In Latin America, stunting prevalence fell from 23.7% in 1990 to 13.5% in 2010, a 43% reduction, and prevalence is projected to fall to 10% by 2020 . However, overall progress may hide important within-country inequalities. In the 12 countries included in the present analyses, the median prevalence among indigenous children was 31.9%, being as high as 61.4% in Guatemala.
Several studies from the Latin American region used data from national surveys to compare stunting prevalence among indigenous and non-indigenous children at a given point in time. An analysis of data from the 1990’s in Bolivia, Colombia, Ecuador and Peru had already signaled the higher prevalence of stunting among indigenous than among non-indigenous children, the latter group including afrodescendants . A PAHO report using data from 2002 to 2008 showed similar results from Bolivia, Ecuador, Guatemala, and Peru . The most recent overview provided data collected since 2018 on seven countries (Bolivia, Ecuador, Guatemala, México, Nicaragua, Panamá and Paraguay), confirming the higher prevalence of stunting among indigenous children . Similar patterns were described in single-country reports on Brazil  and Guatemala .
Other authors described time trends in stunting in the indigenous children for Guatemala (1998–2008) , Mexico (1988–2012)  Bolivia (2003–2008) , Brazil (1996–2006)  and Peru (2007–2012) . In all of these studies, stunting prevalence fell for both indigenous and non-indigenous children, yet the former continued to present significantly higher prevalence than the latter. The exception was Brazil where the samples for indigenous children were small in 1996 and 2006, and the differences were not significant . A separate national survey restricted to indigenous children in 2008–2009 showed substantially a higher prevalence of stunting than was observed for non-indigenous children in the 2006 national survey .
None of the above studies reported on analyses that were stratified by ethnicity and either wealth or residence, nor that separated afrodescendants from the reference group, nor that attempted to explain the ethnic gap by adjusting for wealth or residence.
Our multicountry analyses confirmed that, in all countries studied, indigenous children tended to be more stunted than reference children in the crude analyses. In Brazil and Suriname, the differences in stunting prevalence were not significant, but the differences in terms of mean length/height-for-age curves reached statistical significance for all countries suggesting that linear growth faltering among indigenous children is a population-level problem. We also showed that indigenous children were also poorer and more likely to live in rural areas.
Our findings need to be interpreted in light of a conceptual model for explaining the etiology of undernutrition . In such a model, ethnicity represents a distal or structural determinant of health , as it influences intermediate determinants such as income, wealth, education and place of residence, which is confirmed by our results showing that indigenous children tended to be poorer and more rural than other children. At the proximal level of determination, linear growth faltering is influenced by diets, childcare and the incidence of illness, particularly infections [1, 45].
The unadjusted analyses show the full effect of ethnicity, whereas adjusted models show how much of an effect remains after considering wealth and place of residence, which may be regarded as intermediate determinants, driven by ethnicity and, in turn, affecting nutritional status . Even after adjustment for wealth and place of residence, stunting prevalence remained higher among indigenous children in eight of the 13 countries, suggesting that the differences cannot be explained by these two covariates. Moreover, the well-known association between wealth and stunting was observed within each ethnic group .
These findings on stunting are consistent with a biocultural framework which explains how “the devastating effects of colonization, the loss of ancestral land, and language and cultural barriers for access to health care are among the most salient themes characterizing the poor health situation of indigenous people” . There is ample literature on the high rates of several infectious disease among indigenous people [12, 47, 48]. Survey data from several Latin American countries showed lower health care coverage among indigenous women and children than for the rest of the population, even after adjusting for wealth and residence . Lower coverage has been associated with an organization of health services that is insensitive to the need for intercultural health care .
In addition to illness, diet is a key determinant of growth. Our analyses of dietary indicators of children aged 6–23 months showed that although indigenous children were more likely to be breastfed, they tended to have worse complementary diets than the reference group, particularly in terms of dietary diversity. Specifically, indigenous children receive breastmilk with low-quality complementary foods in an age range when breastfeeding alone is insufficient to meet the nutritional needs of the infant for optimal growth. This finding is in agreement with studies on food security in the Latin American region . Our analyses of complementary diets relied upon the 2008 definitions adopted by WHO , which are currently being updated and revised . For example, the updated version of the minimum dietary diversity indicator will require the consumption of five from eight different food groups, as opposed to the present requirement of four out of seven groups. This will lead to lower proportions of children being classified as adequately fed, but differences between ethnic groups will likely remain. Nevertheless, the same criteria were used for all children so that the comparisons could be valid.
