare the optimal quantity and composition of food rations (Guatemala, five treatment arms) and their optimal timing and duration (Burundi, three treatment arms). We calculated the total and per beneficiary cost, conducted cost consequence analyses, and estimated the cost savings from extending the programme for 2 years. In Guatemala, the programme model with the lowest cost per percentage point reduction in stunting provided the full‐size family ration with an individual ration of corn–soy blend or micronutrient powder. Reducing family ration size lowered costs but failed to reduce stunting. In Burundi, providing food assistance for the full 1,000 days led to the lowest cost per percentage point reduction in stunting. Reducing the duration of ration eligibility reduced per beneficiary costs but was less effective. A 2‐year extension could have saved 11% per beneficiary in Guatemala and 18% in Burundi. We found that investments in multisectoral nutrition programmes do not scale linearly. Programmes providing smaller rations or rations for shorter durations, although less expensive per beneficiary, may not provide the necessary dose to improve (biological) outcomes. Lastly, delivering effective programmes for longer periods can generate cost savings by dispersing start‐up costs and lengthening peak operating capacity.


| INTRODUCTION
Maternal and child undernutrition are global health problems with farreaching consequences that are increasingly addressed through multisectoral nutrition programmes aimed at improving the immediate and underlying causes of undernutrition Ruel & Alderman, 2013). Evidence on programmatic and policy solutions for reducing undernutrition continues to accumulate. Yet evidence on the cost-effectiveness of nutrition interventions remains limited and is especially scarce for multisectoral nutrition programmes. This limits the ability of donors, implementers, and recipient countries to efficiently allocate limited funds. A typical trade-off faced by decision makers is between serving more beneficiaries with a less intensive programme or fewer beneficiaries with a more comprehensive programme. Yet there is currently little evidence to guide such decisions (Schieber, Gottret, Fleisher, & Leive, 2007).
Most previous cost studies of nutrition programmes have been of nutrition-specific programmes that were focussed on single outcome (e.g., iron status; Bhutta et al., 2013;Horton, Shekar, McDonald, Mahal, & Brooks, 2010). Unlike these programmes, multisectoral nutrition programmes typically incorporate inputs from across sectors and aim to improve a range of nutrition and nutrition-related outcomes for multiple beneficiaries within the household (e.g., mother and child).
This complexity makes assessing cost-effectiveness more challenging.
Additionally, the goal of targeting multiple outcomes and the fact that the impacts across these outcomes cannot be easily expressed using a common metric make calculating cost-effectiveness or cost-benefit ratios challenging.
Herein, we report on the costing results of two large-scale food-assisted maternal and child health and nutrition (FA-MCHN) interventions implemented in Guatemala and Burundi. FA-MCHN interventions are popular multisectoral nutrition interventions (Bonnard, Haggerty, Swindale, & Bergeron, 2002) and a programme of choice for the United States Agency for International Development's Office of Food for Peace in food insecure environments with a high prevalence of undernutrition (Food and Nutrition Technical Assistance II Project, 2010). Yet there is little evidence on their cost or cost-effectiveness (Lentz & Barrett, 2013). This costing study was nested in two rigorous programme evaluations that were designed to assess the optimal quantity and composition of the food rations (Guatemala) and the optimal timing and duration of food rations (Burundi) for reducing stunting and its determinants.
The first objective of this study was to determine the cost per beneficiary of the different treatment arm-based programme models (i.e., the five treatment arms that differed in the quantity and composition of the food rations [Guatemala] and three that differed in the timing and duration of food assistance [Burundi]). For each programme model, we calculated costs if it had been delivered at programme scale (i.e., if it had been delivered to all beneficiaries served by the programme). Second, for each of the different treatment arm-based programme models, we estimated the cost per beneficiary per percentage point reduction in stunting and conducted a cost consequence analysis for a wide range of programme outcomes. Finally, because lengthy start-up and closeout periods were required to successfully deliver these programmes, we determined the extent to which these programmes would have operated at a lower cost per beneficiary if they had been implemented for a longer period and thus been able to disperse start-up costs over more beneficiaries and operate at peak capacity for longer.

