Duration of programme exposure is associated with improved outcomes in nutrition and health: the case for longer project cycles from intervention experience in rural Nepal

ABSTRACT

Economic growth and poverty reduction are not always sufficient to improve child health and nutritional status. Heifer International promotes livestock introduction and related training for community development and poverty alleviation. These programmes do not directly address child health or nutrition. To determine effects of its activities on these important outcomes, Heifer conducted a 4-year longitudinal investigation in rural Nepal. The intervention was associated with significantly improved child anthropometry (related to the duration of intervention exposure) and child health. Heifer activities represent a viable ‘nutrition sensitive’ intervention, but these impacts take time to manifest and be sustained.

KEYWORDS:

• Undernutrition
• livestock intervention
• anthropometry
• child health
• Nepal
• sanitation

1. Introduction

In low income countries, the connections among poor nutrition, ill health and poverty are well established. However, growing economic prosperity (linked to poverty reduction) and an improved health environment do not always result in enhanced child nutrition. Complex cultural, gender, education, medical and environmental factors may impede gains in nutrition at both individual and population levels, even as economic and health conditions improve (Demment, Young, and Sensenig 2003; Leroy and Frongillo 2007; Masset et al. 2012; Webb Girard et al. 2012; Haddad 2013; Webb and Kennedy 2014). Since undernutrition affects more than 200 million children throughout the world (stunting and wasting combined), and is responsible for almost 50 per cent of preventable child deaths (UNICEF 2014), it is critical for policymakers to better understand the drivers of improvements in child nutrition.

To try to accelerate gains in nutrition, international aid agencies and national governments have recently called for a greater integration of development efforts across sectors (Ruel and Alderman 2013). In practical terms, this means promoting synergies across agriculture, food systems development, health and nutrition, particularly in rural areas where most of the world’s poor reside. That said, it has been challenging to demonstrate that investments in agriculture tie directly to improvements in child nutritional status (Underwood and Smitasiri 1999, 1999; Suwal 2001; Rosegrant and Meijer 2002; Demment, Young, and Sensenig 2003; Haddad 2013). For example, in a recent systematic review (Webb Girard et al. 2012), effects on child nutritional status were ‘not significant’ in the 4 of the 36 investigations that addressed this outcome. The overall conclusion was that ‘evidence for an effect of agricultural strategies on anthropometric indicators among … children is limited’ (Webb Girard et al. 2012). Other works have also highlighted the complexity of this relationship (Leroy and Frongillo 2007; Randolph et al. 2007; Olney et al. 2009; Masset et al. 2012; Dangour et al. 2013a; Haddad 2013), including the methodological difficulties in empirically establishing this association. Thus, it has not yet been convincingly demonstrated that nutrition-sensitive packages of agricultural interventions can have desirable impacts on child growth.

Part of the challenge in uncovering such effects lies in the fact that agricultural interventions often require many years to demonstrate large scale or sustained impacts on productivity, income and food supply that can be linked to improved diets and nutrition outcomes. Similarly, studies conducted shortly after interventions have taken place cannot capture any potential later effects on nutrition (Masset et al. 2012). The importance of long-term follow-up has, therefore, been emphasised (Gibson et al. 2003; Bezner Kerr, Berti, and Shumba 2010; Ruel and Alderman 2013), especially since some studies have demonstrated that intervention effects on nutrition can be observed as long as 16 years after implementation (Kidala, Greiner, and Gebre-Medhin 2000; Bezner Kerr, Berti, and Shumba 2010; Kuman and Quisumbing 2010). However, methodological issues complicate interpretation of results spread out over decades (Webb Girard et al. 2012).

Recently, we reported the outcomes of child growth after 24 months of household participation in a community development and livestock training programme provided by the NGO Heifer International in Nepal (Miller et al. 2014). That study, using a staggered-introduction design, showed that young children (<5 years of age) living in communities with longer duration of participation in Heifer activities had greater incremental improvements in height-for-age z-score (HAZ) and weight-for-age z-score (WAZ) scores than children whose families joined Heifer activities later in the programme. However, despite those gains, the share of children with various forms of undernutrition (underweight or stunting) did not diminish over the initial 24-month period of intervention activities and observation. Moreover, no gains in child health were specifically attributable to participation in the programme, although household health practices and socioeconomic status (SES) improved in the participant families.

We, therefore, hypothesised that extended follow-up (beyond 24 months) might reveal more significant improvements in child nutrition and health. We also considered the possibility that any positive gains in nutrition and health could decline after the initial active period of intervention ended. (Heifer remained involved in the regions, but to a lesser extent than during the initial period of activity.) To address these questions, we conducted an additional survey in this population 48 months after the initiation of the project.

2. Methods

2.1. Setting

Nepal has a high under 5-year mortality, with 34/100 and 42/1000 children dying before their 1st and 5th birthday, respectively (Ministry of Health and Population Nepal, New ERA & Inc. 2012; UNICEF 2014). While national level nutrition measures have improved recently (Headey and Hoddinott 2015), 20–40 per cent of Nepali children are stunted and 11–30 per cent are wasted (respectively, height-for-age and weight-for-height < –2 standard deviation [SD] from median) (Ministry of Health and Population Nepal, New ERA & Inc. 2012; UNICEF 2014) (Klemm 2014).

2.2. Study design

The study was conducted beginning in 2009 in three districts of Nepal: two in the Terai region (Chitwan, Nawalparasi) and one in the Hills (Nuwakot). These comprise different agro-ecological zones, and have different castes and ethnic groups (Devendra and Thomas 2002; Gittelsohn and Vastine 2003). However, both areas are largely populated by low-income subsistence farmers. Communities in both regions were selected for purposes of comparison.

