Self-reported Health is Related to Body Height and Waist Circumference in Rural Indigenous and Urbanised Latin-American Populations

Body height is a life-history component. It involves important costs for its expression and maintenance, which may originate trade-offs on other costly components such as reproduction or immunity. Although previous evidence has supported the idea that human height could be a sexually selected trait, the explanatory mechanisms that underlie this selection are poorly understood. Despite extensive studies on the association between height and attractiveness, the role of immunity in linking this relation is scarcely studied, particularly in non-Western populations. Here, we tested whether human height is related to health measured by self-perception, and relevant nutritional and health anthropometric indicators in three Latin-American populations that widely differ in socioeconomic and ecological conditions: two urbanised populations from Bogota (Colombia) and Mexico City (Mexico), and one isolated indigenous population (Me’Phaa, Mexico). Results showed that self-reported health is best predicted by an interaction between height and waist circumference: the presumed benefits of being taller are waist-dependent, and affect taller people more than shorter individuals. If health and genetic quality cues play an important role in human mate-choice, and height and waist interact to signal health, its evolutionary consequences, including cognitive and behavioural effects, should be addressed in future research.

indicator/SI.POV.GINI). These national-level statistics, however, hide important within-country differences. In particular, Latin-American people in rural areas tend to be poorer and have less access to basic services such as health and education than people in urban areas.
According to data from the World Bank and the Colombian National Administrative Department of Statistics, in 2017 Colombia was the second most unequal country in Latin-America after Brazil. In rural areas, 36% of people were living in poverty and 15.4% in extreme poverty, while in urban areas, these values were only 15.7% and 2.7%, respectively 48 .
In addition to rural communities, in Latin-America indigenous people tend to have high rates of poverty and extreme poverty 49 , and poorer health 50 , which is less susceptible to improvement by national income growth 51 . In Mexico, there are at least 56 independent indigenous peoples whose lifestyle practices differ in varying degrees from the typical 'urbanised' lifestyle. Among these groups, the Me'Phaa people, from an isolated region known as 'Montaña Alta' of the state of Guerrero, is one of the groups whose lifestyle most dramatically differs from the typical Westernised lifestyle of more urbanised areas 52 . Me'Phaa communities are small groups of indigenous people, composed of 50 to 80 families, each with five to ten family members. Most communities are based largely on subsistence farming of legumes such as beans and lentils, and the only grain cultivated is corn. Animal protein is acquired by hunting and raising some fowls, and meat is only consumed during special occasions but not as part of the daily diet. There is almost no access to allopathic medications, and there is no health service, plumbing or water purification system. Water for washing and drinking is obtained from small wells. Most of the Me'Phaa speak only their native language 53 . In consequence, these communities have the lowest income and economic development in the country, and the highest child morbidity and mortality due to chronic infectious diseases 52 .
These three Latin-American populations can provide an interesting indication about how the regional socioeconomic conditions and intensity of ecological pressures by pathogens may modulate the function of height as an informative sexually selected trait of health and individual condition in each sex. Therefore, the aim of the present study was to evaluate whether human height is related to health measured by self-perception, and relevant nutritional and health anthropometric indicators in three Latin-American populations that widely differ in socioeconomic and ecological conditions: two urbanised populations from Bogota (Colombia) and Mexico City (Mexico), and one isolated indigenous population (Me´Phaa, Mexico). In addition, given the possible immunological effects of testosterone, and that men present higher levels than women, we predicted this relation to be different between sexes in all studied populations. Therefore, we propose that height would be a stronger signal of self-reported health condition in men compared to women. The first sample consisted of 63 subjects (mean age ± standard deviation [SD] = 33.63 ± 9.69 years old) from the small Me'Phaa community -'Plan de Gatica' from a region known as 'Montaña Alta' of the state of Guerrero in Southwest Mexico. In this sample, 24 participants were women (33.46 ± 8.61 years old) and 39 participants were men (33.74 ± 10.41 years old), who participated in a larger study on immunocompetence. Both sexes were aged above 18 years old. In Mexico, people above 18 years old are considered adults. All measurements were collected in the participants' own community. Me'Phaa communities are about 20 km apart, and it takes about three hours of travel on rural dirt roads to reach the nearest large town, about 80 km away. Mexico City is about 850 km away, and the trip takes about twelve hours by road. This community has the lowest income in Mexico, the highest index of child morbidity and mortality by gastrointestinal and respiratory diseases (children aged 0 to 8 years had the highest vulnerability and death risk 52 ), and the lowest access to health services. These conditions were recorded in the National Health Information System 2016 52 .

