Thresholds of socio-economic and environmental conditions necessary to escape from childhood malnutrition: a natural experiment in rural Gambia

Severe growth faltering leading to stunting and underweight remains highly prevalent in low- and middle-income countries [1] (LMICs) and is estimated to play a role in 45% of under-5 year child deaths [2]. An analysis of data from 54 such countries shows that babies are born small and then follow a remarkably uniform pattern of stunting initiated soon after birth and continuing to a nadir at about 24 months [1]. In Africa and SE Asia average height-for-age Z-scores (HAZ) fall to less than − 2 HAZ against the WHO Growth Reference Standard [1]. Weight-for-age Z-scores (WAZ) vary more widely between populations.

Despite many decades of research, the precise aetiology of this growth failure remains obscure. It is ecologically associated with a number of correlates of poverty including generally poor diets of low diversity; food insecurity; poor water supply, sanitation and hygiene; frequent infections and inflammation (including enteric infections that cause a persistent environmental enteropathy that may inhibit nutrient uptake); and constraints on mothers’ time combined with poor parental understanding of the principles of childcare. However, amelioration of some of these factors frequently has a disappointing impact on child growth. For instance, during 35 years of intensive clinical and nutritional interventions in rural Gambia, we have reduced child mortality rates over 10-fold [3] and observed a profound decline in diarrhoeal disease [4], yet the prevalence of stunting is still 30% [5]. Others have shown that careful evaluation of nutritional and educational interventions either singly or in combination achieve improvements in HAZ of less than 0.3 Z-scores [6]; representing less than one sixth of the average deficit.

The excellent prenatal and post-natal growth shown by the cohorts of carefully selected children used to derive the INTERGROWTH Fetal Growth [7], and WHO Child Growth Standards [8] demonstrate the potential for excellent growth when the environmental conditions are right. Furthermore, rates of stunting decline quite rapidly as countries pass through the demographic transition. [9]. So why is it so difficult to improve child growth in low-income settings? The current paradigm is that multiple interventions must be introduced simultaneously [10] and numerous trials combining efforts to improve water, sanitation, hygiene (WASH), diet, breast-feeding rates, infant stimulation and parental education are in progress. Notably, the large-scale WASH Benefits trials in Kenya and Bangladesh have recently reported their outcomes with respect to stunting [11, 12]. The intervention arms that promoted infant and young child feeding (IYCF) showed a significant reduction in stunting but with a very modest effect size (0.13–0.16 Z-score for length in Kenya and 0.13–0.25 Z-score in Bangladesh). The SHINE Trial in Zimbabwe has also reported significant but limited success with IYCF (0.16 Z-score for length) [13]. Disappointingly, none of the WASH interventions reduced stunting.

In this study, we exploited a natural experiment to examine thresholds of socio-economic development associated with growth faltering. Our research centre at MRC Keneba is embedded within a rural subsistence-farming Gambian village. The centre employs over 200 indigenous staff ranging from scientists, physicians and senior management, through nurses, laboratory technicians and fieldworkers to lower grades of support staff such as cleaners many of whom are natives of Keneba or nearby villages. We also studied villagers without any paid employment. These circumstances have created a very wide socio-economic and educational gradient within the same confined ecological setting and allowed us to examine the effects of these gradients on child growth.

On average fathers had 2 wives (range 1–4) and 12 children (range 1–33). Mothers averaged 6 children (range 1–12) and household size averaged 13 (range 4–27). Fathers averaged 3.5 years of formal education (range 0–18 years) and 8.0 years of Arabic/Islamic studies (range 0–24). Mothers averaged 2.0 years of formal education (range 0–16 years) and 4.4 years of Arabic/Islamic studies (range 0–13). Reported incomes ranged from less than US$50 to over US$20,000 per annum. Eight percent of families owned a car, 12% a motorcycle and 55% a bicycle. Seventeen percent had a refrigerator, 32% had a television and 19% had electricity at home. Housing ranged from very basic mud block huts to Western-style housing with hot and cold running water, showers, flushing toilets, gas cooking, refrigerators, mosquito screens, fans and air conditioning.