For this study, among the roster of communities which had approached Heifer requesting inputs, three pairs of comparable communities in each district were identified. Factors taken into account included geographic location, altitude, population size, local natural resources, employment opportunities, availability of health care, type of agriculture practiced and other demographic features (predominant castes, family income and educational levels). Paired communities were nonadjacent to minimise spillover effects.

A staggered-introduction design was used to investigate the effects of a livelihoods-based community development intervention (described in more detail later) (Figure 1). One of each pair of matched communities was randomly assigned to receive Heifer development activities immediately after the baseline survey (Group 1); the second of each pair received the intervention starting immediately after the 12 month survey (Group 2). Thus, at the 48 month survey, Group 1 households had received 48 months of programme inputs, while the Group 2 households had received 36 months of these inputs. The staggered-introduction design allowed a comparison of matched communities with or without intervention for the first 12 months; after that the effect of duration of participation in programme activities on selected outcomes could be assessed. The relatively long follow-up offered an unusual opportunity to monitor change in child growth over time.

Figure 1. Summary of the study design. One of each matched pair of communities in each of three districts was randomised to either Group 1 or Group 2. Exposure to the intervention (inputs from Heifer Nepal) is shown in grey (darker grey indicates more active intervention; lighter grey indicates less active intervention). Group 1 communities started to receive inputs from Heifer starting after the baseline (B) survey. Group 2 communities received the intervention starting after the 12 month survey. The black open arrows indicate the times that surveys were conducted in both Group 1 and Group 2 communities (baseline, 6, 12, 18, 24 and 48 months). The intervention is ongoing in both communities. At the time reported in this paper, Group 1 communities had received the intervention for 48 months and Group 2 communities had received the intervention for 36 months

The study consisted of household surveys conducted every 6 months for 2 years (total, 5 surveys), followed by sixth survey completed 48 months after baseline. Data collection was performed by a local field research NGO (Nepal Technical Assistance Group) independent from the implementing organisation. At each survey time, field enumerators visited each household to complete a 125-item questionnaire with the female head of household or her designee. The questionnaire was based on standardised tools developed by ‘Measure DHS’, specifically the version used in the ‘Nepal Demographic and Health Survey’ conducted by the Government of Nepal and published in May 2007 (Joshi, Sherpa, and Toews 2003; Ministry of Health and Population Nepal, New ERA & Macro International Inc. 2012). At all six time points, data collection also included anthropometric measurements and health information (described later) on all enrolled children. All children who resided in participating households and who were between the ages of 6 months to 8 years at baseline were enrolled, and followed longitudinally. Children in participating households who reached 6 months of age during the course of the study were enrolled as they reached eligibility.

2.3. Intervention

The intervention consisted of an extensive programme of participatory community development activities tailored to rural Nepal. The principal interventions were implemented over 24 months. After these initial 24 months, Heifer remained involved in the communities, but to a lesser extent.

The basic training curriculum focused on poverty alleviation, citizen empowerment and community development, with a strong emphasis on optimisation of livestock management. Notably, the curriculum did not specifically address child nutrition or health. The community development activities were based on women’s Self-Help Groups (sometimes involving other family members), which met biweekly with a trained facilitator and were supplemented by specific interactive instruction, workshops, guidance and training. Training and development activities concentrated on improving animal management (addressing livestock health and husbandry, integration of livestock into the ecosystem and preservation of the environment) and increasing household income. After the initial 24 months of active intervention, Heifer continued with monitoring, occasional technical support, intermittent refresher training, and so forth. Further details of the study design and intervention have been published elsewhere (Heifer International 2014b; Miller et al. 2014)).

2.4. Household characteristics

Demographic information was collected on each household, including SES, animal ownership, annual income and amount of land owned. Scores for SES were largely based on household possessions and quality of housing (for example, housing structure, size, amenities [toilet and water facilities]); these scores were calculated using DHS-Nepal guidelines (Ministry of Health and Population Nepal, New ERA & Inc. 2012). Animal ownership, annual income and amount of land owned were also considered important indicators of household wealth, so this information was also collected. The number of animals owned was converted to a standardised ‘animal score’ using FAO Global Livestock Units (Food and Agricultural Organization 2003).

2.5. Anthropometry

The primary outcome of the investigation was child growth, specifically weight, height, head circumference and mid-upper arm circumference and changes in these measurements over time. Weight was measured with Seca 354 electronic scales (Hamburg, Germany) accurate to 10 g. Before each measurement, all scales were calibrated using standardised weights. Supine lengths were obtained for children ≤3 years and standing heights for those >3 years. Standing height was measured with a portable Seca 213 stadiometer accurate to 1 mm, barefoot and with the head in the auriculo-orbital plane. Supine lengths were measured with a Seca BabyMat 210. Head circumference was assessed with disposable paper tapes at the maximum occipito-frontal measurement. Mid-upper arm circumference was measured with disposable insertion tapes accurate to 1 mm (Harlow Printing Ltd, South Shields, Tyne and Wear, UK) midway between the tip of the olecranon and acromion processes.

Measurement techniques were standardised between enumerators (and within the same enumerators) at each cross-section and across the different survey points. The same team of trained enumerators conducted the surveys. At each survey time, 10 per cent of measurements were verified by supervisors. Regular calibration of equipment was performed to minimise measurement errors. Each measurement was obtained twice, and results were averaged. If results were >5 per cent discrepant, a third measurement was obtained. Results were converted to z-scores (for example, HAZ, WAZ and weight-for-height z-score [WHZ]) using WHO Anthro and Anthro Plus (World Health Organization 2011). The prevalence of underweight (WAZ < –2), stunting (HAZ < –2) and wasting (WHZ < –2) was determined according to World Health Organization standards (World Health Organization 2014).

2.6. Child health

Child morbidity related to fever, diarrhoea or respiratory symptoms was reported by care givers. A standardised interview with the child’s parent (usually the mother) elicited a report of these common symptoms of illness during the previous 2 weeks; for each symptom reported, a score of 1 was given. Thus, a child who had suffered all three symptoms (fever, diarrhea and respiratory problems) would have a score of 3, and a child who had none of these problems would have a score of 0. A dichotomised health score was then devised to identify children with ‘relatively good health’ (score 0 or 1) or ‘poor health’ (score 2 or 3).

