This was a cross-sectional, population-based study that used data from men and women 50 years and older who were enrolled in the National Health and Nutrition Examination Survey (NHANES) 1999–2002. The NHANES performs a continuous, nationally representative health survey of the civilian, non-institutionalized U.S. population and collect data on approximately 5,000 persons each year from in-person interviews, physical examinations, and medical tests. To produce nationally representative estimates, NHANES has adopted a complex, multistage, probability sampling design, and the survey oversamples certain ethnic minorities and individuals aged 60 years and older . In 1999, NHANES began performing dual-energy x-ray absorptiometry (DXA) whole body examinations, and providing data on body composition, such as lean mass for total body and for each arm and leg, head, and trunk. The DXA equipment used was a Hologic QDR 4500A fan beam densitometer . During 1999–2002, NHANES also collected data on muscle strength as measured by the isokinetic strength of the knee extensors using a Kin Com MP dynamometer , which enabled the study of both muscle mass and muscle strength. A total of 2,647 men and women, 50 years of age and older, with muscle mass and muscle strength data available from the NHANES 1999–2002, were included in the present study.
Muscle mass was assessed using lean mass, excluding bone mineral content (BMC), obtained from DXA scan. In previous research, lean mass, excluding BMC, has been shown to validly represent skeletal muscle mass in the extremities . In our study, appendicular skeletal muscle mass (ASM) for each subject was calculated as the sum of lean mass, excluding BMC, in both arms and both legs. A height-adjusted index (aASM) was then calculated by dividing a subject’s ASM in kilograms (kg) by the subject’s height in meters squared (m2) [23, 24]. This index, expressed in kg/m2, was the muscle mass measure used for all analyses.
Muscle strength was established using measurement of the isokinetic strength of the knee extensors at peak force (isokinetic quadriceps strength [IQS]), in newtons . There is evidence that lower extremity strength is associated with physical function, an important component of quality of life . In NHANES 1999–2002, a total of six measurements of right quadriceps muscle strength were taken, three warm-up trial measurements followed by three outcome measurements. If a survey participant completed 4 to 6 measures, the highest peak force was selected from trials 4 to 6. If a survey participant had completed fewer than 4 trials, the highest peak force from the warm-up trials was selected. The IQS, expressed as newtons at one speed (60 degrees/second) on the right side, was the muscle strength measure used for all analyses.
Comorbidity data were obtained from self-reported personal interview questionnaire data collected in the NHANES. Presence of a variety of health conditions, which potentially affect muscle mass and/or strength, was assessed using data from the medical conditions/history questionnaire, including the following: diabetes; coronary heart disease (CHD) (including heart attack/myocardial infarction)/congestive heart failure (CHF);vision problems; obesity; arthritis (including both rheumatoid and osteoarthritis); asthma; osteoporosis; cancer (multiple varieties, excluding skin cancer); and chronic bronchitis/emphysema. Body mass index (BMI) was also investigated and defined as weight (kg) divided by height squared (m2). Consistent with the recommendation from Center for Disease Control and Prevention , the following four categories of BMI were adopted in this study: BMI < 18.5 kg/m2 (underweight), 18.5-24.99 (healthy weight), 25–29.99 (overweight), and ≥30 (obese).
Statistical procedures were employed that made use of the information from the complex, multistage, probability sampling design of the NHANES. Sampling weights were used to create mean estimates that were nationally representative of the older adult population, taking advantage of the oversampling of subjects in particular age, racial, and ethnic groups. All statistical analyses were design-based, except the partial correlation coefficient calculation that used raw, un-weighted data.
Distributions of aASM and IQS, by gender and age group (in five-year intervals), were presented using box plots. The box plots present the mean, 5th, 25th, 50th (median), 75th and 95th percentiles. The Pearson partial correlation coefficient, controlling for age and gender, was used to examine the association between aASM and IQS.
Multiple regression models, using survey strata and weighting, were conducted to examine the effect of each comorbidity on IQS and on the relationship between aASM and IQS. We first constructed a linear regression model on IQS with aASM as an exploratory variable, adjusting for age and gender (Model 1). Next, for each comorbidity, we built two additional linear regression models on IQS, one with the studied comorbidity as an exploratory variable, adjusting for aASM, age and gender (Model 2), and another with all variables in Model 2, plus the interaction between aASM and the comorbidity of interest (Model 3). The regression coefficients for aASM and the studied comorbidity were presented, including the 95% confidence intervals (CIs), along with the significance level of the interaction term from Model 3. Where results suggested a significant interaction between aASM and a given comorbidity, the relationship between aASM and IQS was graphically presented via scatter plot by using the raw non-weighted data, along with the age- and gender-adjusted regression lines, via Model 3 and stratifying on comorbidity status. In addition, subgroup analyses of linear regression models (Model 1) were fit within comorbidity subgroups (e.g., arthritis and non-arthritis subgroups), adjusting for age and gender.
All analyses were conducted using SAS®, version 9.2 survey procedures (SAS Institute, Cary, North Carolina, USA). All hypothesis tests were 2-tailed with a significance level of 0.05, and regression coefficients were deemed significantly different if the corresponding 95% CIs were non-overlapping.