Altitude and its inverse association with abdominal obesity in an Andean country: a cross-sectional study

Background: Abdominal obesity represents an accurate predictor of overall morbidity and mortality, which is worrisome because it is also continuously increasing across Andean countries. However, its relationship with altitude remains unclear. The objective of this study was to assess the association between altitude and abdominal obesity in Peru, and how sociodemographic variables impact this association. Methods: We estimated the prevalence of abdominal obesity in Peru and analyzed its association with altitude using the data from the 2012-2013 National Household Survey (ENAHO). During this survey, a representative sample of Peruvians was screened for abdominal obesity, using waist circumference as a proxy, and the Adult Treatment Panel III guidelines cutoffs. Results: Data were analyzed from a sample of 20 489 Peruvians (51% male). The prevalence of abdominal obesity was estimated at 33.6% (95% CI: 32.5 to 34.6%). In Peru, altitude was significantly and inversely associated with abdominal obesity, decreasing with higher altitudes: 1500-2999 meters above mean sea level (MAMSL) vs <1500 MAMSL, adjusted prevalence rate [aPR]= 0.90 (95% CI: 0.84 to 0.96); ≥3000 MAMSL vs <1500 MAMSL, aPR= 0.78 (95% CI: 0.72 to 0.84), when adjusting by age, gender and residence area (rural/urban). However, this association was significantly modified by age and gender ( p< 0.001). Conclusion: Abdominal obesity is highly prevalent in Peru and decreases significantly with altitude, but age and gender modify this association. Thus, abdominal obesity appears to affect older women from low altitudes more than younger men from high altitudes.

report report report report

Introduction
The increasing prevalence of obesity represents a significant public health problem across low-to high-income countries 1 .
The main reason is that obesity is strongly associated with morbidity and mortality, mostly due to type 2 diabetes, cancer and cardiovascular diseases 2 . However, body fat distribution, particularly that of abdominal obesity, has been reported as a better predictor of overall morbidity and mortality than total adiposity or obesity defined by body mass index (BMI) 3,4 . Furthermore, abdominal obesity is difficult to diagnose in routine clinical care because it requires access to computed tomography 5 or magnetic resonance imaging 6 for precise quantification. Anthropometric measures of abdominal obesity include waist circumference, waist-to-height ratio, waist-to-hip ratio, and the conicity index 7,8 . Thus, the most commonly used surrogate to diagnose abdominal obesity in clinical care and research examinations is waist circumference 9,10 .
In Peru, as in most Latin-American countries, the prevalence of obesity among children, adolescents and adults have grown consistently in recent decades. Among Peruvian adults, estimates of the national prevalence of obesity have grown from approximately 9% in 1975 to 21% in 2017 11 . However, this prevalence seems to vary substantially by altitude 12 .
Epidemiological studies carried out in the United States 13 and Peru 12 among adults and children 14 have described an inverse association between altitude and obesity. A previous study reported that the prevalence of obesity in Peru decreases by approximately 26% at between 1500-2999 meters above mean sea level (MAMSL), and by 46% at over 3000 MAMSL, as compared to at 0-499 MAMSL 12 .
Consequently, this study further assesses the association between altitude and abdominal obesity, when adjusted by standard sociodemographic variables. Additionally, we plan to estimate the prevalence of abdominal obesity by different cutoffs.

Study design
The study employed a cross-sectional multistage study design. Data were accessed from the Peruvian National Household Survey (ENAHO), undertaken annually by the Peruvian National Institute of Statistics and Information (INEI) and the National Center for Food and Nutrition (CENAN) to assess social living conditions. For this purpose, the INEI and CENAN surveyed a representative sample of the Peruvian population using a probabilistic, stratified, multi-stage design, independent for each region, to collect data on participants of ≥2 months of age 15 . Briefly, the ENAHO sample household residents from all regions of Peru (third sampling level), sampling clusters of one or more blocks of ~120 houses (second sampling level) and sampling cities with 2000 or more inhabitants in the urban area or 500-2000 inhabitants in the rural area (first sampling level). ENAHO survey eligibility criteria were Peruvian households inhabitants, including family members, non-family members and domestic workers (with or without payment) that cohabitated during the 30 days prior to the survey, excluding pensions of 10 or more inhabitants 15 . In this study, we used ENAHO 2013 data to assess the prevalence of abdominal obesity and its association with altitude, while adjusting for their primary demographics and design effect. Out of 45 164 observations, people aged 20 years or older were included. We excluded pregnant women, and those observations with unreliable data.