2.7. Child defecation practices

Child defecation practices were selected as a proxy marker for household hygiene (Dangour et al. 2013b; Lin et al. 2013). Mothers were asked if the child (>1 years of age) used either a potty or toilet or practiced open defecation in the yard. For some analyses, baseline practices were used to evaluate the predictive value of these habits in relation to growth outcomes.

2.8. Statistical analysis

The original trial sample size was computed to detect a difference of ≥0.25 in mean WAZ at a power of 87 per cent and a two-sided significance level of 0.05. WAZ was selected as the parameter most likely to show early recovery in a population of children known a priori to be undernourished (Van Ijzendoorn, Bakermans-Kranenburg, and Juffer 2007; Miller et al. 2010; Palacios, Roman, and Camacho 2011; Karakochuk et al. 2012; Ackatia-Armah et al. 2015).

Data were entered and analysed using JMP 11.1 (Cary, NC, U.S.A.) and STATA version 12.0 (College Station, TX, U.S.A.). For some analyses, children were assigned to groups based on chronological age. Age groups were divided into clinically relevant segments: from 6 to 12 months, from 13 to 24 months, from 25 to 36 months, from 37 to 60 months and >60 months. Analysis was conducted at the community, household and individual level, starting with a descriptive analysis of the variables, including t-tests and ANOVA with Bonferroni post hoc tests to correct for multiple comparisons, followed by a series of chi-square tests and correlations to assess collinearity.

Dependent variables were evaluated with histograms to verify normal distributions. Mixed-effect regression models (using Stata command ‘xtmixed’) were utilised to predict the anthropometric z-scores (HAZ and WAZ), with child age, gender, household location (Terai or Hills), time period of exposure to the intervention, baseline measures of household income (per household member), SES, land ownership, animal score and child defecation practices (as a proxy marker for household hygiene) as fixed effects, and with data time point and child as random effects. In addition, we performed quantile regression analyses to explore the relative influence of various baseline household characteristics (child age, sex, location, intervention group, household land owned, income per household member, SES, animals owned and child defecation practices) on end-line (at the 48 month survey time) HAZ and WAZ scores, by median, 25th and 75th quantiles.

2.9. Ethical considerations

This longitudinal staggered-introduction field trial was approved (Reference #845; Renewal #1496) by the Nepal Health Research Council, the organisation responsible for oversight of research involving human subjects in Nepal (Government of Nepal 2014; Office for Human Research Protections (OHRP) 2014), and Tufts University. According to the guidelines of the Nepal Health Research Council, oral consent to participate was obtained at the initiation of the study and again from each respondent at each household visit. Each household was assigned a code; results of survey data were de-identified and entered into a password-protected database using this code.

3. Results

3.1. Participants

Group 1 and Group 2 were defined as having received 48 and 36 months of intervention, respectively. Overall, 3192 individuals (1545 male (M):1626 female (F); gender not recorded for 21 subjects) in 431 families (214 Group 1, 217 Group 2) were enrolled. Ages ranged from newborn to 94 years. There were 604 children (303 M: 301 F) within the target age range of 6–60 months. The number of household members enrolled varied (M ± SD) at the different survey times from 6.71 ± 2.82 at baseline, 6.65 ± 2.91 at 12 months, 6.50 ± 2.73 at 24 months, to 7.27 ± 3.29 at 48 months.

3.2. Household income and SES

At baseline, no differences were found between Group 1 and Group 2 in household income (neither gross household income nor income per household member), SES, land ownership or adult educational attainment. At the first evaluation 24 months after baseline, all households had increased incomes as well as SES. As previously reported, these improvements were greater in Group 1 (at that time, after 24 months of Heifer activities) than in Group 2 (at that time, after 12 months of Heifer activities) (Miller et al. 2014).

When families were re-evaluated at 48 months, substantial incremental gains in income and SES were found compared to baseline in both groups (Table 1), but there were no significant differences between in Group 1 and Group 2.

Table 1. Income and socioeconomic scores in participating households

	Change from baseline to 48 months
Mean ± SE	Income in NPR (U.S. $)	Income per household member in NPR (U.S. $)	Socioeconomic score (SES)
All households	112,586 ± 8032	15,112 ± 1105	0.407 ± .05
	($1103 ± 79)	($148 ± 11)
Group 1	110,252 ± 10564	15,534 ± 1397	0.352 ± .06
	($1080 ± 103)	($152 ± 14)
Group 2	115,141 ± 12254	14,664 ± 1734	0.463 ± .06
	($1128 ± 120)	($145 ± 17)
	Change per month of participation in Heifer activities
All households	2727 ± 200	364 ± 27	0.010 ± 0.001
	($27 ± 2)	($4 ± 0.3)
Group 1	2296 ± 275*	323 ± 38	0.007 ± 0.001**
	($23 ± 3)	($3 ± 0.4)
Group 2	3198 ± 288*	407 ± 39	0.012 ± 0.001**
	($31 ± 3)	($4 ± 0.4)

Total income, income per household member and socioeconomic scores from baseline to 48 months and per month of participation in Heifer activities are shown for households in Group 1 and Group 2, or all together.

*p ≤ 0.05; **p ≤ 0.01.

Gross household income increased by 2727 NPR (27 US$) per month of participation in Heifer activities (equivalent to 364 NPR (4 US$)/household member) (Table 1). In addition, socioeconomic score increased by 0.01 unit per month of participation in Heifer activities. Changes in these parameters (household income and socioeconomic score) per month of participation in Heifer activities were significantly greater in Group 2 than in Group 1 (p = 0.02 and p = 0.01, respectively); that is, improvement occurred over a shorter time period for these parameters.