Methods
The second sample consisted of 60 subjects of over 18 years old (30.27 ± 8.56 years old) from the general community in Mexico City, of whom 30 were women (37.47 ± 5.61 years old) and 30 were men (23.07 ± 3.22 years old). Finally, the third sample consisted of 354 undergraduate students with ages ranging from 18 to 30 years old (20.39 ± 2.10 years old), 184 were women (20.16 ± 2.08 years old), and 170 were men (20.64 ± 2.10 years old) from Bogota, Colombia. All urban participants were recruited through public advertisements.
Participants from both urban population samples were taking part in two separate, larger studies in each country. In Colombia, all data were collected in the morning, between 7 and 11 am, because saliva samples (for hormonal analysis), as well as voice recordings, body odour samples, and facial photographs, were also collected as part of a separate project. Additionally, women in the Colombian and Mexican samples were not hormonal contraception users, and all data were collected within the first three days of their menses.
Participants who were under allopathic treatment and hormonal contraception users from both countries were excluded from data collection. All participants completed a sociodemographic data questionnaire, which included medical and psychiatric history. No women were users of hormonal contraception. Although no participant reported any endocrinological or chronic disease, these health issues were also considered as exclusion criteria.
Given that the indigenous community of 'Plan de Gatica' consists of 60-80 families, each with five to seven members, the final sample for this study could be considered as semi-representative of a larger Me'Phaa population inhabiting in the same community. Nevertheless, the total population of Me'Phaa people inhabiting the 'Montaña Alta' is comprised of 20-30 communities with almost the same number of families as 'Plan de Gatica' . Therefore, it is important to mention that our sample size cannot be considered representative of the total Me'Phaa people inhabiting the 'Montaña Alta' region, but from the specific 'Plan de Gatica' community. Similar condition occurs for participants from the Mexico City and Bogota samples. These participants were recruited at the National Autonomous University of México and Universidad El Bosque campuses, respectively. Therefore, these samples are comprised mostly of bachelor and graduate students, and cannot be considered as representative of a large population of the whole city, which is comprised of about 12 million adult persons in Mexico and about 5 million adults in Bogota.
procedure. All participants signed the informed consent and completed the health and background questionnaires. For participants from the indigenous population, the whole procedure was carried out within their own communities, and participants from the Mexican and Colombian urban population attended a laboratory at either the National Autonomous University of México or Universidad El Bosque respectively, on individual appointments.
Participants from Mexico City and Bogota were recruited through public advertisements on social media and poster boards located along the central campus of the National Autonomous University and Universidad El Bosque. While in Mexico City, participants received either one partial course credit or a payment equivalent to $5 dollars as compensation for their participation, all participants in Bogota were given academic credits for their participation.
For the indigenous groups, recruitment was done through the Xuajin Me'Phaa non-governmental organisation, which is dedicated to the social, environmental and economic development for the indigenous communities of the region (see video from this organisation, https://youtu.be/WOEcGUHjR9Q). Xuajin Me'Phaa has extensive experience in community-based fieldwork and has built a close working relationship with the community authorities. The trust and familiarity with the community customs and protocols have previously led to successful academic collaborations 53,54 . Therefore, Xuajin Me'Phaa served as a liaison between the Mexican research group and the communities for the present study, offering mainly two important factors in data collection: the informed consent of community members and participants, and two trained interpreters of Me'Phaa and Spanish language of both sexes.
First, participants were asked to complete the health and sociodemographic data questionnaires. Subsequently, the anthropometric measurements were taken. The interpreters provided by the Xuajin Me'Phaa organisation administered the SF-36 Health survey in Me'Phaa language. Interpreters used Spanish as the second language and are thoroughly proficient in speaking and reading Spanish. We used the validated SF-36 survey for urban and rural Mexican populations 55 for interpreters to translate Spanish to Me'Phaa language. Given the ethnical customs of Me'Phaa culture, the participants were always interviewed by an interpreter of the same sex to avoid bias in participant responses; for instance, men were interviewed by a male interpreter and women by a female interpreter. The same interpreter interviewed all participants of his/her corresponding sex.
