Real-world persistence, adherence, health care resource utilization, and costs in people with type 2 diabetes switching from a first-generation basal insulin to a second-generation (insulin glargine 300 U/mL) vs an alternative first-generation basal insulin

Abstract

BACKGROUND: People with type 2 diabetes (T2D) who change their basal insulin (BI) may have variable persistence with therapy. Compared with first-generation (long-acting) BI analogs (insulin glargine 100U/mL [Gla-100]; insulin detemir [IDet]), second-generation (longer-acting) BI analogs (insulin glargine 300U/mL [Gla-300]; insulin degludec) have similar glycated hemoglobin (HbA1c) attainment and lowered hypoglycemia risk, which could impact treatment persistence.

OBJECTIVE: To compare persistence, adherence, health care resource utilization (HRU), and costs for individuals switching from neutral protamine Hagedorn insulin or a first-generation BI analog with either the second-generation BI, Gla-300, or an alternative first-generation BI analog (Gla-100 or IDet).

METHODS: We used Optum Clinformatics claims data from adults (aged ≥ 18 years) with T2D who had received BI (neutral protamine Hagedorn, Gla-100, IDet) in the 6-month baseline period, and switched to either Gla-300 or an alternative first-generation BI (Gla-100 or IDet; treatment switch = index date) between April 1, 2015, and August 31, 2019. Participants were followed for 12 months, until plan disenrollment, or until death, whichever occurred first. Cohorts were propensity score matched (PSM) on baseline characteristics. The primary outcome was the proportion who were persistent with therapy at 12 months. Secondary outcomes were adherence (proportion of days covered); change in HbA1c; and all-cause, diabetes-related, and hypoglycemia-related HRU and costs.

RESULTS: PSM generated 3,077 participants/group (mean age: 68 years, 52% female). Cohorts were well balanced except for hospitalization, which was adjusted in models as a covariate. During the 12-month follow-up period, participants who received Gla-300 vs first-generation BI had greater persistence with (45.5% vs 42.1%; adjusted P = 0.0001), and adherence to (42.8% vs 38.2%; adjusted P = 0.0006), BI therapy and a statistically larger reduction in HbA1c at 12 months (−0.65% vs −0.45%; adjusted P = 0.0040). The proportion of participants achieving HbA1c less than 8% (47.2% vs 40.9%; P < 0.0001), but not less than 7% (21.2% vs 20.8%), was significantly higher for Gla-300 vs first-generation BI. All-cause (45.3 vs 65.9 per 100 patient-years [P100PY]) and diabetes-related (21.5 vs 29.1 P100PY), but not hypoglycemia-related, hospitalizations (1.0 vs 1.5 P100PY) were significantly (P < 0.0001) lower for Gla-300 vs first-generation BI. Similarly, all-cause (111.9 vs 148.8 P100PY), diabetes-related (54.8 vs 74.2 P100PY), and hypoglycemia-related (2.9 vs 5.7 P100PY) emergency department (ED) visits were significantly lower for Gla-300 (all P < 0.0001). Costs for all-cause hospitalizations and hypoglycemia-related ED visits were significantly lower for Gla-300 vs first-generation BI. Although pharmacy costs were significantly higher for Gla-300 vs first-generation BI, all-cause total health care costs were not significantly different: $41,255 vs $45,316 per person per year, respectively.

CONCLUSIONS: In this claims-based analysis of people with T2D receiving BI, switching to Gla-300 was associated with significantly better persistence, adherence, and HbA1c reduction compared with switching to an alternative first-generation BI analog. All-cause HRU was significantly lower; despite significantly higher pharmacy costs, total health care costs were similar.

DISCLOSURES: This study was funded by Sanofi US. Medical writing support was provided by Helen Jones, PhD, CMPP, of Evidence Scientific Solutions and funded by Sanofi US. Dr Wright is on the speakers’ bureau and sits on the advisory boards for Abbot Diabetes, Bayer, Boehringer Ingelheim, Eli Lilly, and Sanofi; sits on the advisory board for Medtronic; and is a consultant for Abbot Diabetes, Bayer, Boehringer Ingelheim, and Eli Lilly. Dr Malone is on advisory boards for Novartis and Avalere and consults for Pear Therapeutics, Sarepta, and Strategic Therapeutics. Dr Trujillo sits on advisory boards for Novo Nordisk and Sanofi. Drs Gill, Zhou, and Preblick and Mr Li are employees and stockholders of Sanofi. Mr Huse is an employee of Evidera and a contractor for Sanofi. Dr Reid is a speaker and consultant for Novo Nordisk and Sanofi-Aventis and is a consultant for AstraZeneca and Intarcia.