3.3. Nutritional status of the children

Many children in the study areas were undernourished at baseline, with mean scores (M ± SD) for HAZ −1.48 ± 1.16, WAZ – 1.98 ± 1.19 and WHZ −1.34 ± 1.18 (Table 2). Head circumference z-scores were also low (−2.07 ± 0.96) (described in Miller et al. (2016)). There were no differences at baseline in these measures between children living in Group 1 or Group 2 areas. In both groups, by 48 months, mean HAZ, WAZ and WHZ all had increased significantly (to −1.34 ± 1.03 [p = 0.01], −1.56 ± 78 [p < 0.0001] and −0.98 ± 80 [p < 0.0001], respectively), compared to baseline measurements (Figure 2). Similar results were obtained for children age 6–60 months (Table 2).

Table 2. Anthropometric measurements of children in project area by time of survey

	Time of survey
	Baseline (0 months)		6 Months		12 Months		18 Months		24 Months		48 Months
	N	Mean (SE)	N	Mean (SE)	N	Mean (SE)	N	Mean (SE)	N	Mean (SE)	N	Mean (SE)
Height
HAZ
ALL	606	−1.48 (0.04)	540	−1.33 (0.05)*	635	−1.39 (0.04)	591	−1.28 (0.05) **	609	−1.31 (0.04)**	766	−1.34 (0.03) **
Age 6–60 months only	370	−1.47	302	−1.24	336	−1.31	273	−1.24	250	−1.22	237	−1.29
Prevalence of stunting (per cent)
All	606	32	540	27	635	30	591	25**	609	25**	766	25**
Age 6–60 months only	370	32	302	25	336	28	273	23	250	22	237	22
Weight
WAZ
All	607	−1.98 (0.05)	540	−1.98 (0.05)	634	−1.92 (0.04)	593	−1.87 (0.04)	611	−2.01 (0.04)	767	−1.56 (0.03) ****
Age 6–60 months only	371	−1.99	302	−1.97	336	−1.93	274	−1.91	250	−2.04	237	−1.60
Prevalence of underweight (per cent)
All	607	48	540	47	635	45	593	43	611	49	767	31****
Age 6–60 months only	371	49	302	48	336	47	274	45	250	50	237	35
Weight-for-height
WHZ
All	588	−1.34 (0.04)	498	−1.51 (0.05)*	577	−1.35 (0.05)	496	−1.39 (0.05)	480	−1.63 (0.05)***	662	−0.98 (0.80) ****
Age 6–60 months only	366	−1.28	298	−1.50	329	−1.34	269	−1.45	249	−1.69	238	−1.00
Prevalence of wasting (per cent)
All	588	24	498	33**	577	30*	496	28	480	34***	662	9****
Age 6–60 months only	366	24	298	34	329	31	269	32	249	37	238	11
Age
Age (months ± SD)
All	607	50.3 (25.7)	540	55.9 (26.7)	634	57.9 (28.3)	593	63.8 (31.2)	611	69.7 (32.0)	767	81.0 (37.1)
Age 6–60 months only	371	32.9 (15.6)	302	35.6 (15.0)	335	34.7 (15.3)	284	35.4 (15.5)	248	36.8 (14.6)	242	36.1 (16.5)

Anthropometric measurements are shown at each survey time for all children, and for those ages 6–60 months only. Results for each row are compared to baseline values for that row.

*Indicates p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. For simplicity, SE and indicators of significance are omitted for results of children 6–60 months.

Figure 2. Mean z-scores for height, weight and weight-for-height over 48 months for children in Group 1 and Group 2. The mean z-scores for HAZ (solid line), WAZ (dashed lines) and WHZ (dotted lines) are shown at each survey time. Group 1 is indicated in black; Group 2 in grey. In Group 1, mean HAZ scores increased between baseline and 24 months (a vs. a, p = 0.02); mean WHZ increased from baseline to 48 months (e vs. e, p < 0.001). In both Group 1 and Group 2, WAZ increased from baseline to 48 months (b vs. d, p < 0.001) as well as from 24 to 48 months (c vs. d, p < 0.001). Mean ages of the children at each time point are shown in Table 2

The percentage of children with z-scores < −2 for height and weight were relatively unchanged for the first 24 months of the project, in both the Group 1 and Group 2 project areas (Table 2). However, when children were reassessed after 48 months, there was a significant decrease in the percentage of underweight and wasted children (from ~50% to ~31% and from ~24% to 9%, both p < 0.0001, respectively) (Figure 3). These improvements were seen in both Group 1 and Group 2 areas (after 48 and 36 months of intervention, respectively). Nearly identical results were found when only children <60 months were assessed (Table 2).

Figure 3. The percentage of children with wasting, stunting and underweight assessed over 48 months. The percentage of children with underweight (dashed lines), wasting (dotted lines) and stunting (solid lines) at each survey time is shown. Group 1 is indicated in black; Group 2 in grey. The percentage of stunted children did not significantly change, but the percentage of wasted and underweight children diminished significantly at the 48 month survey, compared to earlier time points as shown. a vs. b, p < 0.0001 and c vs. d, p < 0.0001 for both Group 1 and Group 2 (letters shown only for Group 1 at each time point for clarity; same p values for Group 2 were found). Mean ages of the children at each time point are shown in Table 2

The percentage of children with stunting decreased significantly as well, but this improvement was of lesser magnitude (32–25%, p < 0.0001) and occurred in the first 18–24 months of observation (more rapidly among children in Group 1). There was no further improvement or decline in the percentage of children with stunting at 48 months; this remained unchanged at 25 per cent.

We next examined growth patterns in several additional ways: First, mean z-scores at baseline were compared with mean scores from other survey times for children in Group 1 and Group 2 households. The mean HAZ scores improved significantly in Group 1 at 24 months (−1.49 ± 0.06 to −1.28 ± 0.06, p = 0.02); the mean HAZ for Group 2 also improved but was not significantly different from baseline. No further improvement was noted in HAZ by 48 months for either group.