| The Prevention of Malnutrition in Children under Two Years of Age Program
Detailed descriptions of the Guatemala and Burundi programmes have been published elsewhere (Heckert, Leroy, Bliznashka, Olney, & Richter, 2018;Leroy, Olney, & Ruel, 2016;Leroy, Sununtnasuk, Heckert, & Olney, 2017). The two programmes, which aimed to prevent undernutrition during the first 1,000 days, included three core components: (a) food assistance, composed of a family and an individual food ration targeted to the mother from pregnancy until the child was 6 months old and then to the child from 6 to 24 months; (b) a health, hygiene, and nutrition behaviour change communication (BCC) strategy that included regular smallgroup lessons and other activities; and (c) activities to strengthen the local health care system and promote its use.
In Guatemala, the Maternal and Child Food Diversification Community Program (Programa Comunitario Materno Infantil de Diversificación Alimentaria; PROCOMIDA) was implemented in the Alta Verapaz Department (north central Guatemala) by Mercy Corps. The programme served nearly 53,000 mother-child pairs over the course of implementation. The standard food assistance package, which was a full-sized family ration (FFR) of rice, beans, and oil paired with an individual ration of corn-soy blend (CSB), was delivered in the catchment area of 95 community health centres. One hundred twenty additional community health centre catchment areas were randomly assigned to one of six

Key messages
• Delivering larger rations (Guatemala) and rations for the full first 1,000 days (pregnancy to age 2 years; Burundi) resulted in the lowest cost per percentage point reduction in stunting.
• Smaller ration sizes and shorter periods of ration eligibility reduced per beneficiary costs, but these less expensive programme models were either not effective or less effective than more expensive ones.
• Extending the programmes for 2 years would have saved between 11% and 18% per beneficiary. Implementing effective programmes for longer periods of time will reduce per beneficiary programme costs. study arms for the impact evaluation (five treatment, one control  (Table 1). In addition to the food rations, all programme beneficiaries received monthly BCC sessions and recipe demonstrations delivered by trained programme staff, and health service strengthening activities were implemented in the health centres in the five treatment arms.
Tubaramure (a Kirundi word meaning "let's help them grow") was implemented in the provinces of Cankuzo and Ruyigi in eastern Burundi by a Catholic Relief Services-led consortium, which also included CARITAS Burundi, Food for the Hungry, and International Medical Corps. Approximately 36,000 mother-child pairs were enrolled in the programme. The standard programme provided food rations (family and individual rations of CSB and oil) from pregnancy until the child was 24 months old and was implemented in 205 collines (smallest administrative division in Burundi). An additional sixty collines were randomly assigned to one of four study arms (three treatment, one control). The T24 arm received the standard programme of food rations from pregnancy until the child was 24 months old (Table 1).
Beneficiaries in the T18 arm received food rations until the child was 18 months and all other benefits until the child was 24 months. Beneficiaries in the TNFP (no food ration during pregnancy) arm received all benefits of the standard programme, except for the food rations during pregnancy. Twice each month in all treatment arms and the standard programme, beneficiary mothers received a BCC session delivered by volunteer leader mothers who were trained monthly by programme staff. All community health centres in the two provinces benefited from health service strengthening activities.

| Estimating costs
We first describe how we calculated the cost of programme activities for the entire programme as implemented (standard programme, plus treatment arms) in each country, the cost of delivering each of the treatment arm-based program models if it were implemented at scale, and the cost per beneficiary (including food). We then explain our approach to assessing "cost-effectiveness" in the context of a programme aimed at improving a range of outcomes. Finally, we describe how we calculated the cost per beneficiary for each programme model under a hypothetical 2-year extension of the programme.

