Geographical differences in osteoporosis, obesity, and sarcopenia related traits in white American cohorts

It has been reported that geographical variation influences bone mineral density (BMD), obesity, and sarcopenia related traits in other countries. However, there is lack of similar studies in the US population. In this study, we compared data from three US study cohorts to evaluate geographical variations of BMD and body composition. BMD, fat mass and lean mass were collected from Dual-energy X-ray absorptiometry machine. ANCOVA and Chi-square tests were used to compare the differences between BMDs, obesity and sarcopenia related traits from different regional sites (Omaha, Kansas City and Baton Rouge/New Orleans). Eta-squared was used to measure the effect sizes of these differences. A total of 11,315 Caucasians from our previous three study cohorts were compared. There was no significant geographical difference in BMD for males or females under the criteria of p-values < 0.05 and effect size η2 > 0.01. There were significant geographical differences with medium effect size (p-value < 0.001, 0.01 < η2 < 0.14) for whole body fat mass percentage and index of low muscle mass. For Caucasians in the United States, there is no significant geographical effect found on BMD. The obesity and sarcopenia related traits are significantly different between the three study cohorts.

Geographical differences of BMD. Table 2 shows the geographical differences of BMD among LOS, KCOS, and OOS cohorts. HIP-BMD, FNK-BMD, and SPN-BMD were compared by the ANCOVA model, controlling for age, height, weight, smoking, alcohol use, and regular physical activity. FNK-BMD and SPN-BMD were not significantly different between genders among the three studies under the criteria of p < 0.05, whereas there were geographical differences for HIP-BMD (p = 0.004 for females and p = 0.003 for males). However, due to the limitations of a null hypothesis test, when a large sample size is used, even a trivial effect could be exaggerated with small p-values 29,30 . In the study, the effect sizes (eta-squared) were reported to determine the regional differences. When HIP-BMD values were compared among the three cohorts, the effect size was smaller than Geographical differences in obesity and sarcopenia traits. For all three body composition traits (BMI, WBFP, and ALMI), we identified significant geographical differences (p-values < 0.001) among the three datasets (LOS, KCOS and OOS), even when adjusted for age, height, weight, smoking, alcohol use, and regular physical activity (Table 3). The geographical patterns for BMI were the same for both female and male subjects (KCOS < LOS < OOS). Although BMI showed significant differences among three groups (p-values < 0.001), the effect size was relatively small (η 2 = 0.004 for female and η 2 = 0.003 for male). WBFP showed the same ranking pattern in all three study regions (KCOS < LOS < OOS) as BMI. For sarcopenia, subjects from KCOS had the highest ALMI value and subjects from LOS had the lowest ALMI value. The effect size of geographical differences was large for WBFP and ALMI (η 2 > 0.01). Figure 1 shows the prevalence of sarcopenia and obesity in the three datasets. The sex-specific cutoff values for sarcopenia defined by ALMI were 7.26 kg/m 2 (male) and 5.45 kg/m 2 (female). The cutoff values for obesity defined by WBFP were 30% (male) and 40% (female). For subjects from KCOS, both males and females had the lowest rates of sarcopenia (4.6% for females and 10.94% for males) and obesity defined by WBFP threshold (16.61% for females and 7.73% for males). Subjects from OOS had the highest percentage of obesity in both male and female subjects, regardless of whether it is defined by WBFP or BMI. When obesity was defined by WBFP, significant differences can be found in each pair of datasets; however, when obesity was defined as BMI ≥ 30 kg/ m 2 , there were no significant differences in prevalence of obesity for subjects between KCOS and LOS (female: 24.37% vs 24.51% and male: 23.13% vs 24.81%). This indicates that it is easier to identify dataset differences using WBFP than using BMI.

Discussion
The purpose of this study was to determine whether there were geographical differences in BMD and body composition of non-Hispanic whites in the United States. After adjusting for age, weight, height, smoking, alcohol usage, and regular physical activity, we found no significant differences among the studies from three geographical regions for HIP-BMD, FNK-BMD, and SPN-BMD, if we use stringent criteria p-values < 0.05 and η 2 > 0.01. Significant geographical differences were found for body composition phenotypes, such as WBFP and ALMI.
To the best of our knowledge, in the past decade, this is the only study to compare the geographical differences of both BMD and body composition of non-Hispanic whites in the United States. Our data showed that there was no regional effect on BMD variation in three selected geographical regions in the United States. This is different from studies in other countries. In previous studies, European between-center differences in BMD were observed in the European Vertebral Osteoporosis (EVO) 9 , the Network in Europe on Male Osteoporosis (NEMO) 10 and the Osteoporosis and Ultrasound study (OPUS) 12 . In a study published in 1997 9 of 16 European populations, both males and females aged between 50 and 80 were recruited randomly from local population registers. BMD values  Table 2. BMD of subjects from LOS, KCOS and OOS (mean ± SD). Note: a NS stands for not significant, p > 0.05. b η 2 is eta-squared effect size of ANOVA. In the study, we consider an effect size larger than 0.01 as significant.  Table 3. Body composition of subjects from LOS, KCOS and OOS (mean ± SD). Note: η 2 is eta-squared effect size of ANOVA. In the study, we consider an effect size bigger than 0.01 as significant.