A different pattern was seen for weight. Mean WAZ scores remained relatively unchanged from baseline to 12 and 24 months in both groups. However, by 48 months, mean WAZ scores increased significantly in both Group 1 and Group 2, compared both to baseline (both p < 0.0001) and to 24 months (both p < 0.0001). Thus, improvements in weight appeared in the later phase of the intervention.

Second, we examined changes using matched pairs (for each child individually). WAZ scores increased significantly from baseline for Group 1 children at 18, 24 and 48 months (Figure 4). However, for Group 2 children, the only significant increase in WAZ from baseline was found at 48 months. Similarly, HAZ increased significantly from baseline for children in Group 1 at 6, 12, 18, 24 and 48 months. A significant HAZ increase from baseline for Group 2 was similarly seen only at 48 months. Thus, the children in Group 1 showed earlier and larger improvements in their growth for both height and weight, despite the fact that the intervention did not directly address nutrition.

Figure 4. Pair-matched changes from baseline in HAZ and WAZ. Pair-wise comparisons of HAZ and WAZ from baseline to each of the other survey times (6, 12, 18, 24 and 48 months). In Group 1, HAZ increased significantly from baseline at 6, 12, 18, 24 and 48 months. For Group 2, a significant increase from baseline was seen only at 48 months. WAZ scores increased significantly from baseline for Group 1 children at 18, 24 and 48 months, but for Group 2 children only at 48 months

3.4. Child defecation practices

Child defecation practices (of children between 1 and 5 years of age) were determined at each survey time. The pattern of these practices changed very significantly over time (Figure 5). The percentage of young children using a toilet for defecation increased and open defection correspondingly decreased. Similar results were obtained for older children as well. Findings were similar in both Group 1 and Group 2.

Figure 5. Changes in child defecation practices over 48 months of surveys (Children between 1 and 5 years of age). Child defecation practices became more sanitary over 48 months of observation, with fewer young children in both Group 1 (black) and Group 2 (grey) using the yard (dashed lines) instead of a potty or toilet (solid lines). p-Values were identical for Group 1 and Group 2

3.5. Nutritional outcomes in relation to household and demographic characteristics

Next, to account for differences in individual characteristics (age, gender, location of residence [Terai vs. Hills], group assignment) and baseline household demographic characteristics (baseline land ownership, income, SES, animal score and child defecation practices), mixed effects regressions were performed (Table 3). These regressions assessed the effect of duration of exposure to the Heifer intervention (ranging from 12 to 48 months), regardless of group assignment. Child defecation in a toilet or potty at baseline was positively associated with both HAZ and WAZ. In addition, HAZ was positively related to residence in the Terai, baseline SES and duration of participation in the Heifer intervention (‘dose–response’). Duration of participation in the intervention was predictive of higher WAZ only at 48 months.

Table 3. Regression analysis of factors predicting HAZ and WAZ

HAZ	Coeff.	SE	z	p > |z|	95% confidential interval
Female	−0.1818603	0.1321626	−1.38	0.169	−0.4408942	0.0771737
Age (months)	−0.0073991	0.0042386	−1.75	0.081	−0.0157066	0.0009083
Terai	0.5703652	0.1588322	3.59	0.000	0.2590599	0.8816705
Duration of exposure to intervention (months)
12	0.2428482	0.1082463	2.24	0.025	0.0306894	0.4550071
24	0.4859282	0.1443461	3.37	0.001	0.2030151	0.7688414
36	1.25383	0.8939839	1.40	0.161	−0.498346	3.006006
48	0.5119622	0.223338	2.29	0.022	0.0742277	0.9496967
Household status at baseline
Land owned (M2)	0.0000115	0.000019	0.60	0.545	−0.0000258	0.0000488
Income per HH member (NPR)	3.74e-06	9.64e-06	0.39	0.698	−0.0000152	0.0000226
Socioeconomic status (SES)	0.3990488	0.1510738	2.64	0.008	0.1029495	0.6951481
Animals owned (animal score)	−0.0917985	0.057519	−1.60	0.110	−0.2045337	0.0209368
Child defecation practice (toilet/yard) 1 = toilet, 0 = yard	0.6916057	0.1956108	3.54	0.000	0.3082155	1.074996
Cons	−2.155888	0.331894	−6.50	0.000	−2.806388	−1.505388
WAZ
Female	−0.0125387	0.146983	−0.09	0.932	−0.30062	0.2755426
Age (months)	−0.0003201	0.0046982	−0.07	0.946	−0.0095284	0.0088882
Terai	0.1915439	0.1758516	1.09	0.276	−0.153119	0.5362068
Duration of exposure to intervention (months)
12	0.0462459	0.1037805	0.45	0.656	−0.1571602	0.249652
24	0.1562181	0.1473936	1.06	0.289	−0.132668	0.4451042
36	0.4361097	0.8279034	0.53	0.598	−1.186551	2.058771
48	0.4829532	0.2406149	2.01	0.045	0.0113567	0.9545497
Household status at baseline
Land owned (M2)	−0.0000104	0.0000213	−0.49	0.623	−0.0000522	0.0000313
Income per HH member (NPR)	5.65e-06	0.0000107	0.53	0.597	−0.0000153	0.0000266
SES	0.2323994	0.1684623	1.38	0.168	−0.0977806	0.5625795
Animal score	−0.054513	0.0637893	−0.85	0.393	−0.1795378	0.0705117
Child defecation practice (toilet/yard) (1 = toilet, 0 = yard)	0.7110651	0.2173947	3.27	0.001	0.2849793	1.137151
Cons	−2.526196	0.3675127	−6.87	0.000	−3.246508	−1.805884

Significant predictors are shown in bold.

