Clinical outcomes in high‐hypoglycaemia‐risk patients with type 2 diabetes switching to insulin glargine 300 U/mL versus a first‐generation basal insulin analogue in the United States : Results from the DELIVER High Risk real‐world study

Abstract Aims To compare 12‐month clinical effectiveness of insulin glargine 300 units/mL (Gla‐300) versus first‐generation basal insulin analogues (BIAs) (insulin glargine 100 units/mL [Gla‐100] or insulin detemir [IDet]) in patients with type 2 diabetes (T2D) who were at high risk of hypoglycaemia and switched from one BIA to a different one (Gla‐300 or Gla‐100/IDet) in a real‐world setting. Methods DELIVER High Risk was a retrospective observational cohort study of 2550 patients with T2D who switched BIA to Gla‐300 (Gla‐300 switchers) and were propensity score‐matched (1:1) to patients who switched to Gla‐100 or IDet (Gla‐100/IDet switchers). Outcomes were change in glycated haemoglobin A1c (HbA1c), attainment of HbA1c goals (<7% and <8%), and incidence and event rates of hypoglycaemia (all‐hypoglycaemia and hypoglycaemia associated with an inpatient/emergency department [ED] contact). Results HbA1c reductions were similar following switching to Gla‐300 or Gla‐100/IDet (−0.51% vs. −0.53%; p = .67), and patients showed similar attainment of HbA1c goals. Patients in both cohorts had comparable all‐hypoglycaemia incidence and event rates. However, the Gla‐300 switcher cohort had a significantly lower risk of inpatient/ED‐associated hypoglycaemia (adjusted odds ratio: 0.73, 95% confidence interval: 0.60–0.89; p = .002) and experienced significantly fewer inpatient/ED‐associated hypoglycaemic events (0.21 vs. 0.33 events per patient per year; p < .001). Conclusion In patients with T2D at high risk of hypoglycaemia, switching to Gla‐300 or Gla‐100/IDet achieved similar HbA1c reductions and glycaemic goal attainment, but Gla‐300 switchers had a significantly lower risk of hypoglycaemia associated with an inpatient/ED contact during 12 months after switching.


| INTRODUC TI ON
According to the United States (US) 1 , more than 34 million Americans (10.5% of the population) had diabetes, the vast majority of whom had type 2 diabetes (T2D). 2 Although T2D can be controlled in the early stages with metformin and lifestyle modifications, the progressive nature of the condition means that many patients will progress through successive levels of treatment intensification and will eventually require basal insulin. 3,4 In randomized controlled trials in patients with T2D, the secondgeneration basal insulin analogue (BIA), insulin glargine 300 U/mL (Gla-300), has been compared with the first-generation BIA, insulin glargine 100 units/mL (Gla-100), and has been shown to provide comparable glycaemic control while reducing the risk of some hypoglycaemic endpoints. [5][6][7][8][9] These results have now also been confirmed in various meta-analyses. [10][11][12] Earlier real-world studies of Gla-300 have largely focused on a general diabetes population who were either initiating BIAs (DELIVER Naïve, 13 DELIVER Naïve D, 14 LIGHTNING 15 ) or were switching from another BIA (DELIVER 1, 16 DELIVER 2, 17 DELIVER D, 18 DELIVER D+, 19 LIGHTNING 15 ). While the DELIVER 3 study examined the treatment effectiveness of Gla-300 in older patients, 20 and DELIVER D+ 19 and LIGHTNING 15 reported on some higher risk subgroups, several other important subgroups expected to be at increased risk of hypoglycaemia have yet to be studied in detail.
Additionally, the advantages of second-generation BIAs such as Gla-300 for patients at high risk of hypoglycaemia in real-world settings have not been well characterized beyond 6 months.
Several characteristics have been shown to confer an increased hypoglycaemia risk in patients with T2D, including older age, 21 renal impairment, [22][23][24] the presence of multiple comorbidities including diabetes comorbidities, 25 and cardiovascular disease, 26 long-standing diabetes and/or insulin use, or the use of multiple daily injections or sulphonylureas. 27 The objective of the DELIVER High Risk study was to compare the 12-month clinical effectiveness of Gla-300 versus firstgeneration BIAs (Gla-100 or insulin detemir [IDet]) in patients with T2D who were at higher risk of hypoglycaemia and were switching from one BIA to a different BIA (Gla-100 or IDet to Gla-300, and from Gla-100 to IDet or from IDet to Gla-100).

