Study design
The Prediabetes Detection and Physical Activity Intervention Delivery (PRE-PAID) project focuses on the detection of individuals at high risk for developing type 2 diabetes using a community-based public health approach. The mandate of the PRE-PAID program was to focus efforts on ethnicities known to be at elevated risk for developing type 2 diabetes, which include persons of South Asian, African-Caribbean, Chinese and Aboriginal descent.
Selected communities had an elevated prevalence of type 2 diabetes and a concentrated population of high risk ethnicities. Demographic information was taken from the Institute for Clinical Evaluative Sciences diabetes atlas for the city of Toronto which provided information about diabetes incidence and prevalence by neighbourhood as well as a breakdown of the population by ethnicity [16]. Study participants were recruited through an established network of community partnerships with various organizations that provide public health-related programs to their constituents. Participants were recruited through printed materials, e-mail distribution lists and public diabetes screening events held in high-traffic areas such as shopping malls and community health centres. All participants provided written, informed consent prior to collection of data and all protocols utilized by the PRE-PAID project were approved by the York University Human Participants Review Committee.
Questionnaire design
The FINDRISC and CANRISK questionnaires provided a detailed and well-established framework upon which the PRE-PAID risk questionnaire was modeled. Slight alterations from the CANRISK questionnaire were made to minimize participant burden by removing questions about fruit and vegetable consumption, level of education and giving birth to a large baby. The PRE-PAID investigators opted to streamline the time taken to complete the questionnaire due to the fact that the capillary blood testing immediately followed its completion and some participants may have been lost due to the additional 15 minute commitment for the blood testing component. The questionnaire was also modified in order to include more detailed information (frequency and intensity) regarding the physical activity habits of those completing the questionnaire. These changes were also adopted as a result of the published validation of the CANRISK questionnaire which showed that the question regarding fruit and vegetable consumption, physical activity and macrosomia (birth to a large baby) were not significant contributors to their logistic regression model [15]. The PRE-PAID questionnaire is included as Additional file 1. Upon completion of the seven questions, an overall risk score was tabulated, based on a scoring paradigm similar to that of CANRISK, placing individuals into one of five different risk categories; "Small" (score 0–6), "Moderate" (score 7–11), "High" (score 12–14), "Very High" (score 15–20) and "Extreme" (score over 20). Trained members of the research team assisted study participants with questionnaire completion, and all questionnaire responses were based on self-reported information. BMI charts were provided to simplify the estimation of BMI from body mass and height (kg/m2). Participants were only required to complete the questions that contributed to the calculated risk score. The PRE-PAID questionnaire included space to self-report specific values for height, body mass, age and waist circumference. The inclusion of these values was encouraged to allow future analysis of participant demographics, but not required to attain a complete risk score.
Study participants
Persons were considered eligible for inclusion if they were over 18 years of age and if they did not possess any condition that would preclude them from having a capillary blood test to assess their glycemic control. English language proficiency was encouraged but not essential as the questionnaire was translated into Chinese (simplified and traditional), Punjabi, and Hindi. A total of 691 individuals were recruited in this study.
Blood testing
Point-of-care fingerstick capillary blood testing was performed to validate the risk questionnaire outcomes. HbA1c was selected as the primary blood biomarker because it is a simple, minimally invasive measure that does not require the person to be in a fasted state, thus allowing for flexible testing capabilities. HbA1c is an indicator of three-month glycemic control and is less variable than fasted blood glucose sampling on a day-to-day basis. HbA1c has also been adopted as part of the prediabetes and type 2 diagnostic criteria by CDA as well as the American Diabetes Association (ADA) [6, 17]. For these reasons, HbA1c is a highly appropriate biomarker for the evaluating the validity of the risk questionnaire.
HbA1c was analyzed using the Bio-Rad in2it (Bio-Rad Laboratories, Hercules, CA) point-of-care device and boronate affinity chromatography. All capillary blood samples were collected by a trained phlebotomist and sterile techniques were utilized in accordance with York University biosafety and ethics requirements. In a sub-set of individuals, a second HbA1c sample (from the same fingerstick) was collected using Bio-Rad capillary tubes for analysis using high-performance liquid chromatography (HPLC), a standardized HbA1c analysis criterion method that is in accordance with National Glycohemoglobin Standardization Program regulations. The HPLC analyses described above were performed by Clearstone Central Laboratories (Mississauga, ON) using the Bio-Rad Variant II Hemoglobin testing system.
Results of the HbA1c tests were interpreted based on the 2013 Canadian Diabetes Association clinical practice guidelines diagnostic criteria [6] which define prediabetes using an HbA1c range of 6.0-6.4% and type 2 diabetes using an HbA1c range of ≥6.5% [6]. It should be noted that the ADA use an HbA1c range of 5.7-6.4% for prediabetes and ≥6.5% for diabetes [17]. Participants were informed that the results from the blood tests taken for the PRE-PAID project were not designed to provide medical diagnosis of prediabetes or type 2 diabetes. Individuals who had HbA1c scores ≥6.5% were provided with a letter describing their results and encouraged to see their primary care physician for further confirmatory testing.
Statistical analyses
Descriptive statistics as well as frequencies of questionnaire responses were analyzed for all participants who completed the risk questionnaire. Various exclusions within the dataset took place for further analyses based on missing data that was attributable to participant error, data entry error, or the participant’s unwillingness to provide a blood sample. Figure 1 shows the participant flow diagram for the PRE-PAID risk questionnaire administration.
A comparison of the two methods for determining HbA1c was performed using the Bland-Altman method [18] to detect any potential biases between the two methods of analysis. All analyses described in this investigation were performed using a two-sided 5% level for significance.
ANOVA with post-hoc pairwise comparisons was performed using Tamhane’s T2 approach, which allows for unequal variances to compare risk classification based on the questionnaire score to mean HbA1c measured using the Bio-Rad device. Prior to analysis, the "Very High" and "Extreme" groups were merged because of a very small number of participants falling within the "Extreme" classification. From a clinical perspective, individuals within both of the highest risk groups would be strongly encouraged to visit a physician for further assessment regardless. In addition to the ANOVA, additional analyses including the area under the receiver-operator curve and examination of sensitivity and specificity were performed to examine reliability. These analyses used a cut-point of 6.5 which corresponds to the "moderate" risk category to better describe the ability of the risk questionnaire to predict dysglycemia defined by HbA1c ≥ 6.0%.
Finally, step-wise, backward elimination linear regression was performed to quantify the amount of variance in HbA1c values that was attributable to each of the variables included on the risk questionnaire. The Bland Altman plots were performed using GraphPad Prism 6 and all other analyses were performed using SPSS version 20.