3.6. Child health

Child health scores, reflecting the symptoms reported during the 2 weeks prior to each survey time, were then compared. Scores improved significantly over the 48 months of observation. At baseline, only 79 per cent of children were classified as having ‘good health’; this number increased to 93 per cent by 48 months (Table 4). Similar results were found when only children <60 months were included. Results were similar – showing the same direction and level of significance – when analysed separately by group assignment and gender.

Table 4. Health status

	Per cent (%) of children classified as having ‘good health’ in previous 2 weeks
	All children (%) (N)	Children <60 months of age (%) (N)
Baseline	79 (594)^a	73 (365)^c
6 Month survey	76 (533)	72 (305)
12 Month survey	88 (651)^b	85 (399)^d
18 Month survey	87 (594)^b	81 (274)^e
24 Month survey	92 (601)^b	87 (254)^f
48 Month survey	93 (774)^b	90 (238)^f

The percentage of children classified as having good health in the previous 2 weeks is shown for each survey time. ^a vs. ^b, p < 0.0001, ^c vs^{. d}, p = 0.0005, ^c vs.^e, p = 0.01, ^c vs. ^f, p < 0.0001.

3.7. Changes in growth related to anthropometric status at baseline

We next examined the changes in growth related to child anthropometric status at baseline (Table 5). Children categorised as stunted, underweight or wasted at baseline (HAZ, WAZ and WHZ scores ≤ −2, respectively) showed significantly more improvement in growth measurements than those with baseline z-scores > −2.

Table 5. Change in growth related to anthropometric status at baseline

	Change in HAZ: baseline to 48 months		Change in WAZ: baseline to 48 months		Change in WHZ: baseline to 48 months
	Status at baseline
	Stunted	Not stunted	Underweight	Not underweight	Wasted	Not wasted
Group 1 (N = 231)	0.66 ± 0.11*	0.02 ± 0.07	1.05 ± 0.07*	0.02 ± 0.07	1.31 ± 0.12*	0.06 ± 0.11
Group 2 (N = 260)	0.78 ± 0.10*	−0.15 ± 0.07	1.06 ± 0.08*	−0.10 ± 0.07	1.08 ± 0.07*	0.04 ± 0.06

The magnitude of change in z-scores between the baseline and 48 month surveys was related to nutritional status at baseline. Children who were undernourished at baseline had significantly more improvement in HAZ, WAZ, and WHZ than those whose baseline z-scores were > −2.

*All p < 0.0001 for comparisons of pairs in same row.

Throughout the investigation, children with ‘better health’ had significantly higher gains in WAZ scores than children with ‘poor health’, for both intervention Groups (Figure 6). No such relation was seen between health and HAZ scores. Thus, health and WAZ scores improved in parallel.

Figure 6. HAZ and WAZ related to health status. Anthropometric z-scores for height and weight were compared for children with either ‘poor health’ or ‘relatively good health’ using a composite score from results at baseline, 12, 24 and 48 months. No differences were seen in HAZ scores for these two groups of children, however, WAZ were significantly better in children with better health. Black = Group 1, grey = Group 2. Solid bars, children with poor health; striped bars, children with good health. N = 1782 for Group 1, N = 1915 for Group 2

3.8. Impact of baseline household and individual characteristics on child growth

We next examined the impact of baseline household and individual characteristics on child growth at 48 months. We used quantile regressions as a means of exploring whether some factors were more important at the lower end of the distributions of z-scores (Table 6). The quantile regression analysis showed that location of residence (Terai vs. Hills) was an important factor influencing HAZ for all children across all quantiles. However, for the children whose height at 48 months was in the 25th quantile, other baseline factors significantly predicted HAZ, specifically: child’s age and baseline household SES and income per HH member, which were not significant at the median and 75th quantile. Location of residence in the Terai also was strongly predictive of WAZ for all children. In addition, child defecation practices significantly related to WAZ for children whose weights were in the median or 25th quantiles (but not for those in the 75th quantile). Socioeconomic score was significantly associated with higher WAZ only at the median.

Table 6. Quantile regression of HAZ and WAZ scores at 48 months

HAZ
Regression number	1	2	3
Dependent variable	HAZ	HAZ	HAZ
Estimator	Quantile	Quantile	Quantile
Quantile value	50th = −1.34	25th = −2.017	75th = -0.737
Age (months)	0.215 (0.004)	0.028 (0.005)^a	0.103 (0.005)
Female	0.675 (0.131)	0.555 (0.155)	−0.982 (0.166)
Terai	0.000 (0.152)	0.000 (0.180)	0.000 (0.193)
Intervention	0.110 (0.135)	0.720 (0.159)	0.437 (0.171)
Land owned	0.396 (0.000)	−0.788 (0.000)	0.910(0.000)
Income^b	−0.435 (0.000)	−0.036 (0.000)	0.148 (0.000)
SES	0.052 (0.130)	0.001 (0.153)	0.220 (0.164)
Animals Owned	0.532 (0.047)	0.419(0.056)	0.243(0.060)
Child defecation practice^c	0.244 (0.172)	0.085 (0.203)	0.370 (0.218)
Cons	−0.000 (0.528)	0.000 (0.624)	0.007 (0.670)
WAZ
Regression number
Dependent variable	WAZ	WAZ	WAZ
Estimator	Quantile	Quantile	Quantile
Quantile value	50^th = −1.59	25^th = −2.103	75^th = −1.002
Age in months	0.474 (0.004)	0.167 (0.005)	0.880 (0.005)
Female	0.544 (0.116)	0.244 (0.130)	0.425 (0.140)
Terai	0.037 (0.135)	0.004 (0.374)	0.005 (0.163)
Intervention	−0.834 (0.119)	0.833 (0.024)	0.971(0.146)
Land owned	−0.551 (0.000)	−0.094 (0.000)	−0.353(0.000)
Income^2	0.220 (0.000)	0.398 (0.000)	0.194 (0.000)
SES	0.020 (0.115)	0.187 (0.110)	0.061 (0.140)
Animals owned	0.333 (0.042)	0.378 (0.040)	0.132 (0.050)
Child defecation practice^3	0.001 (0.153)	0.005 (0.147)	0.098 (0.185)
Cons	−0.000 (0.330)	0.000 (0.624)	0.000 (0.401)
N	293	293	293

Quantile regressions were used to examine how variables impacted the growth of children at the lower end of the distributions of z-scores. In addition to location of residence, child’s age and baseline household SES and income were significantly related to HAZ in children whose height at 48 months was in the 25th quantile. Location of residence also related to WAZ for all children. Child defecation practices significantly related to WAZ for children whose weights were in the median or 25th quantiles, but not for those in the 75th quantile.

