Participants
Women were assessed for study eligibility by scrutinizing their medical records when they were admitted for delivery to the antenatal and labor wards of the Obstetrics Unit. Inclusion criteria were women presenting for imminent delivery (in spontaneous labor, scheduled induction of labor or planned cesarean) at our labor or antenatal wards who were aged ≥ 18 years, with a singleton pregnancy and at term gestation (≥ 37 weeks confirmed by ultrasound before 22 weeks gestation). We excluded women with severe-moderate to severe anemia in pregnancy (hemoglobin level < 8 g/dl) [9, 10], known major hemoglobinopathy [11, 12], known gross fetal anomaly (as these characteristics might have major impact on HbA1c assessment or birth weight) and inability to consent due to language difficulty.
Eligible women were approached, provided with the Patient Information Sheet and had oral queries answered by the recruiting investigator (co-author MYNF). Written informed consent to participate in the study was obtained from all participants. All participants’ relevant characteristics including diagnosis of prepregnant diabetes mellitus or gestational diabetes mellitus (GDM), current use of antiglycemic agents, hypertension in pregnancy, positive Group B streptococcus culture during pregnancy, obstetric history e.g., previous Cesarean and parity were transcribed onto the Case Report Form.
Recruitment and interventions
Women who planned delivery at our center were routinely screened for gestational diabetes with the 75-g OGTT (based on Malaysian GDM screening criteria and diagnostic thresholds: fasting ≥ 5.1 and/or 2-h ≥ 7.8 mmol/l) [13] at booking and/or 24–28 weeks gestation depending on risk factors, hepatitis B, HIV infection, and had dating ultrasound in early pregnancy. Women with diabetes in pregnancy were monitored by their blood sugar profiles through self-monitoring of blood glucose. In women with diabetes in pregnancy, delivery (usually by labor induction unless contraindicated) is arranged by no later than 40 weeks gestation or earlier if any concerning clinical factors were present. Women who delivered at our center usually had a full blood count taken at their birth admission amongst other indicated blood tests if any.
Three milliliters of venous blood were drawn from participants, typically piggy-backed to venipuncture for routine bloods or at insertion of an indwelling intravenous catheter for delivery according to our care protocol. The blood was placed in a EDTA (ethylenediaminetetraacetic acid) blood bottle and dispatched to our hospital laboratory for immediate processing to establish HbA1c level. Our laboratory utilized Biorad Variant 2, Chemopharm, Selangor, Malaysia to run the blood samples using high performance liquid chromatography.
The predelivery HbA1c results were not revealed to participants and care providers.
Outcome measures
Primary outcomes were Cesarean delivery and LGA (≥ 90 centile for gestational age birth weight), which were two of the four primary adverse primary outcomes of the original 2008 HAPO study [8] and that are also used in the setting of GDM diagnostic thresholds by IADPSG in 2010 [14].
Sample size calculation
For sample size calculation the following principles were considered: “for regression equations using six or more predictors, an absolute minimum of 10 participants per predictor variable is appropriate” [15] and “a minimum of 10 cases with the least frequent outcome for each independent variable in your model” [16].
We assumed 10 independent variables in the model and probability of the least frequent outcome is 0.10, hence the sample size calculated is 10 × 10 / 0.10 = 1000. In our center, the Cesarean delivery rate was about 30%, hence a sample size of 1000 was expected to yield 300 Cesarean events and with LGA defined as birth weight ≥ 90th centile corrected for gestational age, sample size of 1000 should yield 100 LGA events. Both these number of event estimates should permit a robust binary logistic regression analysis model of up to 10 independent covariables whilst keeping to the 10 event per variable rule.
Statistical analyses
Data were entered into a statistical software package SPSS (Version 23, IBM, SPSS Statistics). The collected participant characteristics were analyzed against the primary outcomes of Cesarean vs. non-Cesarean delivery and LGA vs. non-LGA separately. Birth weight was included as a surrogate for estimated fetal weight [17] in the model for Cesarean delivery. The t test was used to compare means of continuous data, Mann–Whitney U test for non-parametric data and Chi-square test to analyze categorical data to yield crude results. In adjusted analysis, independent co-variables (identified confounders) with p < 0.05 on crude analysis were incorporated into the multivariable binary logistic regression model to control for confounders on the impact of HbA1c level on the primary outcomes. Post hoc adjusted analyses on different subsets of our study population and also for other adverse outcome (postpartum hemorrhage) were performed incorporating HbA1c level and all the initially identified confounders. Two-sided P values were reported and P < 0.05 was considered as significant.