Variables of interest
The study outcome was abdominal obesity: we used waist circumference (WC) as a proxy for its diagnosis. During the ENAHO survey, trained personnel measured the subject's WC at the vertical position of the midpoint between the lowest rib and the border of the iliac crest 10 . We interpreted this measurement by using the cutoffs proposed by Adult Treatment Panel III guidelines (ATP III) for abdominal obesity: WC >102 cm for men and >88 cm for women 16,17 . Additionally, for comparison, we assessed the cutoffs proposed by the Latin-American Diabetes Association (ALAD): WC ≥94 cm for men and ≥88 cm for women 18 and that of the International Diabetes Federation (IDF): WC >90 cm for men and >80 cm for women 19 . Furthermore, we define abdominal obesity as a weight to height ratio (WtHR) ≥0.5 20 and obesity as a BMI ≥30 kg/m 2 . For this purpose, WHtR was defined as subject's waist circumference divided by their height, both measured in cms.; and BMI was defined as the body weight (kg) divided by the square of the body height (m 2 ).
To facilitate comparisons and interpretability, we categorized altitude (measured by GPS) as low (<1500 MAMSL), moderate (1500-2999 MAMSL), and high (≥3000 MAMSL). Likewise, individuals were categorized by age as young adults (20-39 years), adults (40-59 years) and elders (≥60 years). We classified the area of residence as rural using ENAHO/INEI standard definition, which define an area as rural or rural town, if has no more than

Amendments from Version 1
We modified the abstract reporting the PR obtained by the adjusted model without interaction terms.
In the introduction, we include more indicators of abdominal obesity. Likewise, we took into account other suggestions.
We included more information about sampling, how we estimated variables and interaction analysis.
In the result section, we corrected the paragraph related to the residence area associated with obesity. In Table 5, we incorporated a column with the adjusted prevalence rate without interaction terms. Finally, we included the references suggested by reviewers.
Any further responses from the reviewers can be found at the end of the article REVISED 100 contiguous households grouped or have more than 100 households scattered or disseminated without forming blocks or cores 15 . Nutritional status was assessed by BMI and categorized using WHO standard cutoffs as underweight (<18.5 kg/m 2 ), normal (18.5-24.9 kg/m 2 ), overweight (25-29.9 kg/m 2 ), and obese (≥30 kg/m 2 ) 21 .

Statistical analysis
We estimated the prevalence of abdominal obesity by considering survey sampling weights by using STATA survey (svy) commands and excluding registers with missing study outcomes. We assessed bivariate correlation by estimating the Spearman's rank-order correlation coefficient. Considering that the prevalence of abdominal obesity in Peru is not rare 11 , we estimated the adjusted prevalence ratio as a measure of association instead of the odds ratio 22 . Thus, we used a log-binomial regression model that has robust variance, rather than a Poisson regression model, to adjust our prevalence ratio estimates by gender, age group and area of residence 23 . Finally, we tested for interaction between gender and altitude, and between age and altitude using the Wald test because of the consensus that obesity prevalence vary by gender and age 3 . All statistical analyses were performed using STATA/MP 14.0 for Mac (StataCorp LP, College Station, TX), and the results of statistical tests were interpreted and summarized with 95% confidence intervals.

Ethical statement
According to the Regulation of Ethics in Research of the Peruvian National Institute of Health, this study did not require approval or exemption from an ethics committee because the database is publicly available. Study dataset was published using Figshare, which requested to hide subjects' age and to strictly limit the data availability to only the variables analyzed in this study.