^a p (SE), significant results are shown in bold.

^b Toilet = 1.

^cYard = 0.

4. Discussion

In this longitudinal study over 48 months, notable nutrition and health benefits were found for children in households participating in a community development intervention in rural Nepal. Mean WAZ, WHZ and HAZ scores of children significantly improved, and most notably, the percentage of underweight, wasted and stunted children decreased markedly (from ~50% to 31%, from ~24% to 9% and from 32% to 25%, all p < 0.0001). The improvements in WAZ and WHZ occurred between the 24 month and 48 month surveys; the improvement in HAZ occurred earlier (18–24 months). We also found that child health scores generally improved over the 48 months of observation, with an increasing proportion of children classified as having ‘good health’ during the 2 weeks prior to each survey. These health scores corresponded to WAZ scores, underscoring the strong relationship between weight gain and child health (World Health Organization 2014).

This plateau in improvement in HAZ likely reflects multiple factors, which affect long-term nutritional status, including the cumulative effect of multiple illnesses, difficulties in accessing health services, ongoing micronutrient and caloric deficiencies, environmental factors and poor hygiene, erratic food supplies (depending on harvests), household food allocation practices and decision-making about resources and the birth of new children in the household. These factors may have acted in concert to negatively impact child statural growth. Moreover, the Heifer programme provided no specific training related to child nutrition or diet. Thus, although household income and SES improved in participating households, dietary practices may not have changed adequately to further reduce stunting.

Overall, the changes in child growth were predicted by baseline characteristics of the household, including income, SES, land ownership and animal ownership, and one measure of basic hygiene practice. It was striking to find the largest incremental increases in growth among those children who were most undernourished (stunted, wasted and underweight) at baseline. In addition, it was clear that some factors were more important at the lower end of the distributions of z-scores. Individual and household characteristics including child’s age and baseline household SES and income related to HAZ for children in the lowest height quantile, while child defecation practices significantly related to WAZ for children in the median or 25th weight quantiles.

The livelihoods-based intervention conducted by Heifer Nepal focused on improving household income and SES. These parameters indeed improved by 48 months in households participating in the Heifer intervention. (As no survey was conducted at 36 months we do not know when Group 1 achieved these gains.)

It is possible that inputs may have functioned synergistically with a general improvement in the economic situation in rural Nepal during those years. That is, families with better incomes and SES were better able to take advantage of the intervention. However, the results of the surveys performed at 12 months (when Group 2 had not yet begun to receive inputs) showed striking differences between Group 1 and Group 2, suggesting – although not proving – that the activities directly influenced these outcomes. In summary, we indeed found some benefits became apparent only after 48 months of observation, and that the benefits for both groups appeared to converge over time.

Recent policy approaches have emphasised the desirability of food-based, agricultural solutions to malnutrition (Tontisirin, Nantel, and Bhattacharjee 2002; Demment, Young, and Sensenig 2003; Kassa et al. 2003; Berti, Krasevec, and FitzGerald 2004; Leroy and Frongillo 2007; Bezner Kerr, Berti, and Shumba 2010; Webb Girard et al. 2012; Bhutta et al. 2013; Haddad 2013; Ruel and Alderman 2013; Webb and Kennedy 2014), suggesting that this strategy would be most likely to succeed by improving access to increased food supply and achieving dietary diversification. The potential sustainability of this approach is also appealing (Allen 2003; Demment, Young, and Sensenig 2003).

Livestock interventions, such as the one provided by Heifer in this study, have been suggested as one type of agricultural activity with the means to improve child nutrition (Berti, Krasevec, and FitzGerald 2004; Leroy and Frongillo 2007; Masset et al. 2012; Webb Girard et al. 2012), possibly by way of increased availability and consumption of animal source foods. Such dietary changes could favourably impact child growth. However, the efficacy of such interventions has not been rigorously studied (Webb and Kennedy 2014). Recent reviews have noted substantial research gaps and study limitations in understanding the effects of household food production strategies on nutrition outcomes of young children, noting that agricultural strategies with integrated nutrition education, gender and nutrition objectives had the best outcomes (Masset et al. 2012; Webb Girard et al. 2012; Haddad 2013; Webb and Kennedy 2014).

In general, less is known about the effect on child nutrition when agricultural strategies are presented in the context of strong community development and women’s empowerment, as is the case with Heifer programmes. Even without a specific focus on child nutrition, children residing in project areas in rural Nepal experienced modest but significant gains in linear growth 12-24 months (Miller et al. 2014) and marked reduction in underweight and wasting 36-48 after the introduction of the intervention. Similar to the report of Bezner Kerr et al., children in families with a longer duration of participation had better growth (Bezner Kerr, Berti, and Shumba 2010).