The mean anthropometric statistics for the whole sample were HAZ (% stunted), boys = − 1.94 (40%), girls = − 1.55 (35%) (p = 0.018 for sex); WAZ (% underweight) boys = − 1.46 (31%), girls = − 1.32 (21%) (p = 0.052 for sex); and WHZ (% wasted) boys = − 0.64 (5%), girls = − 0.68 (6%) (NS for sex). Notably, The Gambian children’s heights and weights at 2 years of age were very similar to the average heights and weights of WHO reference children at 1 year of age.

Table 2 details the mean anthropometric Z-scores according to the 5 pre-determined SES groupings (see Fig. 1 for the interquartile ranges). HAZ varied strongly with SES with a range from − 0.67 in SES1 to − 2.23 in SES5 (P < 0.001); the proportions of stunted children were 26 and 54% respectively. Surprisingly, the trend between SES2, defined as upper-middle SES, and the lowest category of SES5 only spanned 0.65 Z-scores; and the proportion stunted varied from 37 to 54%. WAZ varied somewhat less strongly with SES with a range from − 0.90 in SES1 to − 1.64 in SES5 (P < 0.001); the proportions of underweight children were 31 and 43% respectively. The range in WAZ between SES2 and SES5 was only 0.44, and the proportions of underweight children were 17 and 43%. WHZ did not vary at all across SES groupings.

Variable	Mean per SES category					P value for trend
	1 upper	2 upper-middle	3 middle	4 lower-middle	5 lower
N	19	46	72	65	28
HAZ score	− 0.67 3,4,5	− 1.58	− 1.69 1	− 2.16 1	− 2.23 1	< 0.001
% stunted	26	37	44	54	54
WAZ score	− 0.90	− 1.20	− 1.35	− 1.67	− 1.64	< 0.001
% underweight	31	17	19	34	43
WHZ score	− 0.63	− 0.56	− 0.63	− 0.77	− 0.68	0.255
% wasted	10	4	4	9	4
Mothers’ heights	159.6	162.0	164.0	161.3	161.9	0.014
Fathers’ heights	173.5 4	171.3 4	170.6	167.8 1,2	170.4	< 0.001

Superscripts indicate the SES groups against which values are significantly different at P < 0.01 by Scheffé’s post hoc test. For the children’s anthropometry, their parents’ heights were included in the model. Child sex was also included in the model

Parental heights were correlated with offspring HAZ (mother’s coefficient 0.66 (SE 0.13) Z-score per 10 cm (P < 0.0001); father’s coefficient 0.52 (SE 0.15) Z-score per 10 cm (P < 0.0001). There was no difference in maternal heights across the 5 or 6 SES groups. Paternal heights in SES1 and 2 were greater than in SES4. When SES1 was subdivided, it was the father’s in SES1B who were tallest and significantly taller than SES4. Father’s heights in SES1A were not significantly different to any other group.

The major WASH Benefits trials in Kenya and Bangladesh [11, 12] and the SHINE trial in Zimbabwe [13] have recently reported that promotion of improved water, supply, sanitation and handwashing had no effect on linear growth. These very disappointing results from large-scale, rigorously implemented trials in appropriate target populations require an explanation. Our current study suggests that the intent was correct but that the environmental hygiene threshold that permits normal child growth is far higher than achieved in WASH Benefits and SHINE. The original selection strategy of participants for the WHO MGRS [17] that forms the basis of the current WHO Growth Standards implicitly recognised this same point.

The key feature of this study is that the participant families had identical access to health care and, by virtue of living in the same small village, shared common exposures to many other health-determining factors. This allowed us to interrogate the influence of 5 socio-economic domains (occupation, income and possessions, education, access to water and sanitation, and housing) with far fewer confounders than would normally exist in such analyses. We were also able to examine a very steep gradient ranging from the poorest village families with very few resources to families in which one or both parents had lived and trained in high-income countries and had advanced educational qualifications and incomes that were several orders of magnitude higher than the villagers. Growth data was routinely measured and recorded for all children in the same anthropometric clinic, enabling an unbiased analysis of their attained size at 2 years of age; the growth nadir in this population.