Plain language summary

Basal insulin products are regarded as "long-acting” (≤ 24 hours) or "longer-acting” (> 24 hours). People with type 2 diabetes may need to change their basal insulin. This study compared change to longer-acting (glargine 300 U/mL) against long-acting insulin (glargine 100 U/mL or detemir). People taking longer-acting insulin were more likely to continue treatment and had a larger reduction in blood sugar and fewer emergency department visits and hospitalizations than those taking long-acting insulins.

Implications for managed care pharmacy

This real-world study examined data from individuals with type 2 diabetes switching from a first-generation basal insulin to either insulin glargine 300 U/mL (Gla-300) or an alternative first-generation basal insulin. Switching to Gla-300 was associated with significantly greater persistence, adherence, and glycated hemoglobin reduction. Although pharmacy costs were significantly higher with Gla-300, all-cause and diabetes-related health care resource utilization were significantly lower. Total health care costs were similar for Gla-300 and first-generation basal insulin.

According to the Centers for Disease Control and Prevention, approximately 34 million Americans—more than 10% of the population—are living with diabetes.¹ Furthermore, another 88 million have prediabetes. Health care costs for Americans with diabetes are 2.3 times greater than for those without diabetes and represent an annual burden of $327 billion, of which $237 billon are direct medical costs.² When considering health care costs for diabetes in the context of overall health care spending, $1 of every $7 spent on health care is used in treating diabetes and its complications.²

As β-cell function becomes more compromised and endogenous insulin secretion wanes, exogenous insulin is often needed. Approximately 7.4 million people in the United States are receiving 1 or more formulations of insulin.³ Insulin therapy is often started with a basal insulin (BI) to provide coverage throughout the day; however, many people find it necessary to change their BI option for a variety of clinical or nonclinical reasons including formulary changes.⁴ Persistence with, and adherence to, BI are often suboptimal,⁵ with the most common barriers being concerns over self-injection, lack of perceived need, and fear of hypoglycemia.⁶ A real-world study in people with type 2 diabetes (T2D) newly initiating BI therapy showed that failure to achieve an HbA1c of less than or equal to 7% during the initial 3 months of treatment was associated with increased risk of failure to achieve glycemic targets at 2 years. Similarly, occurrence of hypoglycemia during the initial 3-month period was associated with longer-term risk of hypoglycemia over 2 years.⁷ Another retrospective study of insulin-naive individuals with T2D showed that those who experienced hypoglycemia within 6 months of initiating BI were more likely to discontinue therapy within the first 12 months.⁸ A primary care database analysis study reported that for new users of BI on top of oral therapy, 2-year persistence rates for people who received insulin glargine, insulin detemir (IDet), and neutral protamine Hagedorn (NPH) were 65%, 53%, and 59%, respectively.⁹

Compared with first-generation (long-acting) BI analogs (insulin glargine 100U/mL [Gla-100] and IDet), the newer second-generation (longer-acting) BI analogs (insulin glargine 300U/mL [Gla-300] and insulin degludec [IDeg]) have a more stable pharmacokinetic profile, longer duration of action (> 24 vs ≤ 24 hours), and more stable action profiles with less within-day and between-day glucose variability.^10,11 Clinical studies have shown that because of these attributes, second-generation BIs have a reduced risk of hypoglycemia compared with first-generation BIs.^12,13 With the average costs of treating hypoglycemia ranging from $212 per prescription for a home glucagon kit to $1,487 per emergency department (ED) visit and $18,961 per hospital admission,¹⁴ reducing hypoglycemia risk may have a large cost-benefit, on top of the obvious health benefits. The results of the DELIVER-2 retrospective cohort study showed a $1,439 per patient per year saving associated with reductions in hypoglycemia-related hospitalizations, ED visits, and outpatient visits for those who switched to Gla-300 vs another BI (Gla-100, IDet or IDeg).¹⁵

The primary objective of this retrospective, real-world observational study was to compare treatment persistence during the 12-month follow-up period between participants with T2D who switched from a previous BI regimen (NPH, Gla-100, or IDet) to the second-generation BI analog, Gla-300, vs an alternative first-generation BI analog (Gla-100 or IDet). Additionally, medication adherence, glycated hemoglobin (HbA1c) change, glycemic goal attainment, and all-cause, diabetes-related, and hypoglycemia-related health care resource utilization (HRU) and health care costs were compared.

