This study was a case control design involving the linkage of data extracted from the Newfoundland and Labrador Diabetes Database (NLDD) with the Live Birth System (LBS). The NLDD is maintained by the Janeway Pediatric Research Unit at the Janeway Child Health Care Centre (JCHCC) in St. John's, NL. The JCHCC is the only tertiary care pediatric hospital in the province. The NLDD includes data for the majority of cases of T1DM diagnosed in NL from 1987 to present. Children are included in the database as part of a prospective provincial study on the epidemiology of T1DM in NL. They have a confirmed diagnosis of T1DM according to Canadian Diabetes Association (CDA) criteria [3
]. The LBS is maintained by the Newfoundland and Labrador Centre for Health Information (NLCHI). Data for this system are acquired from Live Birth Notification forms that are completed in all provincial health care facilities. The forms are provided to NLCHI by the Vital Statistics Division, Service NL, and contain clinical and demographic data for all births (resident and nonresident) in the province. The system currently contains data on all births from 1992 to 2011.
Patient records were individually linked across datasets. Cases included children born in NL from 1992 onwards which have been diagnosed with T1DM before 15 years of age. Children with type 2 diabetes, maturity-onset diabetes of youth, transient hyperglycemia, and diabetes caused by chemotherapy or cystic fibrosis are excluded from the NLDD and thus were not included in the study. Children born prior to 1992 were not included in the study because there are no electronic birth notification data available before this date.
A unique identifier, such as the provincial health insurance plan number, was not available for all children in the NLDD. As a result case subjects were linked to the LBS using child's date of birth and mother's maiden name. Where available, child's name was used to verify the linkage. Of the 301 cases in the NLDD, 23 were excluded because they were born out of province. Of the remaining 278, 6 were excluded as duplicate records. Linkage was possible for all but six children resulting in a total of 266 cases included in the study. Three control subjects (N
= 798) were selected for each case matched on year of birth, sex, and health authority of mother's residence at time of delivery. Power analysis was performed to determine whether this sample size would be sufficient to detect statistical significant associations between T1DM and the risk factors of interest. The power analysis was conducted considering an overall rate of birth by C-section in NL as 30.9% [4
], in order to achieve a power of 80% with a desired odds ratio of 1.5. Using the method described by Kelsey et al. [5
] the power analysis confirmed that a sample size of 266 cases and 798 controls is sufficient to detect statistically significant relationships between T1DM and delivery by C-section.
Cases and controls were grouped into two gestational age categories: preterm and term/postterm. Birth weight in grams was used to classify cases and controls as high birth weight (>4,000 grams) or not (≤4,000 grams). Cases and controls were also classified as small/appropriate-for-gestational-age or large-for-gestational-age using the method described by Kramer et al. [6
]. Method of delivery was categorized as vaginal or C-section. Cases and controls were grouped according to parity or birth order as either 1 and 2 or more. Mother's age in years was classified as ≤34 years or >34 years. Mothers were also classified by their T1DM status and hypertension status. Mother's marital status was categorized as married, single, separated, widowed, or divorced. Mother's education level was classified into three categories: not graduated high school, graduated high school, and education beyond high school.
Descriptive statistics were generated to describe the distribution of cases and controls. Demographic and clinical factors of mothers, including age, marital status, education, place of residence, parity (number of live born children delivered), and complications of pregnancy, were included. Cases and controls were analyzed by sex, place of residence, age of onset, length of gestation, type of delivery, birth weight, size for gestational age, and birth order.
Chi-square tests were used to predict diabetes status on the basis of the independent variables. Conditional multiple logistic regression was used to assess the relationship between T1DM risk and the variables of interest. Two conditional logistic regression models were employed. The first model contained birth weight, gestational age, parity, delivery method, mother's marital status, mother's education level, mother's age, maternal hypertension, and mother's T1DM status. The second model incorporated all variables in the first model with the exception of birth weight and gestational age which were replaced with size-for-gestational-age. Birth weight and gestational age were not included in the same model as size-for-gestational-age as they are components of this variable.
The Statistical Package for the Social Sciences (SPSS) 17.0 was used to generate descriptive statistics and chi-squares. SAS 9.2 was used to conduct the conditional multivariate logistic regressions.
This study received approvals from the Human Investigation Committee of Memorial University, from each of the hospital boards and the Secondary Uses Committee of the Newfoundland and Labrador Centre for Health Information (NLCHI) prior to commencement.