The trial design of ONWARDS 5 has previously been published [1, 5]. Briefly, ONWARDS 5 was a 52-week, open-label, randomized, multinational, phase 3a trial with real-world elements. Eligible participants were adults (≥ 18 years old) with T2D requiring insulin initiation, with an HbA1c value over 7.0%, receiving any noninsulin glucose-lowering medications. Participants were assigned to degludec, glargine U100, or glargine U300 at the discretion of the investigator at the initial screening visit and were subsequently randomized 1:1 to receive either the assigned OD analog (degludec [n = 378], glargine U100 [n = 96], glargine U300 [n = 69]) or icodec with app (n = 542). Real-world elements were incorporated to mimic standard clinical practice including broad inclusion criteria, choice and dosing of the comparator OD analogs at the investigator’s discretion, and a low number of individual trial-site visits (approximately 3 months apart) for both treatment groups.

The ONWARDS 5 trial was conducted in accordance with the principles of the Declaration of Helsinki and the Good Clinical Practice guidelines of the International Conference for Harmonisation. The protocol, consent form, and other relevant documents were reviewed and approved by the appropriate institutional review boards or independent ethics committees. All participants provided written informed consent for participation and publication of data before trial entry.

To create a larger cohort of participants with glargine U300 for comparison with icodec with app, the ONWARDS 5 clinical trial data were augmented with RWD collected between 2015 and 2019 from the US Ambulatory Electronic Medical Records (AEMR) database (Fig. 1). The AEMR database is commercially available, contains information from outpatient clinics and physician’s offices, and provides a comprehensive and robust repository of patient data. The database captures a wide range of data, including demographics, diagnoses, laboratory results, procedures, and prescriptions.

Augmentation of ONWARDS 5 RCT data with RWD to enable a robust comparative analysis of icodec with app and glargine U300 in adults with T2D. glargine U300 insulin glargine U300, icodec with app insulin icodec titrated with a dosing guide app, RCT randomized controlled trial, RWD real-world data, T2D type 2 diabetes

When conducting an augmented study, the supplementary RWD should emulate the components of the RCT protocol as best as possible, particularly with regard to participant eligibility, treatment assignment compatibility, follow-up period, and statistical analyses [12]. Accordingly, the inclusion criteria used for extracting participant data from the AEMR database were: at least one prescription claim for basal insulin (degludec, glargine U100, glargine U300) during the index window (January 1, 2015 to August 31, 2020); at least two fills of the index treatment (including the fill on the index date itself); enrollment at least 360 days before the index date (the 1-year pre-index period); continuous enrollment for at least 360 days from (and including) the index date (the 1-year post-index period); at least 18 years of age at index; consistent and known sex; at least one diagnosis on a confirmatory medical claim for T2D in any period before the index date (including the index date itself) defined by diagnosis or by at least two noninsulin glucose-lowering medication claims; at least one measure of HbA1c in the 90 days before the index date; and at least one measure of HbA1c in the 365 days after the index date.

The primary outcome for this post hoc analysis was change in glycemic control (HbA1c levels) from the index date to 52 weeks after the index date or to drop out (e.g., discontinuation). For ONWARDS 5 participants and AEMR controls, missing data were imputed using multiple imputation (e.g., for individuals who discontinued or were lost to follow-up). Discontinuation in ONWARDS 5 was defined per protocol regarding rescue medication allowance (noninsulin glucose-lowering medications were maintained or altered at the investigator’s discretion, whereas basal-bolus intensification was a reason for treatment discontinuation); for AEMR data, discontinuation was based on individual electronic medical record (EMR) prescription fill/claim start and stop dates. The index date (time 0) was set for AEMR participant data as the start of basal insulin treatment to mimic the ONWARDS 5 trial.

The primary analysis of ONWARDS 5 adhered to the treatment policy estimand for the analysis of HbA1c, and missing values were imputed based on trial participants who had a measurement at the week 52 visit but who discontinued treatment before the week 52 visit (retrieved dropouts). This type of multiple imputation differed from the sequential multiple imputation by Bayesian linear regression performed herein for AEMR missing values (see Supplementary Material for further details).

Propensity score matching between participant data from ONWARDS 5 and EMRs from the AEMR database was performed to ensure appropriate matching between RWD and RCT participant data. The 1:1 propensity score matching was performed based on age, HbA1c, and noninsulin glucose-lowering medication use (glucagon-like peptide-1 receptor agonist [GLP-1 RA], sodium-glucose transport protein 2 inhibitor [SGLT2i], dipeptidyl peptidase 4 inhibitor [DPP4i]) at baseline. Table S1 shows the matches between ONWARDS 5 and AEMR data that facilitated the calculation of trial effects and the treatment comparison of icodec with app against glargine U300.