Methods

Data from the Optum Clinformatics claims database were included from adults (aged ≥ 18 years) with T2D who had received prior BI (NPH, Gla-100, or IDet) in the 6-month baseline period and switched to the second-generation (longer-acting) BI analog, Gla-300, or to an alternative first-generation (long-acting) BI (treatment switch = index date) between April 1, 2015, and August 31, 2019. Participants had to have continuous medical and prescription drug coverage for the 6 months before index date and at least one valid HbA1c value (between 5.0% and 15.0%) during the baseline period. Data from individuals who had a diagnosis of type 1 diabetes identified using an adaptation of the Klompas algorithm,¹⁶ were pregnant, or had bolus claims at any time during the baseline period were excluded. Also excluded were those with any prescription claim for a BI different from the study index BI (Gla-300 or a first-generation BI) or a fixed-ratio combination (iGlarLixi or IDegLira) on index date or within 7 days after index date, or who initiated bolus insulin concomitantly to index BI treatment initiation with Gla-300 or first-generation BI.

Participants had to have switched from their baseline BI (NPH, Gla-100, or IDet) to either an alternative first-generation BI analog (Gla-100 or IDet) or Gla-300. The first-generation BI group comprised those who switched to Gla-100 from NPH or IDet, or to IDet from NPH or Gla-100. Participants were followed from index date for up to 12 months until plan disenrollment, death, end of follow-up (360 days), or last day of available data, whichever occurred first. Data from participants who completed the 12-month follow-up period were censored at 12 months, whereas data from those who did not complete the 12-month follow-up period were censored at the last known follow-up date. Cohorts were propensity score matched (PSM; greedy nearest neighbor matching algorithm using propensity score with logit scale) on baseline demographic and clinical characteristics in a 1:1 ratio. Logistic regression was used to generate propensity scores for each participant. The final set of PSM variables (as listed in Table 1) and their structure (ie, categorical vs continuous) used in the propensity score were selected based on clinical input and model convergence and performance.

TABLE 1.

