The present analysis is based on a cross-sectional study that collected data on social determinants of Indigenous health among university students who identified as Indigenous and attended school in a small city in western Canada (Population size: 98,000). Data collection began in September 2015 and continued over 4 academic terms ending in April 2017. The final sample size was 150 Indigenous adults. Study procedures were approved by the Human Subjects Research Committee at the University of Lethbridge. Data analyzed for the present analysis are available from the corresponding author upon reasonable request.
Indigenous Advisory Committee
This study was conceptualized using a participatory action research framework [23]. An Indigenous Advisory Committee made up of key members of the Indigenous community in Lethbridge was assembled and worked with the research team to set study priorities and make data collection decisions. Through this work, it was determined that salivary rather than blood samples would be taken given blood is a sacred element in many Indigenous cultures and must be respected in ways that may be incompatible with scientific research. As saliva is also a substance that comes from the body, a system was put in place in consultation with Indigenous Knowledge Holders to ensure the wishes of participants were honored. The consent form provided participants the option of having their saliva samples returned to them upon analysis or to have their saliva samples included in an Indigenous ceremony led by an Elder that returned the samples to the Earth.
Procedures
Participants were recruited using posters and ads placed in e-newsletters on campus.
Respondents were asked to confirm eligibility by email/phone (i.e., they identified as Indigenous, current post-secondary students, and 18 years or older). Participants then attended an on-campus study office to complete consent procedures, paper-and pencil surveys, and the physical assessments needed to calculate AL score (mean completion time = 90 min) during standard office hours (9:00am – 4:00pm). To ensure sufficient participant recruitment, we needed to accommodate student course schedules and thus could not standardize a narrow window for data collection across participants, which may have been useful for some biomarkers examined (e.g., DHEA-S, CRP). Saliva samples were collected at 3 time points during this visit using the passive drool technique. Participants rinsed their mouth with water and the first sample was collected after completing a portion of the questionnaire. Remaining samples were taken 30 and 60 minutes later. Whole saliva samples were collected in a 2ml microcentrifuge tube using a Saliva Collection Aid (Salimetrics, State College, PA). During data collection, salivary samples were stored in the in-office freezer and then transferred to a -80°C freezer. Participants were provided with supplies for collecting saliva samples at home for two days, as well contact information for the research assistant who collected data from them during the office visit. Participants were asked to call, email or text with questions, and coordinate a time for sample return.
At home, participants selected two consecutive days with similar wake/sleep times and collected a saliva sample at three time points: immediately upon wake-up, 30 minutes after wake-up, and before bed, and to record the times in which samples were taken on forms provided. Participants were instructed to place the swab under the tongue for three minutes and then place it in a pre-labeled tube and put it in their freezer. When all six samples were collected, the participant contacted the research assistant to coordinate sample return. We used cortisol awakening response (CAR) expert consensus guidelines to increase at-home adherence including clearly explaining the importance of strict adherence to sampling times, emphasizing the importance of collecting sample S1 immediately upon awakening, encouraging participants to ask questions via text/email/phone, providing take-home instructions, having participants record data collection time points in a diary log, advising participants to place kits beside the bed for morning collection, and text messaging the evening before sampling to highlight instructions [24]. Participants returned the samples in an insulated lunch kit with a freezer pack given to them during the in-office visit. Samples received were transferred to a -80ºC freezer. Participants were given an honorarium of $50 for in-office measures and $50 for at-home measures.
Measures
Allostatic load (AL)
AL score was based on a composite of seven biomarkers across four biological domains:
- Cardiovascular markers: Resting systolic and diastolic blood pressure were measured using a Life Source automated sphygmomanometer (Auto Control Medical, Mississauga, ON). The first was taken approximately 15 minutes after the participant arrived, once they had completed the consent process and answered the first part of the survey package in a seated position. This reading was discarded. Two additional readings were taken 15 and 30 minutes after the first while the participant was seated. These two measures were averaged.
- Neuroendocrine markers included DHEA-S and CAR. All were analyzed in duplicate. As per manufacturer’s suggestion for DHEA-S, the three in-office samples were pooled and mixed for analysis. To examine CAR, the wake-up (S1) and 30 minutes post wake-up (S2) samples taken at home on the second day were used to calculate the percent change in cortisol between S1 and S2. Day 1 at-home samples were not combined with Day 2 to produce an average because missing data were higher on Day 1. CAR represents the sharp rise in cortisol levels across the first 30-45 min following morning awakening. In healthy adults, the magnitude of CAR ranges between a 50-156% [25]. The mean CAR magnitude in this study was 65.1% (Table 1).
