2·1 Settings
Participants were recruited among community-dwelling adults (≥21 years) from a large north-eastern residential town of Yishun in Singapore, with residential population of 220,320 (49·4% men), with 12·2% older adults (≥65 years)(21). This is similar to the overall Singapore residential population of 4·02 million (48·9% men), with 14·4% older adults (≥65 years)(21).
2·2 Participants
Random sampling methodology was employed to obtain a representative sample of approximately 300 male and 300 female participants, filling quotas of 20-40 participants in each sex- and age-group (10-year age-groups between 21-60 years; 5-year age-groups after 60 years). Conventionally, the sample size of 30 or greater per age-group is sufficient for normative measures(22). Between October 2017 and February 2019, using a two-stage random sampling method, 50% of all housing blocks were randomly selected, and a random 20% of the units in each block were approached for participant recruitment. Between March and November 2019, 50% of all housing blocks were randomly selected and all units were approached. Up to three eligible participants were recruited from each housing unit using a door-to-door recruitment method. Non-response units were re-contacted a second time at a different time of day on a later date. Older adults above 75 years old were additionally recruited through community sources and from a list of registered participants in four senior activity centres. Exclusion criteria were: individuals with disabilities, injuries, fractures or surgeries that affected function, neuromuscular, neurological and cognitive impairments, or more than five poorly controlled comorbidities. Pregnant women or those planning for pregnancy were also excluded. The estimated overall response rate was 39·0%. Ethics approval was obtained from the National Healthcare Group Domain Specific Review Board (2017/00212). All respondents signed informed consent before their participation in the study.
2·3 Measurements and data collection
Participants answered a socioeconomic and lifestyle questionnaire pertaining to education level, housing type, living arrangement, marital status, history of tobacco use (smoking) and alcohol use, a health and medical questionnaire indicating history of medical conditions and comorbidities, the mini nutritional assessment short form (MNA-SF)(23), a global physical activity questionnaire(24), and the LASA physical activity questionnaire(25).
Body weight to the nearest 0·1kg and height to the nearest 0.1cm were measured using a digital balance and stadiometer (Seca, GmbH & Co. KG, Hamburg, Germany). Waist and hip circumferences were measured using a non-elastic, flexible measuring tape around the navel and widest part of the hips respectively. All participants underwent a DXA scan of the whole body (Hologic Discovery Wi, Hologic, Marlborough, MA, USA). The DXA scan was conducted by experienced radiographers. Body composition information - lean mass, fat mass, and bone mineral content, were obtained from the scan.
2·4 Overweight and Obesity
Classification of overweight and obesity by BMI were derived using WHO international criteria(6), and Singapore MOH Obesity Clinical Practice Guidelines(10, 11). Overweight and obesity were defined internationally as having a BMI 25·0-29·9kg/m2, and BMI ³30·0kg/m2, respectively. Singapore MOH Clinical Practice Guidelines defined overweight as BMI 23·0-27·4kg/m2 and obesity as BMI ³27·5kg/m2. The BF% cut-off points for obesity was set at 25% for men, and 35% for women(6, 13, 26). Waist circumference (WC) for abdominal obesity was defined as above 80cm for women, and above 90cm for men in Singapore(11).
2·5 Statistical analysis
All statistical analyses were performed using SPSS Statistics version 22·0 (IBM, Armonk, NY, USA). Relationship between BMI and BF% was analysed using multiple regression model. BF% was considered the dependent variable; 1/BMI, age, and ethnicity were independent variable. Data was analysed separately by sex. In exploratory analysis, the relationship between BMI and BF% was not linear, hence 1/BMI variable was used to linearise the data and to avoid the need for logarithmic conversion or the inclusion of power(27, 28). Potential interaction variables were explored in model development and a forward-backward stepwise procedure was utilised for the development of the prediction equation models. The dummy variables for ethnicity were E1 and E2. For Chinese E1 = 1 and E2 = 0, for Malays E1 = 0 and E2 = 1, and for Indians E1 = 0 and E2 = 0. Values are presented as mean ± SD, unless otherwise stated.