Characteristics of the study population
We analyzed a population sample of 20 489 subjects from 703 different locations across 25 administrative regions of Peru. To summarize population demographics, most subjects were either female (51.6%), adults between 20 to 39 years of age (39.8%), or inhabitants from urban areas (79.6%) 24 . Of these three demographic measures, both age groups (p=0.0006) and area of residence (p<0.0001) distribution varied significantly by altitude ( Table 1).

Variability of abdominal obesity by different cutoffs in Peru
Estimates of abdominal obesity prevalence vary significantly with altitude and in models that use different standard diagnostic cutoffs. When comparing the estimated prevalence of abdominal obesity using ATP III, ALAD and IDF cutoffs (Table 1 and Figure 1), there were significant differences between them (p<0.001 at each paired comparison). The same variability was observed regardless of age group, gender, and residence area (Table 2). Furthermore, in the correlation analysis (Table 3), we found that using the ATP III cutoff resulted in a stronger correlation with obesity by BMI (Spearman´s ρ = 0.55; p<0.001), as compared with the ALAD (Spearman´s ρ = 0.53; p<0.001) and IDF cutoffs (Spearman´s ρ = 0.37; p<0.001). However, the ATP III cutoff also has a weaker correlation with altitude (Spearman´s ρ = 0.12; p<0.001). Additionally, we found that the prevalence of abdominal obesity, as defined by WtHR >0.5, has only a moderate correlation with the prevalence of obesity by BMI (Spearman´s ρ = 0.43; p<0.001) and a weak correlation with altitude ( Table 3).

Variability of abdominal obesity by altitude in Peru
The prevalence of abdominal obesity and obesity vary significantly by altitude in Peru and are inversely associated with altitude category (trend analysis p<0.001 for both), regardless of age group, gender and residence area (Table 4). Both abdominal obesity and obesity prevalence were significantly higher among females than males (p<0.001 for both) and across urban areas than in rural areas (p<0.001 for both). The prevalence of obesity and abdominal obesity were significantly lower among young adults (20-39 years) than among adults (40-59 years); however, both obesity and abdominal obesity prevalence were significantly higher in young adults than elders (≥60 years old).

Abdominal obesity and its association with altitude in Peru
Regression analyses demonstrated that the prevalence of abdominal obesity was significantly associated with altitude when either unadjusted and adjusted by age groups, gender, and residence. Additionally, we observed significant effect modification   of this association by age group and gender, which seems to be particularly high at altitudes over 3000 MAMSL. Once adjusted by the interaction terms, the association between abdominal obesity and altitude varies significantly by gender, age group and residence area, with different patterns of distribution at different altitudes. At lower altitudes (<1500 MAMSL), the prevalence of abdominal obesity exhibits a positive trend increasing by age group, while above 1500 MAMSL, it exhibits an inverted-u shaped relationship (Figure 2).

Abdominal obesity and its associated factors in Peru
In the regression analysis, we found that altitude, age groups, gender, and residential area were significantly associated with the prevalence of abdominal obesity in Peru (Table 5). Based on our multivariate regression analysis outputs, we observed that the prevalence of abdominal obesity decreased with altitude, increased with age, and is lower among male and rural populations. However, contrary to what was observed for the prevalence of abdominal obesity by altitude in the case of gender and residence area, both of which decrease with altitude, the variability of the prevalence of abdominal obesity by age group exhibits different patterns of distribution at different altitudes.

Discussion
The prevalence of abdominal obesity in Peru is high and decreases with altitude, an association that is modified by age and gender. This prevalence was higher among women over 60 years of age below 1500 MAMSL, and lowermost among men 20 to 39 years of age over 3000 MAMSL, exhibiting an inverted-u shaped relationship. Understanding the intricacies of this association is critical in countries with high elevation such as Peru, where approximately 20% of the Peruvian population lives at or above 3000 MAMSL 25 .
The usefulness of WC as an indicator of abdominal obesity is quite clear; however, there is a permanent discussion regarding the cutoffs for its diagnosis. WC varies by ethnic groups, which has generated the recommendation that each country or