Gender
were determined by DXA from Hologic, Lunar, or Norland 9 . The study demonstrated that there are major differences between BMD values in European population samples from different geographical sites 9 . In another study conducted in 2008 10 , normal hip and spine BMD data of 1,163 males and 329 females aged 19-29 were collected from 17 centers across Europe with strong evidence of heterogeneity among countries, but not among centers of the same country. In 2009 11 , 10,504 hip BMD scans conducted during 1997-2002 from two population-based studies across Norway were used to validate the geographical difference. The authors found that women ≥60 years in Tromsø had higher age-adjusted BMD than women in Bergen, whereas BMD among women <60 years was similar in Tromsø and Bergen. For men, age-adjusted total hip BMD was lower in Bergen compared with Tromsø. In 2011 12 , proximal femur BMD of females aged 20-39 (n = 258) and males aged 55-79 (n = 1,426) from three European centers were compared. A clear geographical difference in hip BMD did exist; however, there was less evidence for femoral neck BMD 12 . Similar research was also conducted in Canada 13 and Mexico 14 and found geographical variation in BMD. Older survey results 15 (earlier than 2000) showed that although there were differences among regions in the United States (BMD levels of whites tended to be lower in the southern United States), the magnitude of this difference was small. They found that mean total femur BMD levels in the North, Central and West regions were only approximately 0.02-0.03 g/cm 2 , or 1-4%, higher than in the South. In our study, we did not find significant differences in FNK-BMD or SPN-BMD among the three groups. For HIP-BMD, although the p-value was less than 0.05, the effect size was smaller than 0.01. This indicates that the geographic differences had minimal effects on HIP-BMD variation. These controversial results can be explained by the adjustment of lifestyle factors in our study, which include smoking, exercise, and alcohol use. These factors were significantly different among the cohorts and may affect BMD variation [32][33][34] .
In this study, we found that geographical locations were associated with WBFP and ALMI, which is used to investigate obesity and sarcopenia, respectively. Our WBFP results were consistent with a previous meta-analysis of literature data from global studies 35 . It showed that European Caucasians had higher (P < 0.05) body fat percentage (28.4%) compared to the American Caucasians (23.4%) 35 . For sarcopenia, a large project using pooled sample (n = 26,625 participants) showed that, for both genders, there were differences in mean ALM and ALMI among the subjects from the studies performed in different regions 28 , including multiple regions in USA and Europe. However, this project did not conduct a detailed analysis of geographical variations in ALM or ALMI, such as dividing the subjects into various geographical areas for further comparison, and also contained a small number of non-white subjects 28 . In our study, only Caucasian subjects were recruited and ALM and ALMI were also found to vary by different geographic locations.
Different definitions of obesity may lead to different results. If we use BMI as a phenotype to define obesity in the study, there was no significant difference in the prevalence of obesity for subjects from LOS and KCOS (Fig. 1). If we use WBFP as criteria for obesity classification, the obesity rate of KCOS was dramatically reduced from 23.13% (by BMI) to 7.73% (by WBFP) for males and reduced from 24.37% (by BMI) to 16.61% (by WBFP) for females. For LOS, the percentage of obesity among female subjects increased from 24.51% (by BMI) to 33.83% (by WBFP) and the obesity rate of male subjects decreased slightly, from 24.81% (by BMI) to 22.24% (by WBFP).
These results indicate that BMI should not be the only standard for obesity, especially when study geographical difference which involve subjects from different regions. A systematic review and meta-analysis 36 found that commonly used BMI cutoff values have high specificity but low sensitivity to identify excessive body adiposity. Thus, WBFP may be a good additional measurement to identify geographical difference for traits related to obesity.
Our study has several advantages compared to previous studies. First, in this study, lifestyle factors (exercise, smoking and alcohol use) were controlled to evaluate the effect of geographical variation on BMD and body composition. Second, the three cohorts were recruited by the same research group using the same inclusion and www.nature.com/scientificreports www.nature.com/scientificreports/ exclusion criteria. This may reduce the selection bias. Third, we used the same standard bone osteodensitometers machine, QDR 4500, the gold standard for bone mass measurement in humans 37 , which may further reduce systematic errors. Fourth, in this study, effect size was considered in order to show the degree of difference. In previous studies, p-value was the only way that data were compared. A proper inference requires full reporting as suggested by SAS in 2016 38 . P-value does not measure the size of an effect or the importance of a result.
Our study had some limitations. First, our sample size was relatively small. Only three regions were included in the study. We did not recruit subjects from western or eastern parts of the United States. Second, we could not investigate the factors that may cause the regional variation of BMD and body composition, like diet or sun exposure. Since we found WBFP variation among the three locations in different latitudes, for the next step, we would like to conduct further studies and try to define if the WBFP variation was caused by the potential mechanisms, like different sun exposure time via vitamin D metabolism, temperature variation in different latitudes or diet differences.
In summary, we found geographical differences of obesity and sarcopenia when defined by WBFP and ALMI respectively. No significant geographical effects on BMD variation were identified. The regional effects on obesity and sarcopenia were still significant after being adjusted for age, weight, height, smoking, alcohol use, and regular physical activity. This indicates that unknown region-related factors may contribute to variations in body composition. We also found that the geographical differences in obesity were easier to identify using WBFP, compared with BMI.

