Integrated database
Individual participant data from 2 randomized clinical trials sponsored by Eli Lilly and Company [20, 21] and 1 randomized trial sponsored by Eli Lilly and Company and Boehringer Ingelheim [22] were used for meta-analyses. These studies were identified by an exhaustive search of Eli Lilly and Company’s integrated clinical trial database based on the following inclusion criteria: 1) participants were insulin-naïve with T2DM, 2) a sufficient number of participants received IGlar (Basaglar®/Abasaglar®, Boehringer Ingelheim and Eli Lilly and Company, or Lantus®, Sanofi-Aventis) for at least 24 weeks, and 3) IGlar was the only insulin component in the antihyperglycemic treatment.
Buse et al. [20] compared the efficacy and safety of twice-daily (BID) insulin lispro mixture 75/25 and once-daily (QD) IGlar in a randomized, open-label, 24-week, non-inferiority trial conducted in 11 countries (Argentina, Australia, Brazil, Canada, Greece, Hungary, India, the Netherlands, Romania, Spain, and the United States). Eligible participants were insulin-naïve adults ≥ 18 years of age with T2DM and taking ≥ 2 oral antihyperglycemic medications (OAMs). Jain et al. [21] compared the efficacy and safety of 2 progressive insulin regimens, QD insulin glargine plus insulin lispro administered up to 3 times daily (TID) versus insulin lispro mixture 50/50 administered up to TID, in a randomized, open-label, 36-week, non-inferiority trial conducted in 9 countries (Australia, Canada, France, Greece, India, Republic of Korea, Mexico, Russian Federation, and Spain). Eligible participants were insulin-naïve adults ≥ 18 years of age with T2DM and taking ≥ 2 OAMs. Rosenstock et al. [22] compared the efficacy and safety of 2 IGlar products, LY IGlar versus Lantus®, in a phase 3, randomized, double-blind, 24-week, non-inferiority trial conducted in 11 countries (Czech Republic, France, Germany, Greece, Hungary, Italy, South Korea, Mexico, Poland, Spain, and the United States). Eligible participants were adults ≥ 18 years of age with T2DM who were either insulin-naïve or previously on Lantus® and taking ≥ 2 OAMs.
In each of these trials [20,21,22], postprandial glucose (PPG) was collected through 7-Point self-monitored blood glucose (SMBG) at weeks 0, 12, and 24 on 3 separate days in the 2-week period prior to each visit. Participants were also given a diary to record hypoglycemia events experienced throughout each study. For each hypoglycemia episode, the participant was asked to record the glucose value, if measured, and to describe the treatment, including if the participant was able to self-treat, and the outcome of the episode. At the onsite visit, the investigator reviewed the diary with the participant to verify and assess any need for treatment adjustment. Further study design details and key outcomes from each trial are summarized in Additional files 1 and 2, respectively.
Statistical analysis
Participants with non-missing HbA1c at 24 weeks were analyzed and classified into 2 responder cohorts (yes vs no) according to the composite HbA1c responder measure: HbA1c < 53 mmol/mol (< 7%) at 24 weeks or reduction in HbA1c from baseline to 24 weeks ≥ 11 mmol/mol (≥ 1%). Responders were the participants who either had HbA1c < 53 mmol/mol (< 7%) or had HbA1c reduction ≥ 11 mmol/mol (≥ 1%) at 24 weeks. Nonresponders did not meet either criterion. For the 3 SMBG profiles obtained at 0, 12, and 24 weeks, the average SMBG value was used for analysis. Missing data at week 24 was expected due to the self-monitoring nature of SMBG and the attrition throughout the trials. Overall, about 60% of participants had non-missing PPG data at week 24. More specifically, among the 1485 patients who had non-missing HbA1c values at 24 weeks, 826 participants had non-missing glucose values post-breakfast and post-midday meal and 829 participants had non-missing post-dinner glucose at week 24. This was considered adequate for analysis with the majority of participants contributing data and a total sample size of > 800.
Heterogeneity across the studies was assessed by study-by-responder interaction. P values for interaction were nonsignificant for the majority of outcomes measured indicating that results in these trials were relatively homogeneous and therefore justified the integration of these data. In 2 of the 3 studies analysed [20, 22], responders had greater reductions in fasting blood glucose (FBG) than nonresponders, while in the 3rd study [21], the 2 cohorts had similar reductions in FBG, thereby resulting in a statistically significant (P = 0.012) study-by-responder interaction (Additional file 2). Baseline participant characteristics and clinical profiles at 24 weeks (HbA1c, FBG, PPG, insulin dose, and hypoglycemia categories [total hypoglycemia (BG ≤ 3.9 mmol/L [≤ 70 mg/dL] or signs/symptoms), documented symptomatic (BG ≤ 3.9 mmol/L [≤ 70 mg/dL] and signs/symptoms), nocturnal (between bedtime and waking), and severe (required 3rd party assistance) were compared between the responder cohorts. A 2-sided P value of < 0.05 was considered statistically significant, with P values based on the Pearson’s Chi-square test for categorical variables and fixed effects meta-regression model for continuous variables. Results presented are model-adjusted mean and standard error (SE). Relationships between improvements in glycemic outcomes as continuous variables (FBG, daily mean PPG, and HbA1c) and baseline variables (HbA1c, FBG), between improvements in glycemic outcomes as continuous variables (FBG, daily mean PPG, and HbA1c) and insulin dose, and between hypoglycemic rate and insulin dose, were explored graphically using scatter plots as post hoc analyses.
Sensitivity, specificity, positive predictive value, and negative predictive value were evaluated to assess if early response at 12 weeks could predict subsequent response at 24 weeks to support current guidelines that recommend evaluation of therapeutic response to pharmacologic interventions at 12 weeks after initiating therapy.
All analyses were performed using SAS Version 9.2® or higher.