Baseline Characteristics

	Propensity score–matched population
	Gla-300 (n = 3,077)	First-generation BI (n = 3,077)	Standardized difference
Age, years
Age group, n (%)
18-49	169 (5.49)	192 (6.24)	0.0320
50-64	767 (24.93)	674 (21.90)	0.0710
65-74	1,352 (43.94)	1,356 (44.07)	0.0026
≥ 75	789 (25.64)	855 (27.79)	0.0480
Mean (SD)	67.96 (10.34)	68.30 (10.79)	0.0320
Median (Q1-Q3)	69 (62-75)	69 (63-75)
Minimum-maximum	20-89	19-89
Sex, n (%)
Male	1,486 (48.29)	1,480 (48.10)	0.0039
Female	1,591 (51.71)	1,597 (51.90)	0.0039
Race/ethnicity, n (%)
Asian	106 (3.44)	108 (3.51)	0.0035
African American	383 (12.45)	443 (14.40)	0.0570
Hispanic	786 (25.54)	710 (23.07)	0.0580
White	1,375 (44.69)	1,382 (44.91)	0.0046
Other/unknown	427 (13.88)	434 (14.10)	0.0066
Health plan type, n (%)
Commercial	417 (13.55)	427 (13.88)	0.0094
Medicare	2,660 (86.45)	2,650 (86.12)	0.0094
Region, n (%)
Midwest	195 (6.34)	232 (7.54)	0.0470
Northeast	280 (9.10)	282 (9.16)	0.0023
South	1,888 (61.36)	1,805 (58.66)	0.0550
West	712 (23.14)	755 (24.54)	0.0330
Charlson Comorbidity Index
Group, n (%)
0	1,375 (44.69)	1,377 (44.75)	0.0013
1	760 (24.70)	706 (22.94)	0.0410
≥ 2	942 (30.61)	994 (32.30)	0.0360
Mean (SD)	1.21 (1.56)	1.31 (1.75)	0.0640
Median (Q1-Q3)	1 (0-2)	1 (0-2)
Minimum-maximum	0-12	0-13
Common comorbidities and diabetic complications of interest, n (%)
Hypertension	2,609 (84.79)	2,659 (86.42)	0.0460
Hyperlipidemia	2,439 (79.27)	2,451 (79.66)	0.0097
Neuropathy	1,293 (42.02)	1,254 (40.75)	0.0260
Nephropathy	1,075 (34.94)	1,094 (35.55)	0.0130
Retinopathy	682 (22.16)	643 (20.90)	0.0310
Obesity	1,073 (34.87)	1,066 (34.64)	0.0048
Depression	580 (18.85)	617 (20.05)	0.0300
Chronic kidney disease	1,128 (36.66)	1,195 (38.84)	0.0450
HbA1c (%) closest to index treatment initiation, n (%)
< 7%	359 (11.67)	431 (14.01)	0.0700
≥ 7% and < 8%	636 (20.67)	676 (21.97)	0.0320
≥ 8% and < 9%	793 (25.77)	663 (21.55)	0.0100
≥ 9% and < 11%	873 (28.37)	813 (26.42)	0.0440
≥ 11%	416 (13.52)	494 (16.05)	0.0710
Severe hypoglycemia during baseline period
Event rate P100PY	3.835	5.135
Incidence, n (%)	59 (1.92)	79 (2.57)	0.0440
BI before switch to index treatment (Gla-300 or first-generation BI), n (%)
IDet	1,268 (41.21)	1,364 (44.33)	0.0630
Gla-100	1,654 (53.75)	1,505 (48.91)	0.0970
NPH	155 (5.04)	208 (6.76)	0.0730
Antihyperglycemic medication, n (%)
Metformin	1,549 (50.34)	1,513 (49.17)	0.0230
Sulfonylureas	907 (29.48)	995 (32.34)	0.0620
DPP-4 inhibitors	533 (17.32)	510 (16.57)	0.0200
SGLT-2 inhibitors	4 (0.13)	1 (0.03)	0.0340
Thiazolidinediones	177 (5.75)	182 (5.91)	0.0069
α-glucosidase inhibitors	19 (0.62)	21 (0.68)	0.0081
Meglitinides	47 (1.53)	49 (1.59)	0.0052
GLP-1 RA	524 (17.03)	419 (13.62)	0.0950
OADs, n (%)
0	979 (31.82)	973 (31.62)	0.0042
1	1,153 (37.47)	1,140 (37.05)	0.0087
2	763 (24.80)	779 (25.32)	0.0120
≥ 3	182 (5.91)	185 (6.01)	0.0041
Health care utilization, n (%)
≥ 1 office visit	3,009 (97.79)	2,998 (97.43)	0.0230
≥ 1 office visit with diabetologist or endocrinologist	641 (20.83)	524 (17.03)	0.0970
≥ 1 hospitalization	486 (15.79)	638 (20.73)	0.1300a
≥ 1 ED visit	881 (28.63)	1,012 (32.89)	0.0920

Bold P values indicate those with a standardized difference approaching 0.1.

^aP > 0.1.

BI = basal insulin; DPP-4 = dipeptidyl peptidase 4; ED = emergency department; Gla-100 = insulin glargine 100 U/mL; Gla-300 = insulin glargine 100 U/mL; GLP-1 RA = glucagon-like peptide 1 receptor agonist; HbA1C = glycated hemoglobin A1C; IDet = insulin detemir; NPH = neutral protamine Hagedorn; OAD = oral antidiabetic agent; P100PY = per 100 patient-years; SGLT-2 = sodium–glucose cotransporter 2.

STUDY OUTCOMES

The primary outcome was treatment persistence, defined as no discontinuation of the index BI until the end of the follow-up period. As this study measured persistence of a titratable injectable therapy with weight-based dosing, rather than using the traditional measure of a fixed gap of “days supply,” which does not accurately reflect BI use, an alternative methodology was used as described by Wei and colleagues,¹⁷ accounting for individual variation in treatment periods because of differences in titration and dose. Index BI was considered discontinued if the prescription was not refilled within the expected time of medication coverage, defined by the 90th percentile of the time for which medication was used. Thus, we applied a data-driven approach in which we calculated the metric quantity of BI supplied between patients’ first and second fills; then we grouped patients according to different metric quantities and then calculated the 90th percentile of the duration between first and second fill within each metric quantity group. Using this stratification process, we were able to measure persistence through assigning the estimated allowable time of medication coverage according to the metric quantity of BI received.¹⁷ Sensitivity analyses were conducted using the 75th and 95th percentiles of time period of medication use as expected medication coverage. Participants who did not have continuous coverage of the index BI during the follow-up period (whether that was through 12 months, plan disenrollment, or death) were considered nonpersistent.

