We analyzed cross-sectional data from the 2005–2006 NHANES, as that was the most recent year with body composition data. The survey uses a multistage, complex, stratified probability sampling design that oversamples minorities and is representative of non-institutionalized adults in the United States, providing excellent external generalizability. The survey has been conducted and managed by the Centers for Disease Control and Prevention since 1971, and its contents and procedure manuals are available online at http://www.cdc.gov/nchs/nhanes/htm. This study was exempt from local Institutional Review Board review due to the de-identified nature of the data analyzed.

We restricted our sample to individuals aged 20–49 years old. We excluded subjects who were pregnant, nursing, or status-post bilateral oophorectomy given changes in body weight, body composition, endocrine hormones, and/or type 2 diabetes risk in these groups. We also excluded subjects who were prescribed testosterone, growth hormone, or glucocorticoids given the known effects of these endocrine hormone deficiencies and/or their replacement on body composition and/or type 2 diabetes risk. Individuals whose height was >192.5 cm and/or whose weight was >136.4 kg were excluded due to limitations of the DXA table. Of 10 348 participants in NHANES 2005–2006, 1764 eligible participants were included in this analysis.

All body composition measures were assessed using a DXA QDR-4500 Hologic scanner (Bedford, MA). Android and gynoid regions were defined by the Hologic APEX software used in the scan analysis. The android region is the area around the waist between the mid-point of the lumbar spine and the top of the pelvis; the gynoid area lies between the head of the femur and mid-thigh. Appendicular lean mass (ALM) was defined as the sum of the muscle mass of both legs and arms.

The following data were self-reported using questionnaires and included in analyses because they have been independently associated with type 2 diabetes risk: age, sex, race/ethnicity [9, 10], education [11], smoking status [12], and physical inactivity [13]. Race/ethnicity was categorized as Hispanic (combining Mexican American and other Hispanic), non-Hispanic White, non-Hispanic Black, and other (including multi-racial). For race-stratified analyses, “other” was excluded. Education was categorized as (1) less than 12th grade, (2) high school graduate or General Education Diploma (GED) equivalent, or (3) higher. For race-stratified analyses, education was collapsed into two categories of (1) less than 12th grade or (2) high school graduate, GED equivalent, or higher. Smoking status was categorized as never smoker (smoked <100 cigarettes in life), former smoker (do not now smoke cigarettes), or current smoker (smoke cigarettes every day or some days). Individuals were defined as physically inactive if they reported no vigorous or moderate activity of at least 10 min over the past 30 days that caused light to heavy sweating or slight to large increases in breathing according to the standard NHANES physical activity questionnaire. This definition of physical inactivity was chosen because prospective studies in both women and men have demonstrated that any level of self-reported physical activity is associated with a lower risk of type 2 diabetes compared to no physical activity [14, 15]. In a secondary analysis, physical activity was coded according to whether the individual met the American Heart Association’s (AHA) recommendations for physical activity in adults (150 min of moderate-to-vigorous aerobic activity per week, or 75 min of vigorous aerobic activity per week) [16]. An electronic digital scale, calibrated in kilograms, was used to assess weight, and a stadiometer was used to measure height after a deep inhalation. BMI was calculated as weight in kilograms divided by height in meters squared. Height was included as a covariate in analyses given the linear relationship between height and muscle mass [17].

All techniques in NHANES followed the guidelines put forth by the American Diabetes Association. A fasting glucose blood test was performed in the morning after a 9-hour fast; subsequently, a 75-g oral glucose tolerance test (OGTT) was performed. Exclusion criteria for oral glucose tolerance testing included hemophilia or chemotherapy safety exclusions, fasting <9 h, taking insulin or oral medications for diabetes, refusing phlebotomy, and not drinking the entire Trutol™ solution within the allotted time. Plasma glucose was measured using a hexokinase method (Roche/Hitachi 911), and samples were processed, stored, and shipped to Fairview Medical Center Laboratory at the University of Minnesota for analysis. Glycosylated hemoglobin (HbA1c) was measured using HPLC (Tosoh Medics, Inc., San Francisco, CA). HbA1c samples were processed, stored, and shipped to the Diabetes Laboratory at the University of Minnesota for analysis.

Diabetes was defined by the presence of one or more of the following conditions: (1) HbA1c ≥ 48 mmol/mol (6.5%); (2) fasting glucose ≥7 mmol/l (126 mg/dl); (3) a 2-h glucose on an OGTT of ≥11.1 mmol/liter (200 mg/dl); (4) self-reported diagnosis of diabetes; or (4) self-reported use of diabetes medications (oral hypoglycemic agents and/or insulin) as previously defined [18]. Because 73 subjects were missing HbA1c, fasting glucose, and 2-h glucose on OGTT, diabetes was defined as self-reported diagnosis or self-reported use of diabetes medications in a secondary analysis.

All data were downloaded, merged according to NHANES guidelines, and analyzed incorporating sampling weights, primary sampling units, and strata as supplied by NHANES. Continuous variables are represented as mean ± SD and categorical variables as count (percent).

To determine sex-stratified differences in clinical characteristics between those with and without diabetes, we performed Wilcoxon tests to compare continuous variables, Chi-squared tests to compare categorical variables, and Fisher’s exact tests to compare categorical variables if count <10 in either cell.

To determine the association between percent ALM/weight (primary predictor) and diabetes prevalence (primary outcome), we performed a series of multivariate logistic regression models: (1) a sex-stratified model (primary hypothesis), (2) a sex-combined model that included an interaction term for percent ALM/weight with sex, and (3) a sex- and race-stratified model. We adjusted for demographic covariates (i.e., age, sex, race, smoking, and education) known to be associated with type 2 diabetes risk, height, and physical inactivity, as well as the ratio of android/gynoid fat because it varies by sex and because android fat is detrimental towards, whereas gynoid fat is protective against, type 2 diabetes. A power calculation for the primary endpoints was as follows: in a sample of 1764 participants, assuming alpha = 0.05, power = 0.8, and a diabetes prevalence of 5%, we could detect an odds ratio of 1.15 between ALM/weight below the median vs ALM/weight above the median. Data are presented as an odds ratio with 95% confidence interval and associated p value for each model.

We used SAS (version 9.2 or 9.3; SAS Institute, Cary, NC) for all analyses and applied procedures to account for NHANES 2005–2006 sampling probabilities and complex sampling design in all models. Multiple imputations was applied to address potential bias resulting from nonrandom missing DXA data [19]. Five complete data files that contained both the non-missing and imputed values (generated using sequential multivariate imputation) were created. A two-sided p value ≤0.05 was considered a statistically significant test of the hypothesis that less skeletal muscle mass is associated with greater diabetes prevalence independent of body fat distribution in both young men and young women in the United States.