The strategy for the analysis was to generate an augmented synthetic glargine U300 data set (by replacing the ONWARDS 5 degludec and glargine U100 control groups) suitable for comparison against ONWARDS 5 icodec with app data. This ‘augmented comparison’ would then supplement the original ‘randomized comparison’ between glargine U300 and icodec with app generated from ONWARDS 5 (Fig. S1). The effects of the two comparisons were weighted based on their relative sample sizes, and a weighted mean was reported.

A trial effect is considered to be a benefit that participants may experience when they are enrolled in RCTs simply because of the fact they are participating in an RCT [13]. These trial effects need to be accounted for when performing comparisons of augmented clinical data. The generation of the augmented data set was conceptually performed in two steps: (1) the estimation of the trial effect based on trial participants who received OD analogs, then (2) the application of the trial effect to real-world glargine U300 matched participants to create glargine U300 synthetic controls. Figure 2 presents a schematic illustrating these two steps. During the initial step, propensity score matching was used to identify individuals from the available AEMR database population who matched the characteristics of the participants from the degludec, glargine U100, and glargine U300 control arms of ONWARDS 5. The change in HbA1c from baseline at 52 weeks between ONWARDS 5 and AEMR participants was estimated to identify the trial effect by stratum (i.e., degludec, glargine U100, and glargine U300). The estimated trial effects were then combined in a fixed effects meta-analysis to produce a single trial effect estimate and associated uncertainty. For the second step, ONWARDS 5 participants assigned to degludec or glargine U100 at screening but randomly assigned to receive icodec with app were matched with AEMR participants who received glargine U300. Subsequently, the change from baseline in HbA1c between the icodec with app group and the glargine U300 synthetic control group was calculated, taking into account the overall trial effect estimated during the initial step.

Conceptual overview of the two-step process to augment the ONWARDS 5 data with RWD from the AEMR database. AEMR Ambulatory Electronic Medical Records, DPP4i dipeptidyl peptidase-4 inhibitor, glargine U300 insulin glargine U300, GLP-1 RA glucagon-like peptide-1 receptor agonist, HbA1c glycated hemoglobin, icodec with app insulin icodec titrated with a dosing guide app, OD once-daily, RWD real-world data, SGLT2i sodium-glucose transport protein 2 inhibitor

For estimation, a joint statistical model for the propensity-score-matched ONWARDS 5 and AEMR data was developed to account for the covariance generated by the contribution of ONWARDS 5 glargine U300 data to both the trial effect calculations and the treatment comparison against icodec with app (see Supplementary Material—statistical appendix for further details).

The primary analysis assumed full augmentation. Acknowledging that clinical observers may differ in their belief about the relevance of augmenting clinical trial data with RWD, an additional parameter, ρ, was introduced to control for augmentation plausibility. The parameter 0 ≤ ρ ≤ 1 corresponded to the belief in the plausibility of augmentation (ρ = 0 corresponded to a belief that augmentation provided no additional information, whereas ρ = 1 corresponded to a belief that augmentation was fully relevant). At the extremes, ρ = 0 represented only the glargine U300 stratum-specific randomized comparison, whereas ρ = 1 represented the fully augmented (primary) analysis.

Sensitivity analyses for trial effect, emulation criteria, and emulation timing factors were conducted to test the robustness of the evaluated outcome. The relevance of using degludec and glargine U100 trial effects to partly represent the trial effect for glargine U300 could be subject to scrutiny; therefore, a sensitivity analysis was conducted by incorporating an additional parameter into the model: 0 ≤ φ ≤ 1, which corresponded to the belief in the weighted mean trial effect estimation (φ = 0 corresponded to a belief that only the glargine U300-matched comparator group sufficiently contributed towards a glargine U300 trial effect, whereas φ = 1 corresponded to a belief that the full weighted trial effect was as calculated in the primary augmented analysis). To test the effect of the imputation method, another sensitivity analysis was conducted while adopting a last observation carried forward imputation strategy. Additionally, to explore how AEMR emulation criteria affected results, further emulations were conducted testing a range (from narrow to broad) of different emulation criteria (Table S2). Finally, timing factors of the AEMR emulation were also explored via a sensitivity analysis, in which AEMR participants were selected with time 0 closer in proximity to the recorded baseline results of participants in the ONWARDS 5 trial.