Secondary outcome measures were treatment adherence, HbA1c goal attainment (< 7% and < 8%), HRU, and health care costs during the 12-month follow-up period, and change from baseline in HbA1c. Treatment adherence was defined as the proportion of days covered (PDC; total days supplied on the claim divided by the number of days in refill interval, assuming all medications are consumed as prescribed) using a cut-off of greater than or equal to 80% to define adherence to treatment.¹⁸ Medication adherence was modeled using logistic regression. Change in HbA1c was calculated between baseline and 12 months, which was specified per-protocol to include only those patients who had at least one valid HbA1c measure between the window of 270 to 390 days post-index. Data at 6 months were also captured. For inclusion in the HbA1c analysis, values at both baseline and 12 months were required, although follow-up HbA1c measurements during the sample selection process were not required. The likelihood of achieving an HbA1c target of less than 8% or less than 7% during the follow-up period was also assessed, along with the time to achieve goal attainment. HRU was assessed during the 12-month follow-up period and included hospital admissions, inpatient length of stay, and ED visits. Crude incidence rates for hospitalizations or ED visit, mean number of events, and event rates (number of events per 100 person-years [P100PY]) were reported, as was inpatient length of stay. Diabetes-related and hypoglycemia-related hospital admissions were defined as those having an International Classification of Diseases Ninth Revision, Clinical Modification (ICD-9-CM) or International Classification of Diseases Tenth Revision, Clinical Modification (ICD-10-CM) discharge diagnosis code of diabetes or hypoglycemia, respectively. Diabetes-related ED admissions were defined as those having an ICD-9-CM or ICD-10-CM primary or secondary diagnosis of diabetes or hypoglycemia, respectively. All-cause, diabetes-related, and hypoglycemia-related hospitalization, ED, and total health care costs and pharmacy costs during the follow-up period were calculated and reported as costs per patient per year (PPPY). Diabetes-related total health care costs were defined as costs from medical claims with a primary or secondary diagnosis of diabetes, antihyperglycemic medication, insulin pumps, glucose meters, and test strips. Diabetes-related total pharmacy costs were defined as those relating to cost from antihyperglycemic medication (eg, insulin, glucagon-like peptide-1 receptor agonists, oral agents, and glucagon).

STATISTICAL ANALYSES

Persistence and mean persistent days (SD) were compared using a chi-square test for persistence rate, and Student’s t-test for persistence days. A P value of less than or equal to 0.05 was considered statistically significant. A Cox proportional hazards regression model with baseline imbalances as covariates was used to compare the risk of treatment discontinuation between the treatment groups, whereas Kaplan-Meier analyses were performed to compare the time to discontinuation. The percentage of participants persistent throughout the 12-month follow-up period and median duration of persistence in days were calculated. Statistical differences in the PDC of greater than or equal to 80% rates were assessed using logistic regression with baseline imbalances as covariates. All-cause, diabetes-related, and hypoglycemia-related events were estimated with a Poisson generalized linear regression model, using a generalized estimating linear equation model with baseline imbalances as covariates,^19,20 and a two-part zero-inflated model²¹ was used to assess costs during the follow-up period. Statistical difference in HbA1c change was assessed using a generalized estimating equation model.