| Study design and data source
DELIVER High Risk was a retrospective, observational cohort study examining patients with a diagnosis of T2D treated in real-world clinical practice settings and deemed at high risk of hypoglycaemia, who were switching from one BIA to a different BIA: from either Gla-100 or IDet to Gla-300 ('Gla-300 switchers'); or from Gla-100 or IDet to IDet or Gla-100 ('Gla-100/IDet switchers'). Only patients switching to and from these BIAs were included; patients switching from neutral protamine Hagedorn insulin were not included. Retrospective data were obtained from Accenture's Predictive Health Intelligence Environment (IBM Explorys, Cleveland, Ohio), which provides electronic medical record (EMR) data for approximately 18% of the US population. This database contains a geographically diverse spectrum of longitudinal medical data from 39 major integrated delivery networks and, importantly, represents a broad mix of patients enrolled in privately insured and government-sponsored healthcare programmes. The database includes information on patient clinical and demographic characteristics, insurance status, healthcare encounters, diagnoses, procedure codes, and associated laboratory values and surgeries, for approximately 50 million patients.

| Study population
Patients were included in the study if they were aged ≥18 years on the index date and had: (a) ≥1 diagnosis of T2D according to ICD-9-CM/ICD-10-CM codes 28 (listed in Table S1); (b) ≥1 prescription of Gla-300, Gla-100 or IDet during the identification period; (c) ≥1 prescription of a BIA (different from the index BIA) in the 12-month baseline period; (d) ≥12 months of baseline EMR activity (defined as any encounter with the healthcare system); (e) ≥12 months of follow-up EMR activity; and (f) valid glycated haemoglobin A1c (HbA1c) measurements (3%-15% 1,29 ) during the 6-month baseline and 9-to 12-month follow-up. All patients also had to have ≥1 of the following seven subgroup characteristics expected to confer a high risk of hypoglycaemia. Of note, patients could have been included in more than one subgroup.
• Subgroup 1: Uncontrolled glycaemia, defined as HbA1c ≥8%.  Patients were excluded if they had type 1 diabetes according to ICD-9-CM/ICD-10-CM codes 28 (listed in Table S1) or prescriptions of more than one BIA on the index date.
Baseline patient data extracted from the EMRs included: age, sex, race, US geographic region, insurance type, comorbidities during 12-month baseline (identified according to ICD-9-CM and ICD-10-CM diagnosis codes detailed in Table S2), HbA1c (last value between 3% and 15% during baseline), hypoglycaemia (defined in Table S1) incidence (% of patients with ≥1 episode) and event rate

| Propensity score-matching
To minimize confounding, Gla-300 switchers were propensity score-matched 30 (1:1) to Gla-100/IDet switchers using a propensity score derived from baseline demographics and clinical characteristics. A greedy nearest-neighbour algorithm was used for propensity score-matching. This algorithm selects a subject switched to Gla-300 and then selects a matched control subject (the subject switched to Gla-100/IDet whose propensity score is closest to that of the selected subject). Once a match was made, patients were not reconsidered for further matching. Propensity scores were matched using two to eight decimal places (with a calliper width of 0.01), sequentially from highest to lowest digit match.

| Outcomes
Outcomes were compared between propensity score-matched Gla-300 switchers and Gla-100/IDet switchers. HbA1c outcomes were follow-up HbA1c, captured as the last value during 9-to 12-month follow-up; HbA1c change from baseline; and the proportions of patients reaching common glycaemic goals (HbA1c <7% and HbA1c <8%) during follow-up. Hypoglycaemia outcomes included the incidence and event rates of any severity of hypoglycaemia ('allhypoglycaemia'; identified by ICD-9-CM/ICD-10-CM codes and/or plasma glucose level ≤70 mg/dL [3.9 mmol/L]), and incidence and event rates of hypoglycaemia associated with an inpatient or ED contact to 12-month follow-up.
Analyses were conducted on the total matched populations of Gla-300 and Gla-100/IDet switchers. Additionally, outcomes were analysed for each of the seven individual hypoglycaemia-risk subgroups (see section 2.2). Matching quality for the individual subgroups was assessed, and where characteristics differed, adjusted analyses accounting for any baseline characteristics were undertaken.

