Trial design

We conducted a multiclinic open-label, comparative effectiveness, RCT at 6 clinic sites. Our protocol was approved by the institutional review board at the McGovern Medical School (HSC-MS-18-0575) and registered in ClinicalTrials.gov before starting enrollment (NCT 03620890). All participants provided written informed consent to participate in the research study.

Participants

Women with a singleton gestation and type 2 or overt type 2 diabetes mellitus who sought obstetrical care at ≤21 weeks’ gestation were screened for eligibility. Overt type 2 diabetes mellitus was defined according to the American Diabetes Association criteria, including people with overt diabetes mellitus diagnosed before 20 weeks’ gestation using either hemoglobin A1c (HbA1c) of ≥6.5 or fasting glucose of ≥126 mg/dL or random blood glucose of ≥200 mg/dL or 2-step method (50 g glucose challenge test of >135 mg/dL followed by 100 g glucose challenge test with at least 2 values above thresholds: fasting blood glucose of >90 mg/dL, 1-hour postprandial value of >180 mg/dL, 2-hour postprandial value of >155 mg/dL, and 3-hour postprandial value of >140 mg/dL). ^,

Those with type 1 diabetes mellitus, age younger than 18 years, major fetal anomaly, allergy or previous adverse reaction to NPH or insulin detemir, diabetes nephropathy (Cr≥1.5), or diabetic proliferative retinopathy were excluded. We defined the inclusion and exclusion criteria during our study design phase, before the study started, and made no changes after enrollment started.

Randomization and masking

Participants were randomly allocated in 1:1 ratio to receive either insulin detemir or NPH prescriptions by the treating clinicians. A computer-generated randomization sequence was prepared, with permuted blocks, and was stratified according to study site by the study statistician. The medication based on the computer-generated list was typed out on a piece of paper with the insulin regimen. The piece of paper was placed in an opaque envelope and numbered according to the computer-generated list. Envelopes were kept secured at the clinic sites and were managed by the research team. Owing to the nature of the intervention, treating clinicians and the participants could not be blinded for their assigned allocation.

Trial interventions

Following allocation, the participants were instructed to start treatment on the same day. Individual’s weight was obtained at the visit and the initial daily total insulin dose of detemir or NPH, and the short-acting insulin was determined on the basis of the weight and treatment according to the recommended guidelines. In the first trimester of pregnancy, the patient weight in kilograms was multiplied by 0.7, in the second trimester of pregnancy by 0.8, and in the third trimester of pregnancy by 0.9 for the total daily units of insulin. ^, In the insulin detemir group, the total daily dose of insulin was divided into 50% insulin detemir and 50% short-acting insulin. Based on clinician’s preference, insulin detemir dose could be divided in half and then injected in the morning and evening, 12 hours apart. Short-acting insulin was divided with one-third each being injected with breakfast, lunch, and dinner meals. In the NPH insulin group, approximately 60% of the total daily dose was given in the morning and 40% in the evening. For the total morning dose, two-thirds was NPH and one-third was the short-acting insulin. The total evening dose was halved: the short-acting insulin injected at dinner, whereas the NPH was injected at bedtime.

Other than the type of long-acting insulin, participants were managed according to usual obstetrical care. ^, Participants were instructed to check glucose levels at fasting and 1 or 2 hours after meal. The treating clinicians adjusted the insulin dose to maintain good glucose control based on accepted targets: fasting glucose of <95 mg/dL, 2-hour postprandial values of <120 mg/dL, overall mean glucose of <100 mg/dL. According to our usual practice, each individual had at least 1 meeting with a registered dietitian and diabetes care and education specialist to learn how to self-administer insulin and were followed up in a Maternal-Fetal Medicine clinic. Participants were followed up every 1 to 2 weeks, or as necessary, determined by their glucose control, until delivery.

Use and type of short-acting insulin, insulin adjustment, and the timing of delivery were at the discretion of each participant’s maternal-fetal-medicine subspecialist.