Figure 2. Prevalence of abdominal obesity (by Adult Treatment Panel III guidelines cutoffs) by age group and altitude.
The association between abdominal obesity and altitude vary greatly by gender and age group, which behave as effect modifiers.
region produces its cutoffs 26 . Worldwide, the most used cutoffs for WC are the ones proposed by the ATP III, which are primarily specific for adult European Caucasian populations 16,17 .
There are some efforts in Latin America to propose WC cutoffs for their population. A recent study carried out in five Latin American countries recommended using cutoffs of 90-92 cm for women and 94 cm for men 27 . In Peru, the PREVEN-TION study proposed WC cutoffs at high altitude (~2600 MAMSL) of 87 cm for women and 97 cm for men based on abnormalities of intima-media thickness and cardiovascular manifestations 28 . Similarly, different countries have proposed their cutoffs for WC, including Portugal (91 and 97 cm) 29 , China (80 and 84 cm) 30 , and South Asian countries (84 and 88 cm) 31 . In our study, different cutoffs produced a wide range of estimates for the prevalence of abdominal obesity. We observed that when using ATP III cutoffs, the estimated prevalence of abdominal obesity was over three times higher among women than in men (51% vs 15%).
Furthermore, regardless of altitude, these differences seem to be even larger ≥3000 MAMSL (40% vs 7%). These differences are similar to those reported previously 32 , so we believe they can be explained by both the altitude effect and the cutoffs itself, which are gender-differentiated. Further studies are needed to assess the necessity of specific cutoffs corrected by altitude, gender, and age.
Another important finding of our study is that the prevalence of abdominal obesity varies significantly between urban and rural areas, a difference that remains consistent at different altitudes. As reported elsewhere, the prevalence of abdominal obesity in Peru is higher in urban areas than in rural areas 33 , but also shows a slower increase in time in WC compared to rural areas 34 . However, such a difference between urban and rural areas seems to increase with higher altitudes, ranging from 1.7:1 at <1500 MAMSL to 2.1:1 at ≥3000 MAMSL. This finding is relevant in countries with large populations living over 3000 MAMSL, due to the cardiovascular risk that this could imply.
Regardless of WC cutoffs utilized, the mean WC in the Peruvian population living at high altitudes is high. In our study, at >3000 MAMSL the mean WC among men was 87.1 cm and among women 86.0 cm, which are lower than those reported at ~3 600 MAMSL in La Paz-Bolivia (93 cm in women and 93 cm in men) 35 and close to those reported at ~3 660 MAMSL in Tibet (84.5 cm overall) 36 .
According to our results, by both WC and WtHR, Peruvians who live at higher altitudes have a lower prevalence of abdominal obesity than those living a lower altitude. This finding concurs with previous reports 12,37 ; moreover, a higher percentage of overweight (36.3% vs 25.3%), obesity (17.5% vs 8.5%), hypercholesterolemia (18.9% vs 14.6%), low HDL (45.7% vs 40.3%), hypertension (9.8% vs 3.9%) and glycemia >126 mg/dL (2.9% vs 0.9%) were observed in people living above 3000 MAMSL vs below 1000 MAMSL 37 . Overall, the lower cardiovascular risk observed at higher altitudes could be explained in part by the lower levels of urbanization and income, commonly reported in developing countries 38 . Also, it might be explained by the variability in the progress of the epidemiological transition in Peru observed at different altitudes 39 .
It is important to highlight that a WtHR >0.5 seems to overestimate Peruvian abdominal obesity. Regardless of the evidence 20 , if we use a cutoff of 0.5, over 80% of the Peruvian population is classified as having abdominal obesity. Further studies are needed to assess the usefulness of such an indicator in Latin-American countries such as Peru.
We should mention as a limitation that the ENAHO is a crosssectional survey that was meant to represent Peru's nutritional status, and the sample might not represent all altitudes of the country. Likewise, it is essential to emphasize that Peru is one of the few countries with many large populations over 3000 MAMSL. Therefore, the association between altitude and obesity could remain unnoticed at low altitude countries. Another limitation is the absence of variables such as socioeconomic status, education level, physical activity and diet. However, the area of residence (urban and rural) is a variable that encompasses socioeconomic and educational aspects in our country.
In conclusion, our study found that abdominal obesity is highly prevalent in Peru and that abdominal obesity varies substantially by altitude, age, gender, and urbanization. Overall, the prevalence of abdominal obesity decreases with altitude, but age and gender modify such association; abdominal obesity seems to affect older women from low altitudes more than younger men from high altitudes. These findings should help to guide interventions to reduce Peruvian's cardiovascular risk, which should be a matter of more significant concern in future years. Peruvian population prevails people with short stature), the waist waist index, the conicity index, which are usually more accurate than mere measurement than the abdominal circumference. I recommend that this investigation be continued in a larger sample , that it is even necessary to evaluate risk factors such as physical activity and diet, which are major confounders.