| Statistical analysis
Categorical variables are presented as frequencies and percentages, and continuous variables as means and standard deviations (SDs).
Patients with missing data were classified as 'unknown' for any missing variables. HbA1c reduction was analysed using Student t-tests, and HbA1c goal attainment was compared using Fisher's exact tests.
Hypoglycaemia incidence was calculated using logistic regression and adjusted odds ratios (aORs), along with 95% confidence intervals (CIs) and P-values, were calculated. For event rates of hypoglycaemia, least-square mean (LSM) differences, 95% CIs, and P-values were calculated using a generalized linear model.

| Sensitivity analyses
Sensitivity analyses were performed for HbA1c outcomes by considering latest HbA1c laboratory observations during 3-6 and 6-9 months post-index date as the follow-up HbA1c. For the hypoglycaemia outcomes, a sensitivity analysis was performed using only hypoglycaemia events identified by ICD-9-CM/ICD-10-CM codes.
All analysed subjects had at least two HbA1c measurements, with the measurements closest in time to the index date and follow-up milestone selected.

| Study population
As shown in the study flow diagram (Figure 1 Before propensity score-matching, there were some statistically significant differences in baseline characteristics between Gla-300 switchers and Gla-100/IDet switchers, including age, race, insurance type, HbA1c, BMI, some diabetes therapies and other concomitant medications, some comorbidities (hyperlipidaemia, obesity, neuropathy), hypoglycaemia, and healthcare utilization (Table 1). After propensity score-matching, all baseline characteristics had a standard mean difference <0.1, indicative of a good balance in baseline characteristics between the cohorts (Table 1).
There were no significant differences (with an alpha level set conservatively at p < .15) in any of the baseline characteristics between the matched cohorts.

| HbA1c
During the 12-month follow-up period, HbA1c decreased significantly from baseline in both cohorts ( Figure 2A
Patients in the Gla-300 switcher cohort had a significantly lower risk of experiencing hypoglycaemia associated with an inpatient/ED contact than patients in the Gla-100/IDet cohort (8.2% vs.

| Sensitivity analyses
The HbA1c sensitivity analysis results (ie, during 3-to 6-and 6-to 9-month follow-up) were very similar to the main results (ie, during 9-to 12-month follow-up) (Supporting Information Text S1).
When restricting hypoglycaemic events to those identified only by ICD-9-CM/ICD-10-CM codes, there were approximately half as many hypoglycaemia episodes, and the significant differences in inpatient/ED-associated hypoglycaemia incidence and event rate in favour of Gla-300 were lost (Supporting Information Text S2).
Further, patients in the Gla-300 switcher cohort experienced significantly more all-hypoglycaemic events than those in the Gla-100/ IDet switcher cohort (p = .022).