| Cost of programme activities
We used a prospective costing design and calculated costs using the activity-based costing ingredients method (ABC-I; Kaplan & Anderson, 2004;Tan-Torres Edejer et al., 2003). The first step was to identify activity-based cost centres (AB-CCs), which aggregate costs based on programme activities. The AB-CCs were defined so that they were mutually exclusive and exhaustive of all programme activities. A description of programme activities and staff responsibilities was developed based on the programme proposals developed by the implementing non-governmental organizations. For each programme, key programme staff participated in an initial workshop to further elaborate the programme description, and annual interviews thereafter were used to update it. The programme descriptions included nine AB-CCs for PROCOMIDA and eight for Tubaramure (listed in Table 2 and described in Table S1). The first four were related to the delivery of the core programme components. The remaining AB-CCs were the activities to support, monitor, and manage the core activities.
Activities for each year were classified as either start-up (those conducted during the development of the programme that would not need to be repeated to sustain the programme, such as the design of the BCC lessons) or post-start-up activities.
The second step was to collect data on the type (e.g., labour and materials) and quantity of inputs required for each activity from programme documents, workshops, individual interviews, and observations of programme activities. Annual interviews were conducted with individual members of the programme staff. For field activities (e.g., food distribution and delivery of BCC lessons), data on the allocation of labour and materials were collected by direct observation at randomly selected field sites. The implementing non-governmental organizations provided detailed finance information, which was used to determine the cost of each input.
The quantity of inputs required for each activity and their unit costs were entered into spreadsheets. Inputs were classified as either a capital cost (one-time expenses, such as equipment) or recurrent cost (for inputs such as labour and materials). Capital costs were annualized by allocating them across the remaining programme years using a discount rate of 3%. Recurrent costs were directly allocated to the year that they were incurred. We calculated the cost of each activity in each year according the amount of each input required and the cost of the input. To account for indirect costs, we added 15% to the cost of each activity. Because the costing literature provides no clear guidance on which indirect rate to use, we used 15% as an estimated average of allowable indirect costs across programme funders and implementers. Importantly, the choice of the indirect cost rate does not affect the primary results of our study (i.e., the difference in costs between programme models).
The costs of start-up and post-start-up activities were handled separately. The cost of each start-up activity was annualized across the remaining programme years and converted to the base year of 2009 by adjusting for annual inflation using a GDP deflator of 3% (Dhaliwal, Duflo, Glennerster, & Tulloch, 2013;Tan-Torres Edejer et al., 2003). Then, the present value of the annualized start-up activity cost was calculated by dividing the annual discounted cost by the annualization factor and converting it to the year of analysis (2015; Dhaliwal et al., 2013). These values were then equally divided among the years the activity was implemented and each subsequent year.
The start-up activity costs were summed for each AB-CC to calculate the annual AB-CC-specific start-up activity costs. In Guatemala, the full monthly ration was 6 kg of rice, 4 kg of beans, and 1,850 g of fortified vegetable oil, and the reduced was 3 kg of rice, 3 kg of beans, and 925 g of fortified vegetable oil. In Burundi, the monthly family ration was 12 kg of CSB and 1,200 g of fortified vegetable oil.
b Provided to women who were pregnant or had a child between 0 and 5 months old and children 6-23 months.
c Women and children received 4 kg each month. d Women received 30 sachets (20 g), and children received 60 sachets (10 g) each month.
e Women and children received 60 sachets (2 g) each month.
f Women received 6 kg of CSB and 600 g of fortified vegetable oil, and children received 3 kg of CSB and 300 g of fortified vegetable oil each month.
CC were summed to obtain post-start-up activity costs for each AB-CC for each programme year (Phillips & Fiedler, 2007). The total cost of each AB-CC was the sum of its start-up and post-start-up costs.
The total cost of programme activities was the sum of the cost of all AB-CCs. All costs are reported in 2015 U.S. dollars (USD).