This study had several strengths and weaknesses. The major strength was the number of children who were followed longitudinally, with 4 years of follow-up of anthropometry in the context of detailed family demographic information. The study was conducted by a well-trained and consistent group of field enumerators, with built-in controls for quality assurance and extreme care taken to obtain accurate anthropometric measurements. Possible weaknesses of the study include the staggered intervention design, which did not permit observation of an ‘unexposed’ control group over the entire 4 years. However, the clear differences that emerged between Group 1 and Group 2 households allowed us to determine the impact of duration of exposure on growth outcomes. Although the per cent of underweight children was reduced in both groups by 48 months, the incremental differences between Group 1 and Group 2 bolsters the assumption that this improvement was related to exposure to the intervention. Other possible concerns relate to methodologies applied. For example, it is not known if the health scores reported here, although calculated every 6 months for the previous 2 weeks, accurately reflect overall health during the entire 4 years. However, methods were designed to be as accurate as possible and it was expected that any measurement error was random across families. There is also the possibility that some of the measurements were inaccurate. However, care was taken to reduce this possibility. Refresher courses were conducted for field staff prior to each survey, the same teams of field staff conducted each survey, and the same calibrated equipment was used throughout. All measurements were obtained in duplicate (or more, if discrepancies were revealed). In addition, a subset of children was remeasured at each survey by field supervisors to maintain ongoing quality control. In addition, we were unable to control for the self-selection process inherent in such a programme; it is possible that participants in Heifer activities differed from non-participants in motivation and actions. Finally, we recognise that the prevalence rates of undernutrition (particularly underweight) in our baseline survey differ from results reported in the Demographic and Health survey of Nepal. However, our target population was not a nationally representative sample. Rather, these communities were selected because they were particularly disadvantaged. It is possible that these communities were suffering from a specific, acute hunger crisis that could have increased the rate of underweight among the children, although we have no specific data to support this. However, other researchers have reported similar (Pramod Singh et al. 2009) or proportionally similar prevalence rates among specific populations in Nepal (Ghosh et al. 2009; Sapkota and Gurung 2009; Manohar et al. 2014; Pokhrel et al. 2016).

Although there is great interest in the impact of agricultural interventions on child nutrition, there have been few long-term studies which examine an integrated approach, as recommended by many commentators (Leroy and Frongillo 2007; Randolph et al. 2007; Masset et al. 2012; Webb Girard et al. 2012; Haddad 2013; Ruel and Alderman 2013; Webb and Kennedy 2014). Our intervention consisted of a livestock and livelihoods approach, with strong social capital development (including gender awareness), but without a specific nutrition or health focus.

In addition to improved income and SES, Heifer’s strong focus on gender empowerment may have indirectly benefitted child nutrition through increase in the social status of women participants, allowing more control over household resources (Olney et al. 2009). Moreover, the community development focus of the intervention emphasised behaviour change at household and community levels. This intensive approach is in accordance with the recommendations of many researchers (Leroy and Frongillo 2007; Webb Girard et al. 2012; Bhutta et al. 2013; Haddad 2013; Ruel and Alderman 2013; Webb and Kennedy 2014) .

These findings suggest that a nutrition-sensitive package of agricultural interventions can have significant and important impacts on child growth despite lacking explicit targeted nutrition interventions as part of its design; some of these impacts may take several years to be manifest. The programme was designed to increase household wealth and SES in the context of strengthening community development and social capital. In addition to achieving these goals, improvements in child nutrition not included in the programme’s explicit goals were documented over 4 years of observation.

Thus, holistic approaches to community development may be effective in improving child nutrition, even when this is not a specific aim of the activity. Improved understanding of this process may benefit other programmes to maximise such effects. Longer project cycles may reveal gains over time; such benefits may be linked to more sustainable improvements after intervention activities.

Acknowledgements

Contributions by Deepak Thapa and Preeti Subbha, and the Nepal Technical Assistance Group team are deeply appreciated. Financial support from Heifer International and the Nutrition Innovation Laboratories is also acknowledged. Funding was provided by the USAID Feed the Future Innovation Laboratory for Collaborative Research in Nutrition for Asia (award number AID-OAA-l-10-00005) and Africa (award number AID-OAA-L-10-00006) to the Friedman School of Nutrition Science and Policy, Tufts University. The content of this paper is solely the responsibility of the authors and does not necessarily represent the official views of USAID.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by Heifer International and the USAID Feed the Future Innovation Lab for Nutrition [award numbers AID-OAA-l-10-00005, AID-OAA-L-10-00006] awarded to Friedman School of Nutrition Science and Policy, Tufts University.

Notes on contributors

Laurie C. Miller

Laurie C. Miller is Professor of Pediatrics and Adjunct Professor of Nutrition and Child Development at Tufts University.

Neena Joshi

Neena Joshi is the Director of Programs for Heifer International Nepal.

Mahendra Lohani

Mahendra Lohani is the Senior Vice President of Programs for Heifer International. Under his leadership, programs in Africa, Asia and Europe are focused on building strong social capital in addition to livestock development and wealth-creating value chains.

Beatrice Rogers

Beatrice Rogers is Professor of Economics and Food Policy and Director of the Food Policy and Applied Nutrition Program at Tufts University.

Meghan Kershaw

Meghan Kershaw is a program manager and researcher at the Friedman School of Nutrition at Tufts University with a research interest in impact evaluations for maternal child health and nutrition interventions.

Robert Houser

Robert Houser is a statistician at Friedman School of Nutrition Science and Policy.

Shibani Ghosh

Shibani Ghosh is a public health nutritionist with over 15 years of experience working with malnutrition in all its forms, and formulating and implementing evidence-based interventions.

Jeffrey K. Griffiths

Jeffrey K. Griffiths is an infectious diseases physician by training and is considered a leader on neglected biological pathways linking agriculture and nutrition, as well as an expert on water sanitation.

Shubh Mahato

Shubh Mahato is the Country Director of Heifer International Nepal.

Patrick Webb

Patrick Webb is a food policy and nutrition programming expert, engaged in research and policy guidance around the globe, as a professor of nutrition at Tufts University and as a PI on a variety of grants.