Trial outcomes

The primary outcome was a composite of adverse neonatal complications and consisted of 1 or more of the following: shoulder dystocia (defined as the need for any extra maneuvers, other than gentle downward traction of the fetal head to deliver the fetal body after the fetal head has been delivered), LGA (defined as birthweight over the 90th percentile of the expected value according to gestational age [GA]), NICU admission, neonatal hypoglycemia (blood sugar of <40 mg/dL in the first 24 hours of life or <50 mg/dL after or need for medical therapy), respiratory distress (need for at least 4 hours of respiratory support with supplemental oxygen, continuous positive airway pressure (CPAP) or ventilation at the first 24 hours of life). Neonatal secondary outcomes included GA at delivery, small for GA (defined as birthweight below the 10th percentile for GA), 5-minute Apgar score of <7, lowest glucose level, need for intravenous glucose, respiratory distress syndrome (RDS; defined as the presence of clinical signs of respiratory distress with oxygen requirement and a classical chest x-ray finding), need for mechanical ventilation or CPAP, neonatal jaundice requiring therapy, brachial plexus injury, and hospital length of stay.

Secondary maternal outcomes included hypoglycemic events (defined as <60 mg/dL), antepartum hospital admission for glucose control, hypertensive disorder of pregnancy, maternal weight gain (from randomization to delivery), cesarean delivery, and postpartum complications, such as wound infection and endometritis.

The maternal demographic, antepartum, labor course, and outcomes were extracted from the electronic medical record and collected using Research Electronic Data Capture (REDCap) software. Furthermore, the neonatal outcomes until discharge were culled and entered in REDCap.

Sample size

The sample size was calculated on the basis of previous data on composite adverse neonatal outcomes of 43% in women with type 2 diabetes mellitus. We used the Bayesian statistics to estimate a sample size using a probability of 75% or greater for superiority. Given the evidence of noninferiority or superiority of detemir compared with NPH, the fact that both treatments are commonly used, there is no evidence of important differences in side effects, and there is no difference in insurance coverage during pregnancy, we decided that a threshold of >75% posterior probability of reducing neonatal adverse outcomes by any amount would justify the use of detemir in this population. A sample size of 108 provided 80% power to identify the most effective treatment, assuming the NPH group had a true adverse outcome rate of 43% and the true adverse rate was 29% in the detemir group (33% reduction). Of note, we designed this study as a Bayesian RCT to be able to provide probabilities of benefit or harm from either intervention. It is not possible to calculate these quantities from frequentist statistics. Furthermore, P values and confidence intervals (CIs) are often misinterpreted and incorrectly dichotomized into “significant” and “nonsignificant” results. To avoid these pitfalls, we chose a Bayesian design and analysis that is less prone to these issues while providing probabilities that can be directly used for decision-making.

Data analysis

Intention-to-treat analyses were conducted under a Bayesian framework. The probability and magnitude of treatment benefit of insulin detemir compared with NPH were assessed for each outcome. There were 3 different types of outcomes: binary (eg, composite primary outcome), discrete (total days in hospital), and continuous (eg, weight gain). Binary outcomes were evaluated with a logistic model. A negative binomial model was used to assess discrete outcomes. For both cases, we conservatively set a neutral prior probability centered at a relative risk (RR) of 1.0 (ie, a priori no effect on the outcome) and a 95% credible interval (CrI) of 0.25 – 4.0. Continuous outcomes were analyzed with a linear regression model using the default prior distribution provided by the brms package. All models included the stratifying variable clinic site as a covariate. We report the estimated posterior median RR or mean group difference with a 95% CrI for each outcome. A 95% CrI is interpreted as there being a 95% probability of the true RR being in this interval. The probability of treatment benefit was determined by the posterior probability that an RR (for binary and discrete outcomes) or a mean group difference (for continuous outcomes) is less than 1 or 0, respectively. To validate that there is no effect of fetal loss (ie, first-trimester miscarriage, termination of pregnancy); we conducted a posthoc sensitivity analysis. Under the same Bayesian framework, the probability of treatment benefit and its CrI were obtained on basis of the outcome as a composite of adverse neonatal outcomes and/or fetal death.

All analyses were conducted in R software (version 3.6.2) using the brms package.