[ad_1]
Population
This cross-sectional secondary analysis used data from a nationwide survey that examined the knowledge, attitudes, and practices of diabetes in Singapore. The methodology was described in detail previously [29]. Participants were Singaporeans or permanent residents, ≥ 18 years, living in Singapore during the survey period and fluent in English, Malay, Chinese, or Tamil. The survey excluded individuals below 18 years old, those who could not be contacted due to missing or incomplete addresses, lived outside of Singapore, or were institutionalized during the study period. In addition, individuals who had difficulty completing the survey due to physical, mental, or cognitive impairment were also excluded. The survey was conducted face-to-face in either one of Singapore’s four main languages: English, Malay, Chinese, or Tamil. Responses were recorded via computer-assisted personal interviews (CAPI) using handheld tablets.
The recruitment period was from February 2019 to March 2020. This study excluded participants recruited between April 1st, 2020 and September 1st, 2020 (n(28) because of the COVID-19 pandemic and the resultant self-imposed restriction of movement, as well as social distancing measures, implemented that may have influenced physical activity level and sedentary behavior.
The study was approved by the Institute of Mental Health’s Institutional Research Review Committee (IRRC) and the National Healthcare Group’s Domain Specific Review Board (Ref: 2018/00430). All participants provided written informed consent. Parental consent was also obtained for participants aged 18 to 20 years.
Sampling and sample size
A disproportionate stratified sampling design was used to sample participants from a national administrative database that comprises all Singapore residents. The proportion of participants in each ethnic group (Chinese, Malay, and Indian) was fixed at approximately 30%. In each age group (18 – 34 years, 35 – 49 years, 50 – 64 years, 65 years and above), the proportion of respondents was specified at 20%. This survey design ensured that the different ethnicities and age groups were well represented. Additionally, survey weights were incorporated to ensure that the sample was representative of the population.
The sample size in the survey was 3000, which was determined by the prevalence of diabetes knowledge in Singapore (60%), as reported in a previous study [30]. Moreover, it was calculated using a power of 0.8, type I error of 0.05, and adjusted for design effects. The margin of error was 2.5% for the overall prevalence and 4.5–5.0% for prevalence stratified by age groups and ethnicities.
Recruitment strategy
The participants were sampled from a national administrative database that included all residents in Singapore [29]. An invitation letter was sent to the participants 1 to 2 weeks before the household visit by the interviewer [29]. The letter provided information about the study and a contact number to address any inquiries about the study [29]. A maximum of 10 visits were made to reduce the survey’s non-response rate [29]. If the participant was not at home during the visit, a card with the survey firm’s contact number was left in the letterbox [29]. An inconvenience fee of SGD40 was given to the participant after the survey [29].
Barriers to physical activity
Respondents were asked to rate 12 barriers on a three-point Likert scale: ‘not really a barrier’, ‘somewhat a barrier’, ‘very much a barrier’. Internal barriers comprised ‘a disability or injury’, ‘young children or family needs’, ‘work’, ‘lack of time’, ‘age’ and ‘feeling tired’. External barriers consisted of ‘the weather’, ‘pollution’, ‘safety concerns’, ‘limited accessibility’, ‘cost of exercising’ and ‘lack of footpath, cycle lanes or parks’. An item was considered a barrier if the respondent indicated it as either ‘somewhat a barrier’ or ‘very much a barrier’. These barriers were identified through literature review and consultation with clinicians, an epidemiologist and policymakers working in the diabetes prevention domain [31,32,33]. Moreover, the content validity of the questionnaire was assessed through cognitive interviews. We asked the participants if they felt the items in the questionnaire represented the most important barriers and whether we had left out anything relevant.
Physical activity and sedentary behavior
Information on physical activity and sedentary behavior was collected using the Global Physical Activity Questionnaire (GPAQ). It is a 16-item self-reported questionnaire that assesses two types of physical activity intensity in three different domains: moderate (work, transport and recreation) and vigorous (work and recreation) intensity [34]. Moderate intensive activities are those that involve a small increase in breathing or heart rate for at least 10 min continuously (eg, brisk walking), whereas vigorous intensive activities are those that involve a large increase in breathing or heart rate for at least 10 min continuously. 10 min continuously ‘(eg, running) [34]. The energy expenditure for each domain was computed by multiplying the metabolic equivalent (MET) values to time variables. For moderately intensive activities, the MET value is 4. The MET value for vigorous intensive activities is 8 [35].