Data availability
Otherwise I catalogue the article as excellent.
Is the work clearly and accurately presented and does it cite the current literature? Partly

Are sufficient details of methods and analysis provided to allow replication by others? Yes
If applicable, is the statistical analysis and its interpretation appropriate? Yes Page 6, second paragraph: it is stated that abdominal obesity and obesity prevalence were higher across rural areas than in urban areas; possibly the contrary should be reported. In Discussion it is reported that differences in socio-economic status might be captured by the urban/rural variable. Are there also differences in ethnic background by altitude?
Page 9, findings are summarized of a previous study reporting a higher percentage of overweight and other cardiovascular risk factors in people living above 3000 m vs. those living below 1000 m. However, these results seem to be in contrast with the present study; please explain.

If applicable, is the statistical analysis and its interpretation appropriate? Partly
Are all the source data underlying the results available to ensure full reproducibility? Partly

Are the conclusions drawn adequately supported by the results? Yes
No competing interests were disclosed.

Antonio Bernabe-Ortiz CRONICAS Center of Excellence in Chronic Diseases, Cayetano Heredia University, Lima, Peru
Overall comment: This paper looks for the association between altitude and abdominal obesity. Although the idea is not novel, extra reports using existing data can help to understand better this. I would expect more details regarding the variables (exposure and outcome) used in the analysis. Moreover, any other kind of models, not that using only three categories for altitude would be more relevant, or verifying if the association is similar for overweight and obesity. This analysis can be improved by looking extra ways to see the association more than the traditional form to check that, for example, using linear regression if possible or polynomic models (quadratic at least).

Major concerns:
Abstract: Please change it accordingly to comments below.
Introduction: Obesity has different indicators and in the reference is used similarly for BMI (overall obesity), waist circumference (abdominal obesity), etc. Please be careful and consistent with words used. Some longitudinal studies have been published showing the association of interest and they have not been considered here (e.g. PMID: 29472520 ). What is the novelty in this paper? Is it altitude? Apparently not as shown in the longitudinal paper... is it the rurality? Usually rural areas are in high altitudes. Is waist circumference an easy measure to do? Is it routinely done? How the usefulness of waist circumference in high altitudes is evaluated in this paper? I think that is no part of this study as pointed out in the last part of the first paragraph. Any reference for the last sentence of the second paragraph? Third paragraph: is the risk of obesity decreasing due to altitude? The reference is a cross sectional study, so is it possible to talk about risk? Methods: Study design: why this study is multicentric? How many centers were included? Please explain. If the ENAHO is conducted yearly, why to use the 2013 ENAHO? Why not to use the last one available? Please explain how the sampling was done? Stratified by what? How many stages does the sampling have? Was any criterion related to time living in high altitude (e.g. at least 6 months living in the city/area)? Are 30 days enough to see the potential impact of altitude on health? Why households with 10 or more inhabitants were excluded? Any explanation? What proportion of households in Peru has 10 dwellers or more? Are these decisions biasing results? How altitude was measured? Was GPS used for this? Only a simple calculation of the altitude of the city was used? Please explain. Usually, 2500 meters is used as the cutoff and not 3000... Since this variable is the main exposure, details should be given to understand if any misclassification could be introduced... Only those aged 20 years and more were included in the analysis? How about those between 18 to 20 years or those younger? How the rural index was built? Explain please... Any reference helping to understand this? I am pretty sure the ENAHO stratifies the sample by urban/rural settings... was not the case this time?
Categorization of BMI is presented in different way compared to how it was analyzed... please be 1