Materials and Methods
All the methods were conducted in accordance with the rules and guidelines of the Institutional Review Boards of University of Missouri Kansas City, Creighton University and Tulane University. The Institutional Review Boards of University of Missouri Kansas City, Creighton University and Tulane University approved the study. Written informed consent was obtained from all participants before inclusion in the study.  41 . All samples were approved by the respective institutional ethics review boards and all participants signed informed-consent documents before entering the studies. Only healthy people (defined by the exclusion criteria) were included in the analyses. All of the study subjects were Caucasians with age > 18 years. The exclusion criteria are:
A medical history questionnaire was given to each subject to obtain self-reported information on age, sex, family history, physical activity, drinking, and smoking history.

Measurement of BMD and body composition.
All BMDs at regional sites, including hip (HIP-BMD), spine (SPN-BMD), femoral neck (FNK-BMD) were measured, as were body composition including fat mass and lean mass at all regional sites, using Hologic QDR-4500 DXA scanners. DXA instruments were operated by trained technicians and calibrated daily by measuring spine phantoms. Long-term precision in our hands, the coefficient of variation (CV), was less than 1.6% for BMD measurement.
WBFP was calculated as whole body fat mass divided by whole body mass. ALM was determined by the sum of lean mass of both arms and legs (in unit of kilograms). The ALMI was calculated as ALM/height 2 (in unit of kilograms/meter 2 ). BMI was calculated as weight in kilograms divided by the square of height in meters.
Definition of sarcopenia and obesity. Sarcopenia was defined here using ALMI 25 . If a subject's ALMI fell into lower than 2 SD below mean of young adults, the subject would be considered as sarcopenia in the study 24,25 . In this definition, sarcopenia was significantly associated with physical disability in both genders, and (2019) 9:12311 | https://doi.org/10.1038/s41598-019-48734-9 www.nature.com/scientificreports www.nature.com/scientificreports/ was independent of ethnicity, age, morbidity, obesity, income and health behaviors 24 . The cutoff values for definition of sarcopenia were ALMI ≤ 7.26 kg/m 2 for males and ≤ 5.45 kg/m 2 for females. These cutoff values were broadly accepted in a number of sarcopenia studies (e.g. 24,25,42 ).
We defined obesity using BMI or WBFP. The WHO defines obesity as BMI ≥ 30 kg/m 2 . When we used WBFP to define obesity, cutoff values of 30% and 40% were proposed to define obesity for males and females, respectively. These cutoff values were adopted in a wide range of multiple studies (e.g. 43,44 ).
Data and statistical analyses. We performed a Grubbs' test to remove outliers. We applied a one-tailed test to detect outliers, which have values statistically different from other values (p-value < 0.05). We removed outliers and reanalyzed data as needed until no outliers were detected. ANCOVA was performed to detect geographical differences of BMD and body composition traits. In this study, we choose age, weight, height, smoking 32,45,46 , alcohol use 47 , and regular physical exercise 33,48 as covariates in the ANCOVA model. The comparison of percentage of regular physical exercise, alcohol use, and smoking were performed by Chi-square test. All other basic characteristics in this study were compared using ANOVA. Eta-squared based on ANOVA was also reported to reflect the effect size of differences due to its importance to empirical studies 34 . Given a sufficiently large sample size, a non-null statistical comparison will always show a statistically significant result. Therefore, in this study, we used both p-value and effect size as criteria to define statistical significance. The effect sizes were defined as small (η 2 = 0.01), medium (η 2 = 0.06), and large (η 2 = 0.14) based on benchmarks suggested by Cohen 31 . A test is significant if η 2 > 0.01 and p-values < 0.05.

Brief summary.
There is lack of studies on geographical variation influencing bone mineral density (BMD), obesity, and sarcopenia related traits in the US population. By comparing three study cohorts of various latitudes, we found no geographical effects on BMD. Obesity and sarcopenia related traits were significantly different among the three study cohorts.