The physical activity level in each domain was calculated by summing the energy expenditure. Overall physical activity level was computed by adding the energy expenditure from the three domains. Sedentary behavior was determined by asking the amount of time spent sitting or reclining per day. GPAQ was used to assess physical activity and sedentary behavior because studies in Singapore have demonstrated that it correlated moderately with accelerometer-measured physical activity and sedentary behavior [36, 37]. Moreover, GPAQ is inexpensive to administer in population-based studies [36].
Confounders
Based on a previous analysis by Lau et al. [38] and two systematic reviews by O’Donoghue et al. and Bauman et al. [39, 40], we adjusted for the following confounders when examining the association between barriers to physical activity and physical activity-related outcomes: age, sex, ethnicity, monthly personal income, chronic physical conditions, and sedentary behavior. Moreover, the systematic review by Bauman et al. found that significant environmental factors of leisure-related activity include safety concerns, access to facilities and the presence of pavement [40]. Hence, ‘safety concerns’, ‘limited accessibility’ and ‘lack of footpath, cycle lanes or parks’ were included as confounders regardless of their importance in multivariable regression analyzes for leisure-related activity.
For sedentary behavior, we adjusted for the following confounders: age, education, marital status, monthly personal income, body mass index, chronic physical conditions, and physical activity. Confounding environmental factors included ‘safety concerns’, ‘weather’, and ‘limited accessibility’. These confounders were obtained from previous analysis by Lau et al. and the systematic review by O’Donoghue et al. [38, 39].
Sociodemographic factors were included in the analysis with the following categories: age (18 – 34 years, 35 – 49 years, 50 – 64 years, 65 years and above), sex (male and female), ethnicity (Chinese, Malay, Indian, Others), educational qualification (primary or below, secondary, pre-university / junior college, vocational institute, diploma, degree and above), marital status (single, married, divorced, separated / widowed / divorced), monthly personal income (no income / below SGD 2000, SGD 2000 – 3999, SGD 4000 – 5999, SGD 6000 – 9999 and SGD10 000 and above) and WHO Classification of body mass index (Underweight, Normal weight, Overweight and Obese).
Chronic conditions were assessed using a self-reported checklist of 18 chronic conditions. These conditions included asthma, arthritis, back problems, cancer, chronic inflammatory bowel disease, chronic lung diseases, congestive heart failure, diabetes, heart disease, hyperlipidaemia, hypertension, kidney failure, migraine, neurological conditions, Parkinson’s disease, stomach ulcer, stroke, and thyroid disease. These responses were divided into three categories: no chronic condition, one chronic condition, and two or more chronic conditions.
Statistical analysis
Survey weights were included in the analysis to account for disproportionate sampling, non-response bias, and post-stratification by age and ethnicity. Weighted percentages and unweighted frequencies were used to summarize categorical variables. The outcomes examined were domain-specific physical activity (work, transport and leisure) and sedentary behavior.
All physical activity-related outcomes were positively skewed and had many zeros. Hence, associations between barriers to physical activity and these outcomes were determined by considering four regression models: Poisson regression, negative binomial regression, zero-inflated Poisson model and zero-inflated negative binomial model [41]. The zero-inflated negative binomial model was selected as the best model as it had the lowest Akaike Information Criteria (AIC) [41].
The zero-inflated negative binomial model is a two-part model. The first portion follows a negative binomial distribution and relates to the change in physical activity level among those who were physically active [42]. The exponential form of the regression coefficient denotes the proportional change in physical activity level (in MET-minutes) for each unit increase in the independent variable [42]. The second portion follows a logit probability process to distinguish between physically inactive respondents (represented by excessive zeros) and physically active respondents [42]. The exponential form of the regression coefficient indicates the odds of being physically inactive for each unit increase in the independent variable [42]. The exponential form of the regression coefficients (eβ) and 95% confidence interval (e95% CI) were calculated for all zero-inflated negative binomial models.
Associations between the barriers and sedentary behavior were assessed using a multivariable linear regression. Linear regression assumes that the residuals have constant variance and a normal distribution. These assumptions were tested by inspecting the residuals versus fitted values plot and the quantile – quantile plot of the residuals. These plots showed that the assumptions had not been seriously violated. Beta-coefficients (β) and 95% confidence interval (CI) were reported for sedentary behavior.
All regression models were modeled based on the parsimony principle. Initially, all barriers to physical activity were included. Statistically insignificant barriers were removed from the multivariable model in a stepwise manner. The final multivariable model for each outcome included significant barriers to physical activity and established confounders. Due to the disproportionate stratified sampling design, standard errors were estimated using Taylor series linearization.
The analysis was conducted using Stata / SE 17.0 (College Station, Texas), with a two-sided test at a 5% significance level. Missing data were removed in a listwise manner.
[ad_2]
Source link