| DISCUSS ION
In this large, retrospective real-world study of EMR data of patients with T2D deemed at high risk of hypoglycaemia and treated with a BIA, switching to a different BIA -either second-generation (Gla-300) or first-generation (Gla-100 or IDet) -was associated with comparable glycaemic improvement and all-hypoglycaemia incidence and event rates during 12 months of follow-up, but significantly lower hypoglycaemia incidence and event rates associated with inpatient/ED contact.
In the current study, Gla-300 switchers had significantly lower event rates associated with inpatient/ED visits than Gla-100/IDet switchers in all seven risk subgroups. A reduction in event rates for hypoglycaemia associated with an inpatient stay or ED visit with Gla-300 has also been found in the LIGHTNING study, 15  -uncontrolled HbA1c, ASCVD, and sulphonylurea users. The sulphonylurea user group generally had the lowest incidence and event rates (for all-hypoglycaemia and inpatient/ED-related hypoglycaemia) for either treatment group relative to the other risk subgroups, whereas the ASCVD subgroup generally ranked as the highest risk group in terms of incidence and event rates. Hypoglycaemia is thought to be associated with increased cardiovascular risk, potentially due to oxidative stress, inflammation, the development of atherosclerosis, etc. 31,32 It is also known that cardiovascular (CV) events increase the risk of severe hypoglycaemia events (SHEs) as well as the risk of CV events after SHEs, validating the bidirectional relationship between CV events and SHEs in patients with high comorbidity scores. 33 Therefore, it may be that patients with ASCVD in the current study have a complex cycle of hypoglycaemic episodes with worsening ASCVD or increased frailty.
When examining the effect of Gla-300 on HbA1c levels among patients at high risk of hypoglycaemia, mean HbA1c reductions from baseline among Gla-300 switchers varied somewhat across the seven risk groups, from -0.33% (sulphonylurea use) to -0.88% (uncontrolled HbA1c), but there were no significant differences between Gla-300 switchers and Gla-100/IDet switchers in each subgroup. Similar results were found in high-hypoglycaemia-risk subgroup analyses in DELIVER D+, 19 in which HbA1c reductions in the Gla-300 arm varied from -0.54% (age ≥65 years) to -0.98% (uncontrolled HbA1c) (ASCVD and sulphonylurea use were not studied), with no significant differences between the Gla-300 and insulin degludec arms. The mean overall HbA1c reduction in the Gla-300 arm in the current study was -0.51% over 12 months, which was very similar to that found in general T2D adults in DELIVER D+ 19 and DELIVER 2 17 (-0.63% and -0.51%, respectively) over the shorter period of 6 months. Further, attainment of HbA1c <7% and <8% was similar in the current study and in DELIVER D+ 19 and DELIVER 2, 17 albeit over different follow-up periods, indicating that Gla-300 can be used effectively in high-hypoglycaemia-risk patients.
In the current study, attainment of HbA1c <7% and <8% increased considerably from baseline to follow-up (approximately 10% to 18% and 30% to 43%, respectively, in both arms) implying that despite having a higher hypoglycaemia risk, patients can still make important progress towards achieving glycaemic targets.
In the present study, 12-month follow-up data show that, despite higher rates of hypoglycaemia in this high-risk population, patients on Gla-300 generally had numerically lower incidences and event rates for all-hypoglycaemia, and mainly significantly lower incidences and event rates for hypoglycaemia associated with inpatient/ED contacts, while at the same time reducing HbA1c compared with baseline, allowing more patients to achieve HbA1c goals.

| Strengths and limitations
To our knowledge, this study is the first to provide real-world insights into the characteristics of high-hypoglycaemia-risk patients and their clinical response to Gla-300 vs Gla-100/IDet over a 12-month follow-up period. Such patients are often excluded from randomized controlled trials, so these results provide valuable information for clinicians, healthcare-delivery systems, patients and payers.
However, the results of DELIVER High Risk should be interpreted with some caution because of its retrospective design and relatively short follow-up (12 months). Further, diagnoses were based on ICD-9-CM/ICD-10-CM codes, but as EMR data may not link the actual diagnosis name, this could have resulted in some misclassification.
Hypoglycaemias may have been underreported in the study (ie, there were no self-monitoring-blood-glucose or continuous-blood-

| Conclusions
In conclusion, in patients with T2D at high risk of hypoglycaemia and treated with a BIA, switching to Gla-300 versus a first-generation BIA (Gla-100 or IDet) resulted in similar HbA1c reductions and glycaemic goal attainment, but with a significantly lower risk of hypoglycaemia associated with inpatient/ED contacts (overall and in most risk subgroups) during the 12 months after switching.

H U M A N S TU D I E S A N D S U BJ EC TS
The IBM Explorys data services provide aggregated, de-identified data in accordance with the requirements of HIPAA. Given the observational and retrospective nature of this study, individual consent was not required after ensuring for anonymization of data.

DATA AVA I L A B I L I T Y S TAT E M E N T
Aggregate data are provided in the manuscript. Scripts used for pulling data from IBM Explorys platform are available on request from Rishab.a.gupta@accenture.com. Restrictions apply to the availability of the source patient-level data, which were used under licence for this study.