- Metabolic markers included body mass index (BMI) and waist circumference. To calculate BMI, height and weight were measured to the nearest 0.5cm using a Health O Meter mechanical beam scale and stadiometer, and to the nearly 0.1kg using a weighbeam scale; respectively. Waist circumference (WC) was measured at the top of the iliac crest, to the nearest 0.5 cm. Although correlated (Pearson’s r = 0.87 in this sample), both measures were included in the AL score as each are independently associated with health risk.
- Immune marker: We measured CRP using the third in-office saliva sample.
Cortisol, DHEA-S, and CRP concentrations were assessed using enzyme-linked immunosorbent assays (ELISA) (Salimetrics, LLC., State College, PA). Average intra-assay variability was 3.9 % for cortisol, 6.6% for DHEAS, and 4.3% for CRP. Average inter-assay variability was 9.2% for cortisol, 12.8% for DHEAS, and 8.3% for CRP. For CAR, all samples from the same participant were analyzed in the same plate, to minimize the effect of inter-assay variability. AL risk assessment was based on the distribution of the study sample for salivary CRP and DHEA by dividing the sample into sex-specific quartiles with high risk defined by the highest quartile for CRP and the lowest quartile for DHEA-S. As shown in Table 1, we used standard cutoffs for all other biomarkers [26,27]. Consistent with prior studies, one point was assigned if the variable was in the high-risk quartile and 0 if not. Scores were summed across each system type (neuroendocrine, metabolic, immune, and cardiovascular) to create a total score for AL.
Racial discrimination
The Experiences of Discrimination (EOD) Scale is a valid and reliable measure of self-reported racial discrimination that has been used across many ethnic groups [28]. The situation score is derived by counting the number of situations (1 to 9) in which racial discrimination was experienced. Previous research suggests Indigenous Canadians experience high levels of racial discrimination, and that a scale adjusted to measure experiences in the past 12 months is recommended to ensure sufficient variability [4,29]. Consistent with previous research, each question in the present study was worded as follows, with information in brackets reflecting words added, and X reflecting the situation tested: (In the past 12 months) have you experienced discrimination, been prevented from doing something, or been hassled or made to feel inferior at X because of your (Aboriginal) race, ethnicity or colour? Internal consistency of the measure in this study was good (Cronbach’s alpha = 0.82). Racially-motivated HD was operationalized using the following adapted question from the EOD: In the past 12 months, have you experienced discrimination, or been hassled or made to feel inferior getting or maintaining housing because of your Aboriginal race, ethnicity, or color? Response options were 0 = No and 1 = Yes. [28].
Covariates
Exact age, gender, parenthood and income were collected as part of the survey package. Categories used to examine each covariate are outlined in Table 2.
Missing data
Data were collected from 150 participants, 35 of whom were removed from the analysis because they chose to not complete and/or return at-home samples. An additional 8 were removed because the timing of at-home sampling was completed in ways that did not follow procedure resulting in the inability to calculate valid CAR [24]. As well, two participants were removed for not completing questions about discrimination in the past 12 months, and one was removed for not reporting their age. There were no missing data on survey questions about gender or income The final sample size included in this analysis was N = 104. Independent-samples t-tests confirmed the mean age, income, and HD experience of participants included and excluded from the analysis due to missing data were not statistically different, nor was the gender balance different between groups. We conducted a supplementary reanalysis of the main findings excluding CAR from the AL calculation, which reduced our ability to understand the impacts of housing discrimination on neuroendocrine function, but increased the sample size to N = 144.
Analysis strategy
Bootstrapped linear regression models (k = 5000) examined the association between past-year HD (yes or no) and the continuous form of AL. Bias-corrected and accelerated (BCa) bootstrap intervals were used to adjust for potential skew. Potential confounders were carefully considered and tested before inclusion in models to reduce model overfitting, and keeping in mind that analyses that follow the “more control variables is better” approach to improve causal inference have been debunked [30–33]. Thus, potential confounders were tested using individual regression models before entry into the main model. Those associated with AL at p < 0.10 were retained [34] which included age, income, and parenthood; but not gender or marital status.
Previous studies have controlled for health-risk behavior and mental health when examining associations between discrimination and AL. We disagree with such an approach given racial discrimination predicts later mental health and health-risk behaviour problems in more than 30 longitudinal studies across a variety of ethnicities and age groups worldwide [1,35,36]. This large body of longitudinal evidence suggests mental health and health-risk behaviour should be examined as part of the causal chain linking various forms of racial discrimination to AL (i.e., as mediators), rather than noise that is controlled in statistical models. Controlling for such intermediate variables introduces bias by decomposing the total effect of x on y into its parts [32]. If that is the goal, one should conduct a mediation analysis [37]; however, such analyses can produce endogenous selection bias [38] and would require a larger sample size. All analyses were run using SPSS 25.0.