| Cost of activities for each treatment arm-based programme model
We calculated the hypothetical programme activity cost if each treatment arm was implemented at programme scale (i.e., if all programme beneficiaries received that treatment). The detailed programme description was used to identify which activities would have been conducted differently, to adjust the quantity of the inputs needed to implement these activities, and to estimate the cost of programme activities for each treatment arm-based programme models (Puett et al., 2013). Because the programmes had target beneficiary numbers and monitored enrolment to reach these goals, we assumed for those treatment arms with lower programme uptake that the corresponding treatment arm-based programme model would have expanded services to reach the target number of beneficiaries and adjust inputs accordingly.

| Cost of programme activities per beneficiary
The total cost of programme activities for each treatment arm-based programme model was divided by the total number of beneficiarymonths to produce the monthly cost per beneficiary. To calculate HECKERT ET AL.

of 16
bs_bs_banner the total number of beneficiary-months, we used data collected by each programme on the number of beneficiaries served each month ( Figure S1). The per beneficiary cost was then multiplied by the average months of programme participation to calculate the total cost of programme activities per beneficiary.

| Cost of food rations and supplements per beneficiary
To accurately calculate the per beneficiary cost of food rations and supplements, which varied in quantity by treatment arm and by who was receiving the individual ration (the mother or the child), the cost of the food rations and supplements was calculated separately from the pro- rations and supplements (calculated separately for whether the primary beneficiary was the woman or the child) was multiplied by the average number of months that the ration was received and summed to determine the cost per beneficiary for the duration of the programme.

| Total cost per beneficiary
The per beneficiary costs of programme activities, food rations, and supplements were summed to calculate the total cost per beneficiary.

| Comparing programme costs and impacts
For estimates of the impact of PROCOMIDA and Tubaramure, we drew on results from the cluster-randomized controlled programme evaluations. We first calculated the cost per beneficiary per percentage point reduction in stunting. Then, to address the shortcomings of using a single cost-effectiveness ratio for a multisectoral nutrition programme with impacts across multiple outcomes at the child, mother, and household level, we used a cost-consequence approach to compare the full scope of programme impacts (Mauskopf, Paul, Grant, & Stergachis, 1998). To do so, we drew on the programme impact pathways that were identified at the outset of the programmes (Olney et al., 2013a;Olney et al., 2013b) and compared them across the study arms.

| Effect of increasing programme duration on cost per beneficiary
To determine the extent to which programme duration affected the cost per beneficiary, we estimated the hypothetical programme costs if the Guatemala and Burundi programmes had operated for two additional years at peak capacity. Two years represent a meaningful increase in programme length (around 33%), while not being so long that programmes would need to replace large capital goods (e.g., vehicles) or face new start-up costs (e.g., to refresh the BCC strategy).
We assumed that during the two additional years, the programme would have implemented the same post-start-up activities at the same

| Programme activity costs
The total cost of PROCOMIDA activities, which included the standard programme in the nonstudy area and all five treatment arms in the study area but did not include the cost of rations and supplements, was 27.7 million USD (Table 2). Distributing the food rations and supplements was the most expensive AB-CC (19.1% of programme activity costs); when combined with the supply and logistics of food commodity and supplement distribution, the delivery of food rations accounted for 30.0% of total programme activity costs ( Figure 1). The second most costly AB-CC was BCC development and execution (17.1%), which required extensive staff time to deliver and employed the largest number of staff members. Management, planning, and administration, which required the staff time from managers and headquarters staff with higher salaries, ranked third in the percentage of total activity cost.
Start-up costs accounted for 17.3% of total programme activity costs; the four AB-CCs with the largest start-up costs were food ration and supplement distribution, development and execution of the BCC strategy, monitoring and evaluation, and systematic information management ( Table 2). The other AB-CCs either incurred no or minimal start-up costs.