For the present study, both urban and indigenous participants only answered items corresponding to the dimension defined as general health (i.e. Item numbers 1, 33, 34, 35 and 36), except for item 35. This item informs about the expectation for future health. Since the grammatical compositions of Me'Phaa language do not consider 'infinitive' and 'future' as verbal tenses 56 , an interpretation of this question was not possible for the Me'Phaa people, therefore, this item was excluded. In Colombia, we used a Spanish version of the SF-36 questionnaire 57 , that was previously validated in the same country 58 .
To obtain the self-reported health rate, all items were recoded following the instructions on how to score SF-36 57 . We calculated the final factor by averaging the recoded items. To make this data compatible with the Mexican database, item 35 was excluded because it cannot be answered by the Mexican Indigenous population, and the general health dimension was calculated by averaging only items 1, 33, 34 and 36.
Anthropometric measurements. All anthropometric measurements were measured thrice and subsequently averaged to obtain the mean value (for agreement statistics between the three measurements of each characteristic, see section 1.3 in the Supplementary Material). All participants wore light clothing and had their shoes removed. The same observer repeated the measurements thrice.
We measured the body height in cm, to the nearest mm, by using a 220 cm Zaude stadiometer, with the participant's head aligned according to the Frankfurt horizontal plane, and feet together against the wall.
Anthropomorphic measurements also included waist circumference (cm), weight (kg), fat percentage, visceral fat level, muscle percentage and body mass index (BMI). The waist circumference was measured midway between the lowest rib and the iliac crest in cm by using a flexible tape and was recorded to the nearest mm. These www.nature.com/scientificreports www.nature.com/scientificreports/ anthropomorphic measurements have been used as an accurate index of nutritional status and health, especially waist circumference. Metabolic syndrome is associated with visceral adiposity, blood lipid disorders, inflammation, insulin resistance or full-blown diabetes and increased risk of developing cardiovascular disease 59-61 , amongst Latin-American populations 62 . Waist circumference has been proposed as a crude anthropometric correlate of abdominal and visceral adiposity, and it is the simplest and accurate screening variable used to identify people with the features of metabolic syndrome 63,64 . Hence, in the presence of the clinical criteria of metabolic syndrome, increased waist circumference provides relevant pathophysiological information insofar as it defines the prevalent form of the syndrome resulting from abdominal obesity 60 .
Weight, fat percentage, visceral fat level, muscle percentage, and BMI were obtained using an Omron Healthcare HBF-510 body composition analyser, which was calibrated before each participant's measurements were obtained. Statistical analysis. We used linear models (LM) to test the association between height and self-reported health. The dependent variable in this model was the health factor and predictor variables included participant sex, age, sample (Bogota, Mexico City, Me'Phaa), height and waist and anthropometric measurements (hip, weight, fat percentage, BMI and muscle percentage) as fixed, main effects, as well as all possible interactions between height, waist, sample, and sex. For all models, the continuous regressors involved in interactions (waist and height) were centred.
Although sample could be thought as a random factor (i.e. fitting linear mixed models instead), we treated it as a fixed-effects categorical predictor in the models because there were only three levels (Bogota, Mexico City, Me'Phaa), and a minimum of five levels is recommended. To test the residual distribution, generalised linear models (GLM) were fitted, but in all cases, residuals were closer to a normal or gamma (inverse link) distribution, for each sample. Models here included were fitted using the lm function in R, version 3.6.1 65 .
The most parameterised initial model (Model 1) was then reduced, by excluding the main effects of hip, weight, fat percentage, visceral fat, BMI and muscle percentage (as these are phenotypic markers associated either with height or waist circumference), and keeping the main effects of age, as well as the main effects and all possible interactions between any combination of height, waist, sample, and sex, consistent with our predictions. This, still highly parameterised model (Model 2), was further reduced using the functions dredge (https://www. rdocumentation.org/packages/MuMIn/versions/1.43.6/topics/dredge) and model.sel (https://www.rdocumentation.org/packages/MuMIn/versions/1.43.6/topics/model.sel) from the package MuMIn: Multi-Model Inference 66 . The dredge function fitted a set of 334 models with combinations (subsets) of fixed-effect terms from the second model, that were then compared using the function model.sel based on the Akaike Information Criterion (AICc) and Akaike weights, allowing us to select the best model (Model 3). This, best-supported model (i.e. the model with the lowest AICc with a ΔAICc higher than two units from the second most adequate model), is reported 67 .