Our results reflect the social exclusion pattern that systematically affects indigenous people in Latin American, and that persists in spite of progress in indicators reflecting health and socioeconomic conditions . Discrimination against indigenous populations, affecting their access to health and other services, and their lack of political representation also play a role in explaining the higher prevalence of stunting in this population . Language and culture are important issues as well, being essential components of the identity and worldviews of indigenous peoples that exert a major influence on their health and nutrition . Specifically, language often represents a significant barrier to the communication between indigenous patients and health-care providers, thus restricting access to careseeking and limiting the effectiveness of patient-provider interactions . In short, social exclusion, discrimination, language and culture are key drivers of ethnic inequalities in nutrition and health indicators [10, 11, 54]. Therefore, a comprehensive and inclusive intercultural health model should be a key component of health services in the Latin American countries .
Our comparisons of linear growth in afrodescendants and the reference group showed few differences between the two groups. In five out of the seven countries with data, afrodescendants were more likely to be urban, and in half of these countries they were richer than the reference group. After adjustment for wealth and residence, two countries – Colombia and Ecuador – showed a lower prevalence of stunting in afrodescendants than in the reference children, and no country showed higher prevalence among the latter. The published literature on this comparison is scarce; we were able to find a study from Colombia reporting that adult afrodescendants were taller than individuals classified as indigenous or “others” , and a birth cohort from Southern Brazil where one-year old children born to white and black mothers had similar prevalence of stunting .
Our analyses have a number of limitations. Although standardized child health and nutrition surveys are available for most countries in the Latin American region, several surveys failed to collect information on either anthropometry or ethnicity (i.e., Costa Rica, Dominican Republic and Panama). Moreover, sample sizes for indigenous children were very small, as was the case in Brazil and Paraguay, and some surveys were carried out before 2010, such as in Bolivia, Brazil and Nicaragua.
An important limitation is the nature of the information on ethnicity, which is recognizably an important issue in all studies of ethnicity and health based on self-classification . In Paraguay and Peru, the language spoken at home was used as a proxy variable, but it is possible that indigenous families also speak Spanish. In Paraguay, Guarani is taught in all public schools in the country, so many non-indigenous women speak both Spanish and Guarani. In Brazil the information was on self-reported skin color, with a single specific category for “indigenous. This represents the standard classification endorsed by the Brazilian Institute of Geography and Statistics (https://ibge.gov.br). In addition, due to sample size considerations, a single category was employed for indigenous children, even though in several countries this group includes a broad variety of indigenous communities, each with unique world views, languages, traditions, feeding practices and traditional healing systems. This also applies to afrodescendants to some extent, as there were countries that could have grouped very different cultures into one category . It should be noted that survey sampling frames were not designed specifically with ethnicity in mind. Nevertheless, this does not preclude the comparisons among ethnic groups, just as it is possible to stratify results by wealth or education even though these dimensions did not represent specific sampling domains in surveys.
Due the complexity of ethnic group classifications , the proportions of the survey samples classified in each group might differ from those measured in population censuses or in other surveys. In addition, differential fertility may result in larger proportions of children being classified as indigenous or afrodescendants than would be the case for national censuses that count all individuals. Different types of questions and categories used in each type of survey may lead to the aforementioned discrepancies in the survey samples and the small number of observations when the information is disaggregated into subgroups . Thus, our results must be interpreted carefully, although the fact that our classification showed large and mostly consistent results for indigenous children suggests that the categories included in the analyses are able to discriminate population subgroups.
The quality of anthropometric measurements in routine surveys has been questioned . In Additional file 1: Supplementary Table 3, we show that this is unlikely to have biased our results, because nearly all standard deviations of height for age (expressed in Z scores) ranged from 1.0 to 1.5, which is the expected range in high-quality studies .
Lastly, given intense miscegenation, which is characteristic in the Latin American region, the reference group includes substantial proportions of children with mixed ethnic backgrounds, whose families did not identify themselves as being indigenous or afrodescendants. In spite of the above limitations, the present analyses comprise the most comprehensive report on ethnic differences in linear growth failure and feeding practices across several countries in the Latin American region, relying upon standardized survey methods and analytical approaches.