| Programme activity costs by treatment arm-based programme model
The least expensive programme model was RFR + CSB at an estimated 26.1 million USD total and 721 USD per beneficiary, because the programme stored and transported smaller quantities of food (Table 3).
The NFR + CSB programme model was the most expensive at 28.4 million USD total and 785 USD per beneficiary; the increased expenses are attributable to lower programme participation, which would have required the programme to expand geographically to reach the same number of beneficiaries (Heckert, Leroy, Bliznashka, Olney, & Richter, 2018). The total cost of programme activities for any of the three programme models that provided the FFR along with CSB, LNS, or MNP was approximately 27.5 million USD total and 760 USD per beneficiary.
Total start-up costs were between 15.0% and 16.2% of programme activity costs and highest in the programme models delivering LNS and MNP.
Underlying the differences in programme activity costs were differences in the costs of seven of the nine AB-CCs (Table 3)

| Cost of food rations and supplements per beneficiary
The total monthly cost of the FFR was 8.68 USD compared with 5.05 USD for the RFR. The monthly costs of the MNP supplement were 2.11 USD for children 6-23 months and 2.67 USD for mothers. The monthly costs of CSB and LNS were 2.38 and 2.64 USD, respectively (neither varied by whether the mother or child was the recipient).
The largest differences in the total cost of the family and individual rations over the course of the programme were driven by the size of the family ration, which was smaller in the RFR + CSB and nonexistent in the NFR + CSB programme models; the cost of the combined family and individual rations for these programme models were 69.05 and 215.57 USD, respectively. The total cost of food rations and supplements was highest in the programme models that included the FFR with either CSB, LNS, or MNP (320.82, 328.23, and 319.14 USD, respectively).

| Total cost per beneficiary
The total combined cost of the family and individual food rations and programme activities was similar in the FFR + CSB, FFR + LNS, and FFR + MNP programme models and ranged from 1,078 USD to

| Comparing costs and impacts of PROCOMIDA
PROCOMIDA reduced stunting by 11.1 and 6.5 percentage points in FFR + CSB and FFR + MNP, respectively, at 24 months of age ( In addition to reducing stunting, PROCOMIDA improved other outcomes along the programme impact pathways. It increased adoption of optimal infant and young child feeding and childcare practices, improved maternal health and nutrition knowledge, and in some treatment arms improved household cleanliness and reduced household hunger. In addition to these positive programme effects, however, PROCOMIDA increased maternal and child anaemia in the FFR + CSB arm and led to increased maternal bodyweight at 24 months postpartum in all three CSB arms. These potentially negative effects were not found in the arms that provided LNS or MNP as the individual ration.

| Programme activity costs
The total cost of Tubaramure programme activities (not including food rations) was 23.6 million USD. Food ration distribution accounted for the largest share (30.4%) of programme activity costs (Table 2).
When combined with the supply and logistics of food commodity    distribution, the delivery of food rations made up 44.0% of total programme activity costs. The next most costly activities were management, planning, and administration (18.8% of total activities) and the development and execution of the BCC strategy (12.9%).
Start-up costs constituted 10.1% of programme activities costs. The development and execution of the BCC strategy and the management, planning, and administration activities incurred the largest share of start-up costs.

| Programme activity costs by treatment arm-based programme model
The T24 treatment arm-based programme model was the most expensive at 23.9 million USD total (or 453 USD per beneficiary) if implemented at scale (Table 5). The TNFP model was only slightly less expensive at 23.7 million USD total and 449 USD per beneficiary.
The T18 model was 22.6 million USD total and 427 USD per beneficiary. At around 10% of total activity costs, start-up costs were similar for all three. The largest differences in costs across treatment armbased programme models were attributable to the food distribution activities; T18 and TNFP stored and transported smaller amounts of food, and food distribution activities ended 6 months earlier in T18.
The differences among programme models for the costs of monitoring and evaluation and the training and supervision of programme staff were relatively small, and in both cases, T18 would have been the least expensive, and T24 would have been the most expensive. The remaining AB-CCs did not differ across programme models.

