3. Hypothesis and Research Methods
3.7. Parameter Selection for the T2D Model Without Adjustment for POP
Before testing for an association of POP plasma concentration with the OR of having diabetes, I specified a basic model without a pollutant parameter using
established determinants of T2D. The following parameters were considered for inclusion in the basic model for predicting the OR of having T2D based on a priori knowledge and quality of the data.
3.7.1. Measures of obesity
Obesity is a known risk factor for diabetes.193,195,197,229 Figure 2 illustrates that
obesity is expected to be a predictor of the OR of having T2D after adjusting for other predictors. The CHMS employed trained examiners for anthropometric measurements during each respondent’s visit to the mobile examination clinic. Selecting the measures to adjust for obesity in the model was based on a priori knowledge and assessment of model fit. The measures of obesity selected as predictors in the models for the OR of T2D include body mass index (BMI), waist circumference (WC), and waist to hip ratio (WHR). Body Mass Index (BMI) is a comparison of "weight" relative to the "height" of
respondents. BMI was calculated by dividing weight in kilograms by height in metres squared as BMI = weight (kilograms) / height (metres)2. Each of the measures has
reported advantages and limitations as discussed in this section below.
Abdominal obesity as measured by waist circumference and waist-to-hip ratio has been reported to be a better predictor than was overall body mass for T2D,195,230
metabolic syndrome,231 and cardiovascular risk.232–234 The American Heart Association
also recommends use of WC or WHR as better predictors of health risk than BMI when a trained examiner is taking measurements.235 An analysis of the Australian Risk Factor
Prevalence Study found the waist to hip ratio provides a superior measure of central obesity with low measurement error, high precision, and no bias over a wide range of ethnic groups.236 However, some clinicians have pointed out that ratios such as the
waist-to-hip ratio can lead to misclassification particularly for some females with a relatively larger hip circumference.235
Health Canada reports that WC and WHR are most predictive of additional risk with a BMI between 18.5 and 34.9. The WC and WHR measurements are less
predictive of additional health risk when BMI is greater than 35.0.237 To adjust for this
expected diminishing risk for BMI above 35, an interaction term between BMI and WC or WHR is included in the models as specified below.
3.7.2. Measures of physical activity, sedentariness and fitness
Physical activity, sedentariness and fitness are factors known to be related to T2D.196,197 There is a disease pathway through physical activity and/or fitness as shownin figure 2 after adjusting for other parameters. Therefore, physical activity/fitness is included in the model of T2D. Several measures were considered as indicators of physical activity, sedentariness and fitness. Variable selection was based on a priori knowledge and assessment of data quality and model fit.
The CHMS includes many direct measures of physical activity and fitness from an accelerometer monitor and fitness testing at the mobile examination clinic. The CHMS provides data from an accelerometer device worn by respondents for up to a week. The accelerometer data indicates general daily activity levels. Respondents also
were asked to participate in a series of fitness tests based on a modified version of the Canadian Fitness Test (mCAFT).
The direct measured data from the activity monitor and/or the fitness testing would be preferable to use in the models for their accuracy. However, non-response and missing data are significantly higher for the direct measures data than for the self- reported data. The respondents who provided direct measures of activity or fitness were substantially different in demographic and other characteristics from the respondents with missing data for the direct measures. Respondents with direct measures had significantly higher self-reported physical activity levels as compared to respondents who did not provide direct measures of activity or fitness (results are not shown).
In contrast, the self-reported measures of physical activities have a very low proportion of missing values. Therefore, a self-reported measure of activity was
selected for the models over a direct measure to minimize the loss of respondents from the analysis due to missing data and to avoid introducing selection bias into the models. The daily leisure activity energy expenditure measure derived from self-reported physical activity information is the conceptually most appropriate self-reported measure. Daily leisure activity energy expenditure was calculated by the CHMS using the frequency and duration per session of the physical activity and the metabolic equivalent task (MET) value of the activity.238 The MET is a value of metabolic energy cost expressed as a
multiple of the resting metabolic rate. For example, an activity of 4 METs requires four times the amount of energy as compared to when the body is at rest. The measure was calculated as:
EE (Energy expenditure for each activity) = (N * D * MET value) / 365
Where: N = the number of times a respondent engaged in an activity over a 12 month period. D = the average duration in hours of the activity. MET value = the energy cost of the activity expressed as kilocalories expended per kilogram of body weight per hour of activity (kcal/kg per hour) / 365 days. MET values are generally expressed in three intensity levels (i.e. low, medium, high). This calculation is adopted from the Canadian Fitness and Lifestyle Research Institute.
3.7.3. Measure of age
Age is a potential confounder for the relationship between POPs plasma concentration and the OR of T2D. The process of aging could itself affect the onset of T2D.192 Additionally, POPs are known to bioaccumulate in tissues with age and are
therefore expected to have a strong positive correlation with age. To assess if there was a potential for collinearity between age and each exposure measure in the models, the correlation between age and each exposure measure was estimated. High correlations, for example above 0.9, indicate a potential for collinearity between variables when included together in a regression model. Another indicator of collinearity in a model is the change in the standard error of the estimate of the exposure parameter after adding age to the model. If the parameter estimate’s standard error is substantially inflated by adding age then there may be a problem of collinearity.
The date of birth and date of data collection were included in the CHMS data. The age in years at the date of both the questionnaire interview and the date of physical examination were derived from these dates. Age in years at the time of the physical examination was used as a covariate in the models.