Finally, we compared the three models selected model (Models 1, 2 and 3) using the ICtab function from the bbmle package 68 . Once a final model was selected, model diagnostics were performed (collinearity, residual distribution, and linearity of residuals in each single term effect; see section 3.3 in the Supplementary Material).
Interactions in the final model were explored and via simple slopes analysis and Johnson-Neyman intervals 69

Results
All data and code used to perform these analyses are openly available from the Open Science Framework (OSF) project for this study (https://osf.io/5rzfs/). Descriptives. Descriptive statistics of age, waist circumference, hip, height, weight, fat percentage, visceral fat, BMI, muscle percentage and self-reported health and reported in Table 1.
The distribution of all measured variables is shown in Fig. 1. Age, waist, height, visceral fat, and self-reported health strongly varied in both women (Fig. 1a) and men (Fig. 1b)  Models to predict self-reported health. To establish the relationship between height and self-reported health, we fitted three linear models (Table 2). For all models, the continuous regressors involved in interactions (waist and height) were mean-centred.
In the first model (Model 1), we included as predictors all measured variables as main effects, as well as all interactions between height, waist circumference, sample, and sex. The first model was initially reduced by excluding hip, weight, fat percentage, BMI and muscle percentage. We decided to include waist circumference instead of visceral fat or fat percentage for two reasons: first, because these three variables are strongly correlated in women and men (r > 0.79 in all cases; see Supplementary Tables S3 and S4,  www.nature.com/scientificreports www.nature.com/scientificreports/ In the second model (Model 2), we therefore included age, height, sample, sex, waist circumference, and all possible interactions between combinations of height, waist circumference, sample, and sex. This second model was further reduced by the implementation of the functions dredge and model.sel from the package MuMIn 66 (for details, see the Statistical analysis section in the Methods). These functions fitted and compared a total of 334 models with different combinations of fixed terms from Model 2; these compared models and their relative probability to be the best model are shown according to their relative Akaike weights (w i (AICc)) in Fig. 2.
This analysis revealed that the best model (labelled 159 in Fig. 2), included height (centred), sample, sex, waist circumference (centred) and the interaction between height (centred) and waist (centred). However, to account for the age differences between samples, we selected the second-best model (labelled 160 in Fig. 2), because it also included age as a regressor, and had a ΔAICc of less than 2 units (≈0.98) compared to the best model. This model, including age, was therefore selected as our final model (Model 3).
The three selected models were compared using the AICc, Akaike weights (w i (AICc)) and ΔAICc ( Table 3). The analyses revealed that Model 3 is not only the most parsimonious of the three selected models, but has higher R 2 adjusted and F values (Table 2), as well as a lower AIC and higher Akaike weight 67 (Table 3) than the previous two models. In fact, Model 3 is close to 464,686 times more likely to be the best model compared to Model 2, and about 35,141,683 times compared to Model 1 (Model 2, was around 76 times more likely compared to Model 1).
Furthermore, for Model 3 (the final, minimum adequate model), Generalised Variance Inflation Factors (GVIF) 66 revealed no concerning cases of collinearity for any of the predictor terms (GVIF ≤3, and a GVIF 1/ (2 × Df) ≤1.6 in all cases; for details, see Supplementary Table S8; residual distribution by sample and linearity in each single term factor are shown in Supplementary Fig. S2).
Moreover, a significant interaction between waist and height (Table 4; t = −2.30, p = 0.022) was revealed, indicating that the negative association of waist circumference with self-reported health was height-dependent (Fig. 3b); the best predicted self-reported health was for tall participants with small waists, and while the  www.nature.com/scientificreports www.nature.com/scientificreports/ association between height and self-reported health is positive for people with small waist circumferences, it decreases for people with increasingly large waists. Furthermore, the Johnson-Neyman procedure 69,70 (Fig. 3c), indicated that height is only a significant, always positive, predictor of self-reported health for people with relatively small waist circumferences of less than 73.51 cm (centred: −4.51).