The highly significant gradient in HAZ and WAZ across SES groups is unsurprising and validates the relevance of the SES metrics. The remarkable feature of the data is that the only children growing close to the WHO Growth Standard were those living in Western-type accommodation within the MRC campus (HAZ + 0.68, WAZ + 0.36, WHZ − 0.03). The children of staff of almost similar rank who, by virtue of a shortage of housing on the campus, lived in rented housing in the village fared substantially less well (HAZ − 1.65, WAZ − 1.64, WHZ − 1.07). The HAZ and WAZ advantage of children in SES1A was not explained by differences in parental heights suggesting that it is unlikely to reflect an inter-generational advantage.

Our analysis is unique to date and has the great strength of removing the influence of the healthcare and environmental variables listed in Box 1 that are concordant across all families and that otherwise would confound interpretation of the effects of the SES gradient studied. Its chief weakness is the limited sample size, and replication is needed if a similar natural experiment can be found elsewhere. The sample size limitation does not limit the key observation that child growth was far from desirable even in families with high levels of education, very good incomes and absolute food security. In a comparison of group SES1A and SES2 combined (n = 57) against the SES4 and SES5 groups combined (n = 93) HAZ varied by only 0.59 Z-scores from − 1.59 to − 2.18 and the prevalence of stunting from 39 to 54%. WAZ varied by only 0.38 Z-scores from − 1.28 to − 1.66. WHZ did not differ. An additional weakness is that the SES data were collected up to 15 year after the anthropometric data. This is of limited concern because we allocated occupation and salary ranking in MRC staff according to their grade when their child was 24 m and social mobility is very slow in villagers. Additionally, any misclassification would tend to nullify associations rather than introduce bias.

The modifiable components of children’s growth (i.e. the non-genetic components) are influenced by a myriad of nutrition-specific factors (breast-feeding and complementary feeding practices, food availability and quality, and parental knowledge of the principles of child feeding) and nutrition-sensitive factors (sanitation and hygiene, access to health care, infections and stress). This likely explains why interventions aimed solely at improving complementary feeding, and diet have shown limited efficacy, even when conducted under the rigorously controlled conditions of research trials. Meta-analysis of trials involving feeding and/or educational interventions demonstrate average HAZ responses of less than 0.3 Z-scores; equivalent to about one sixth of the usual deficit by 2 years of age in LMICs [6]. Following these disappointments, it was argued that the quality of nutrition interventions had to be improved still further with the provision of a comprehensive range of micro- and macro-nutrients, and large trials of lipid-based nutritional supplements have recently been completed. Unfortunately, these have, by and large, also yielded disappointing outcomes ranging from no significant impact [18–21] to maximum benefits again not exceeding 0.3 Z-scores for height [22, 23]. Very recently, the WASH Benefits and SHINE trials have again confirmed this result. The groups receiving promotion of the WHO IYCF guidelines showed improvements of 0.13 (Kenya) and 0.25 (Bangladesh) Z-score in the nutrition group and of 0.16 (Kenya) and 0.13 (Bangladesh) Z-score in the nutrition plus WASH groups against the average HAZ in the control groups of − 1.54 in Kenya and − 1.79 in Bangladesh [11, 12]. The SHINE trial has recently reported very similar results with an improvement of 0.16 for HAZ in the IYCF group and 0.02 in the IYCF+WASH against a background level of − 1.59 Z-scores in the standard of care group [13].