| Cost of food rations
The monthly cost of the combined individual and family rations for households with a pregnant mother or a child younger than 6 months was 13.08 USD. For a household with a child between 6 and 23 months, the cost was 10.90 USD per month. The total cost of food rations per beneficiary (i.e., for the duration of programme participation) was highest for T24 (314 USD), followed by TNFP (262 USD), and T18 (249 USD).

| Total costs per beneficiary
The total combined cost of programme activities and food rations per beneficiary was highest in T24 (766 USD) and slightly less expensive in TNFP (711 USD) and T18 (676 USD).

| Comparing costs and impacts of Tubaramure
Tubaramure had a significant impact on stunting in the T24 (−7.4 percentage points) and T18 (−5.7 percentage points) arms and a marginally significant impact in TNFP (−4.6 percentage points; Olney et al., 2018). The cost per beneficiary per percentage point reduction in stunting was estimated at 103, 119, and 155 USD, respectively (Table 6). In addition to reducing stunting, Tubaramure had positive impacts on a range of child-, maternal-, and household-level outcomes.
The programme improved child haemoglobin and development, reduced child morbidity, and improved infant and young child feeding practices. In mothers, the programme was found to reduce anaemia, increase dietary diversity, and improve prenatal care seeking.
Tubaramure also improved maternal knowledge in several health and nutrition-related domains, improved household food security and handwashing practices, and decreased the proportion of households not treating their drinking water.

| Changes in the cost per beneficiary when increasing programme duration
Under the 2-year extension scenario, the total cost per beneficiary of delivering the programme decreased by approximately 15% (from between 721 and 785 USD to between 612 and 683 USD) for PROCOMIDA and by approximately 30% (from between 427 and 452 USD to between 300 and 312 USD) for Tubaramure (Table 7).
Once combined with the cost of the food rations and supplements (which would not have changed), the total cost of PROCOMIDA per beneficiary was approximately 11% lower for each of the programme models, and the cost per beneficiary per percentage point reduction in stunting was 87 USD for FFR + CSB and 147 USD for FFR + MNP. For Tubaramure, after including the cost of food, the cost per beneficiary was 18% lower, and the cost per beneficiary per percentage point reduction in stunting was 86, 96, and 125 USD for T24, T18, and TNFP, respectively.