This interaction was replicated when fitting an alternative version of Model 3 (Model 3 A), replacing waist circumference for visceral fat, by following the same method to select it (i.e. fitting an alternative Model 2, and repeating the same selection process; see section 4.3 in the Supplementary Material). Similar to Model 3, this alternative Final Model, also included an interaction between height and, in this case, visceral fat, in which height was found to be a positive significant predictor of self-reported health, only for people with low levels of visceral fat (see Table S9 and Fig. S3 in the Supplementary Material). Furthermore, in this model the interaction was more extreme than when using waist circumference, and height becomes a significant, negative predictor of self-reported health for people with high visceral fat (see Fig. S3c in the Supplementary Material).

Discussion
The present study provides new insights into the relationship between height and health in men and women by studying three Latin-American populations, which included urban and indigenous populations with marked differences in access to basic needs and services like food and health.
Contrary to our initial hypothesis, height was not a significant predictor of self-perceived health but interacted with waist circumference. Most results in favour of a direct relationship between height itself and health were carried out in small modern populations and specific Western ethnic groups more than twenty years ago. New studies with non-traditional population groups have failed to verify the positive relationship between height and health, especially associated with cardiovascular and autoimmune diseases 72, 73 . For example, studies on Native Americans, Japanese, Indians, and Pakistanis showed that shorter people had a lower prevalence of cardiovascular disease than the tallest people in each population 73 . These findings were similar in Sardinian inhabitants, a European population with the lowest physical stature recorded in Europe in recent years 72 .
Interestingly, our results suggest that there is a main negative effect of waist circumference on self-perceived health. This is congruent with a broad range of studies done in different human populations 74 . In fact, waist circumference has been proposed as one of the most important biomarkers of metabolic syndrome that predicts health condition in terms of cardiovascular diseases 60 . Nevertheless, we found that the positive association between height and self-reported health was waist-dependent in our studied samples. This is, height predicted self-reported health differently for people of different waist circumferences: while being taller predicts better  20  30  40  50  60  3  6  9  15  20  25  30  35  40  20  25  30  25  50  75  100   20  30  40  50  60  80  100  80  90  100  110  120  140  150  160  170  180  40  60  80  www.nature.com/scientificreports www.nature.com/scientificreports/ self-reported health for individuals with relatively small waists, this association was not found in people with larger waist circumferences. Furthermore, while there is a cost of abdominal and visceral adiposity for tall people, there is no predicted cost for shorter persons. Interestingly, epidemiological studies have widely implemented an integration of both phenotypical components in the form of waist to height ratio (WtHR). In general, waist circumference has a stronger negative impact on the health of short individuals than on tall ones 75 , contrasting with our results. These differences might be due to WtHR has been mainly used to predict health in terms of metabolic and cardiovascular diseases (CVD), while our study used a general status of health, which could include more than metabolic and CVD. In addition, we use these phenotypic variables as continuous and independent predictors because the aim of our study argued that human height by itself would be an honest indicator of general health, which we would not be able to evaluate with WtHR as predictor. Therefore, our results argue for the importance of considering a phenotypic independent integration of different human features that could be involved in health or physiological conditions, when a possible sexually selected trait is being evaluated as a signal of individual condition.