A chief target of the WASH interventions has been to ameliorate the gut damage that is caused by poor living conditions [10]. Our previous work in The Gambia identified persistent and chronic damage to the gut (now commonly termed environmental enteric disease (EED)) as a likely mediating factor in growth failure [24, 25]. Enteropathy impairs nutrient absorption, increases energy and nutrient losses and causes chronic inflammation that suppresses the growth hormone/IGF1 somatotrophic axis [10, 26]. This chronic gut damage affects most children in poor settings and is believed to be initiated and sustained by repeated exposure to microorganisms (and possibly other aggravating factors) in a contaminated environment. A study in Bangladesh has found differences in biomarkers of EED between ‘clean’ and ‘dirty’ households and a 0.54 HAZ difference in growth in favour of the clean households, though it should be noted that average growth was still very poor at − 1.66 HAZ [27].

Overcoming such influences has formed the rationale for numerous WASH interventions. However, a Cochrane systematic review and meta-analysis of such trials published in 2013 concluded that efficacy was poor [28]. The analysis of randomised trials including over 4600 children studied over 9–12 m intervention periods revealed no evidence of any beneficial effect on weight-for-age or weight-for-height and only a marginally significant impact on height-for-age of less than one tenth of a standard deviation (+ 0.08 Z-score, 95%CI 0.00–0.16). An individual participant data analysis (n = 5375) was conducted for 5 RCTs. For HAZ, this identified a significant age interaction with children under 2 years gaining most benefit. Nonetheless, the benefit in this selected age band was still only 0.25 (95% CI 0.14–0.36) Z-scores [28]. The WASH Benefits and SHINE results are even more discouraging insofar as their null outcomes arose from very large, rigorously implemented trials in appropriate target populations [10–12].

One hypothesis about the lack of effects of the WASH interventions in WASH Benefits and SHINE is that they did not target the whole community; they only worked within individual households. Therefore, spillovers from the un-intervened households could still contribute to infection and inflammation in the study children. There is evidence that community-wide total sanitation interventions may have bigger effects [29–31], possibly related to herd protection. A similar effect may be at play within the MRC compound though it should be stressed that there is free movement into and out of the compound with multiple opportunities for cross-exposure between the SES1A children and villagers through social visits, markets, clinics and by sharing of transport.

The emergent consensus is that it has proven extremely difficult to shift child growth patterns by more than one sixth of the total deficit that accrues by age 2 years in LMICs. The Lancet Series on Maternal and Child Nutrition reinforces the conclusion that nutrition-related interventions will have limited impact [32]. The authors estimated that, even if scaled up to 90% coverage, the implementation of all of the currently identified evidence-based nutrition and nutrition-related interventions would eliminate around 20% of stunting globally.

The current study suggests a resolution to this paradox. Our data indicates that there is a very high threshold of living conditions that has to be surpassed for stunting to resolve. Our evidence suggests that despite adequate parental knowledge of the principles of hygiene and good nutrition within SES1B and SES2, and despite income levels allowing access to moderately good housing and virtually any foods, the children of such families still show profound growth faltering in early life. An explanation might lie in two domains. First, there are many fewer animals within the MRC compound and they are kept out of the houses, so it is much easier for parents to maintain their children in a setting with minimal faecal contamination; a factor considered a likely driver of infections and enteropathy [10, 33]. Second, studies have found that the volume of water usage is highly dependent upon time taken to access the water [33]. Even if water is available at an adjacent standpipe requiring just 2 min for collection, water usage has been found to average 20 l per person per day, whereas usage rises to 60 l per person per day when water is piped directly into the home [34]. We propose that the disappointing outcomes from the WASH trials [10–13] have arisen, not because the theoretical premise behind the trials is incorrect, but because the interventions have fallen short of the very high threshold indicated by our current data.

On the basis of these observations, we infer that improvements in housing conditions including, and perhaps especially, the provision of piped water into the home, may be a key factor in explaining why national stunting rates remain so persistent in low-income settings yet decline so markedly as countries pass through the economic transition. This could be tested using cluster-randomised or step-wedge intervention designs of what we term WASH++ (and others are terming ‘Transformative WASH’) and, if true, would lead to a major shift in priorities for implementation of nutrition-specific and WASH-type interventions.