| DISCUSSION
This study, carried out in Guatemala and Burundi, is the first prospec-   b Impact estimates were taken from Leroy, Olney, and Ruel (2018). c Impact estimates were taken from Leroy, Heckert, Cunningham, and Olney (2014). Estimates refer to children 0-23 months old unless otherwise indicated.
found that the cost per beneficiary of delivering a family ration with ready-to-use supplementary foods for 6-to 36-month-olds was 220.40 2010 EUR per month (Puett et al., 2013). Emphasizing the cost per percentage point reduction in stunting may lead to the erroneous conclusion that an investment of 100 to 150 USD per beneficiary will result in a 1 percentage point reduction in stunting that twice that investment would lead to a 2 percentage points reduction, and so on. The returns to such investments, however, are non-linear, such that a given increase in programme spending (say, doubling) will not automatically lead to a doubling in the percentage point improvement in linear growth. This is the case for two distinct reasons. First, the proportion of fixed costs (i.e., those that do not vary with the number of beneficiaries served) relative to total costs generally decreases as the number of programme beneficiaries increases. Second, the association between inputs and biological outcomes (such as linear growth) often follows a sigmoid-shaped curve.
A low programme dose (either delivered for shorter duration, with less intensity, with fewer inputs, or with poorer quality) will not result in improved outcomes, as not all limiting nutrients and conditions have been adequately addressed; only at higher doses are observable effects found. Improving the outcome when it gets closer to its optimum level becomes more difficult and requires proportionally larger doses (and thus higher costs for each additional unit of improvement).
Programme impact results from Guatemala provide evidence of this: when comparing the Guatemala programme models that varied the amount of food households received, only the ones providing the FFR had a measurable effect on linear growth .
This impact could have been the result of the larger food ration (e.g., reduced sharing of micronutrient-fortified food and increased availability of household resources), or the FFR may have incentivized programme participation, which may have led to greater exposure to micronutrient-fortified food, BCC, and health-strengthening activities. This finding is important for policy makers who are often faced with the choice between serving more people with a lighter programme or serving a smaller group of beneficiaries with a more intensive programme. Our findings suggest that below a minimum investment level (in this case, not much lower than the effective programme models), the returns drop to 0.
We intentionally did not calculate the cost per stunting case averted. Stunting is useful for population assessment and impact evaluation, but it is a poor metric of the absolute number of children affected (Leroy & Frongillo, 2019;Perumal, Bassani, & Roth, 2018).
In addition to the lack of a biological or clinical basis for the arbitrary in Burundi) appears paradoxical. The cost savings, however, did not only occur via the redistribution of start-up costs over two additional years. A second mechanism is that the programme operated at peak beneficiary capacity for longer and thus delivered a more cost-efficient programme for longer. The per beneficiary cost of food distributions during the programme roll-out and closeout phases, for instance, was much higher than during peak enrolment as they served fewer people for only slightly lower transportation and staff costs. Interestingly, both programmes operated at peak enrolment for very short periods of time ( Figure S1). In Guatemala, this was because of the time it took to initiate activities across the whole programme areas. In Burundi, this was to ensure that all beneficiaries would reach 24 months by the end of the programme.
Given the multisectoral nature of the programme and thus the multiple programme impacts, we chose to focus the analyses presented in this paper on comparing programme costs to a wide range of programme impacts. Thus, we did not conduct a full costeffectiveness analysis for any specific outcome (i.e., an analysis that assessed the sensitivity of the results to changes in input parameters).
Important strengths of this study are the experimental design of the impact study, the prospective costing approach, and the comparison of programme costs to multiple important programme impacts.
These findings also have implications for the how investments in international assistance programmes, such as those in this study, can be modified to deliver programmes at a lower cost per beneficiary.
These programmes often receive funding for fixed periods of time, during which they build a complex programme but are unable to maintain it at peak capacity for a meaningful length of time. Renewing a programme grant late in the programme cycle would not resolve this problem, as programmes need to scale-down operations well in advance of closure. One approach could be for a 5-year programme to be assessed after 2 years for its potential for impact before transitioning into a 7-or 10-year programme. Although these conclusions are based on two multisectoral nutrition programmes, they may also apply to other international assistance programmes.

| CONCLUSION
This prospective costing study of two FA-MCHN programmes makes several contributions that further the understanding of how to optimally design and plan multisectoral nutrition programmes. To answer the question of whether to allocate a more intense programme to fewer beneficiaries (or vice versa), we found that the returns on investments in multisectoral nutrition programmes do not scale linearly and that a minimum level of investment per beneficiary is required for programmes to be effective. For example, reducing programme inputs-the size of the family ration in Guatemala and the duration of food assistance in Burundi-only marginally decreased programme activity costs (by 13-21% in Guatemala and 7-12% in Burundi), led to a lack of impact on stunting in both cases, and thus did not improve the cost per beneficiary per percentage point reduction in stunting.
Further research is needed to identify the minimum investment required to achieve expected impacts on key outcomes and to better understand how differing contexts may modify this minimum investment. Further research should also explore the dynamics of programme duration and cost per beneficiary to understand how to optimize the duration of programme implementation. This study also provides insights into the cost savings of extending the duration of programme implementation. Using the ABC-I method, we showed that lengthening the period over which the Guatemala and Burundi programmes were delivered reduced the cost per beneficiary by 11% in Guatemala and 18% in Burundi. These findings support increasing the length of investments for the delivery of effective development assistance programmes and should be considered by decision makers who wish to maximize impacts and cost-effectiveness.

Implementation of the Tubaramure programme was led by Catholic
Relief Services in consortium with Food for the Hungry, International Medical Corps, and Caritas Burundi, and implementation of the Rosenvaigue provided administrative support.