On the other hand, given that height is the human anatomical feature most sensitive to environmental and socioeconomic conditions 18,38 , we expected a stronger association between health and height for the indigenous population where the cost to produce and maintain this costly trait is greater than for inhabitants from urbanised areas. Nevertheless, we did not find inter-population differences in the magnitude of this relation. Urban populations reported better health than the indigenous population, and the shortest participants tended to be from the indigenous Me'Phaa sample. These results could, in fact, suggest different life-history strategies. Compared with modern Western societies, different life strategies could take place in harsh environments 76 , for instance, investing   Table 3. Performance criteria of the three selected models. Note. Models are in descending order from the best to the worst fitting. ΔAIC is the change in AIC between each model and the best model. Akaike weights (w i (AICc)) are conditional probabilities for each model being the best model 67 . relatively less energy in growth and reallocating it towards reproduction 18 . In addition, a relative increase in the intensity or number of infectious diseases (including paediatric diseases in Me'Phaa) and higher tendency to early sexual maturity could negatively impact growth, resulting in a lower average height 77,78 . These trends could be compensations between life-history components 35 . Finally, fast and prolonged growth implies high costs for the organism 1 . Rapid growth may influence mortality risk 79 and growing for a longer time delays the onset of reproduction, increasing the risk of death and producing fewer offspring 1 . This perspective of life strategies allows us to understand the relationship between height, health, and reproduction. This suggests the importance of addressing factors such as ethnicity, socioeconomic status, level of urbanisation in populations where there is great heterogeneity in access to food, health, and pressure from pathogens, for instance, in Latin-American populations in which this relationship has barely been directly explored. Although our results show that height and waist circumference are important predictors of self-perceived health, we did not evaluate any immunological mechanism that may underlie the self-perception responses of the participants. This limitation makes it hard to directly evaluate human height as an honest signal of individual condition. In the present study, the questionnaire used to evaluate general health (SF-36) is far from being a direct indicator of immune condition, since the participant´s perception responses could be influenced by components different than the individual's ability to deal or resist to infectious pathogens, such as skeletal disorders, cancers, cardiovascular or metabolic abnormalities. Nevertheless, studies that have evaluated a more direct approximation of immune condition have led to controversial results. For instance, antibody response to a hepatitis-B vaccine as a marker of immune condition has been positively associated with height in men (but not in women) up to a height of 185 cm, but negatively in taller men 80 . Furthermore, height is uncorrelated with components of innate and adaptive immune system functioning, such as lysozyme activity, neutrophil function, IgA and IgG 81 . One possible explanation for this disparity is that, while participants of the former study 80 lacked any previous exposure to the hepatitis B virus, participants of the latter study 81 were previously exposed to flu and tetanus. Finally, in relation to sex differences, women reported lower average health than men in all communities, which is concordant with reports and normative SF-36 data in other populations, especially in younger people 82,83 . These results could consolidate the idea that height is a reliable signal of health in men 35 , while it could reflect reproductive success in women 84 in terms of labour and birth, and to a lesser extent, function as an indicator of health 85 . It has been seen that taller women experience fewer problems during the labour process due to a lower risk of mismatch between foetal head size and size of the birth canal 85 . Nevertheless, this speculative idea warrants further studies on comparing health, reproductive success and female height.
It is important to consider that the mode of survey administration may be another limitation in our study, and it could have led to confounding effects. For example, it is possible that indigenous people have different understanding and thresholds about their general health perception, which we were unable to evaluate without previous validation of translated items, and it could have explained the lowest values of general health reported by indigenous people. Nevertheless, it could also reflect the real health conditions in Me'Phaa communities and not a misunderstanding of the survey. Other national indicators of health, such as morbidity and mortality by gastrointestinal and nasopharyngeal infectious diseases, have reported that Me'Phaa communities also present the poorest health in Mexico 52 , which is consistent with our results. In fact, items for the dimension of general health perception have the lowest standard deviation and coefficient of variation in the entire SF-36 survey, in both validated Spanish 55,58 and English versions 86 , which makes this dimension the most understandable one.
In addition, in order to consider obvious differences in language and perception of health, statistical models in this study assumed these inter-population variations a priori. The effects of the sample were considered in all performed LMs. We found that although samples differ considerably, the associations between height, waist circumference and self-perceptions of health were predicted to be in the same direction for all populations (i.e. not interacting with the sample).
Finally, we did not have any information regarding potential pregnancy history in women. This is important because each pregnancy can affect waist circumference, so future studies should collect and control or include this variable in all fitted models.
The present study contributes information that could be important in the framework of human sexual selection. If health and genetic quality cues play an important role in human mate-choice 87 , and height and waist interact to signal health, its evolutionary consequences, including cognitive and behavioural effects, should be addressed in future research. This could be done by studying the interaction between waist circumference and height, in relation to reproductive and/or mating success, as well as mate preferences and perceived attractiveness, in populations with both Westernised and non-Westernised lifestyles.

Data availability
All data used for this article are openly available at the OSF 88 . Code to perform all analyses, data manipulation, tables and figures is available in PDF ('Supplementary_Material.pdf ') and R Markdown ('Supplementary_ Material.Rmd') formats, so that it can be fully reproduced and explored in depth 89 .