Results

BASELINE CHARACTERISTICS

From an initial sample of 4,944,260 individuals with a diagnosis of T2D between April 1, 2015, and August 31, 2019, 4,548 participants who used BI and switched to Gla-300 and 7,967 who switched to an alternative first-generation BI were identified as eligible for PSM (Supplementary Table 1, available in online article). After PSM, there were 3,077 participants in each group. The mean age was 68 years, 52% were female, and the majority (86%) had Medicare insurance coverage (Table 1). Baseline characteristics were well balanced (standardized difference < 0.1) except for the proportions of participants experiencing hospitalization (16% vs 21%), which were higher in the first-generation BI group. The study protocol was prespecified to include any imbalanced baseline variable after PSM as covariates to assess whether the inclusion of covariates would affect the model results. Thus, baseline hospitalization was adjusted for in models as a covariate. Approximately 68% of participants in both treatment groups had received 1 or more oral antidiabetic agents (OADs) during the baseline period, and 31% had received 2 or more OADs. At baseline, 51.3% of participants were receiving Gla-100, 42.8% were receiving IDet, and 5.9% were receiving NPH. In the Gla-300 and first-generation groups, 82.6% and 76.0% of participants had complete, 12-month follow-ups. Among the matched groups, the median follow-up time was 360 days for both groups, and the average follow-up time was 326 days and 313 days for the Gla-300 and first-generation BI groups, respectively.

PRIMARY ENDPOINT: TREATMENT PERSISTENCE

Using the 90th percentile of time of medication coverage, during the 12-month follow-up period, a significantly higher proportion of participants who received Gla-300 vs first-generation BI (45.5% [n = 1,399] vs 42.1% [n = 1,296]; adjusted P = 0.0001) were persistent with therapy (Figure 1). The mean (SD) number of persistent days was 234.6 (130.1) for Gla-300 and 218.7 (131.9) for first-generation BI (unadjusted P < 0.0001). In the sensitivity analysis, there was no significant difference at the stricter 75th percentile analysis (which permits a shorter gap to achieve persistence; n = 1,910; 30.3% vs 31.8%; P = 0.23), but at the 95th percentile (n = 2,997), persistence was significantly higher for Gla-300 vs first-generation BI (51.2% vs 46.3%; P = 0.0001). Thus, participants who switched to Gla-300 vs a first-generation BI analog were less likely to discontinue their medication during the follow-up period with a hazard ratio (HR; [95% CI]) of 0.87 [0.82-0.93]; P = 0.0001). The median time to discontinuation in those who discontinued therapy was 31 days longer for those taking Gla-300 (336 [95% CI = 325-344] days) vs a first-generation BI analog (305 [95% CI = 287-317] days). At 6 months, 30.8% of those who switched to Gla-300 discontinued medication compared with 34.6% for first-generation BI. At 12 months, these proportions increased to 54.5% and 57.9%, respectively.

FIGURE 1.

Treatment Persistence (90th Percentile Time of Medication Coverage) and Adherence up to 12 Months of Follow-Up

SECONDARY ENDPOINTS

A significantly higher proportion of participants who received Gla-300 vs first-generation BI were adherent to therapy (≥ 80% PDC: 42.8% vs 38.2%; P = 0.0006; Figure 1). The mean (SD) number of adherent days was 214.5 (111.1) for Gla-300 and 194.5 (112.6) for first-generation BI. Multivariate logistic regression analysis revealed that those who switched to Gla-300 had approximately 20% higher odds of being adherent to treatment (odds ratio [95% CI] = 1.197 [1.081-1.326]; adjusted P = 0.0006).

A total of 2,817 participants (Gla-300 n = 1,535; first-generation BI n = 1,282) had a valid baseline and follow-up HbA1c value at 12 months. Change in HbA1c from baseline to 12 months was significantly greater for those who received Gla-300 compared with first-generation BIs (−0.65% vs −0.45%; adjusted P = 0.0040; Figure 2), as was the proportion of participants who achieved HbA1c less than 8% at 12 months (47.2% [n = 1,453] vs 40.9% [n = 1,258]; log-rank P = 0.0002). Cox model analysis showed that participants who switched to Gla-300 were approximately 20% more likely to achieve this target (log HR [95% CI] = 0.1849 [0.109-0.2608]; P < 0.0001), There was no significant between-group difference in the proportions of participants achieving HbA1c less than 7% at 12 months (Gla-300, 21.2% [n = 653]; first-generation BI, 20.8% [n = 640]; log HR [95% CI] = 0.0288 [−0.0808 to 0.1384]; P = 0.61), nor was there a significant difference between groups for change in HbA1c from baseline to 6 months (Gla-300 [n = 1,806]: −0.53%; first-generation BIs [n = 1,530]: −0.43; adjusted P = 0.07). All-cause (45.3 vs 65.9 P100PY) and diabetes-related (21.5 vs 29.1 P100PY) hospitalizations (Figure 3) were significantly lower (both P < 0.0001) for Gla-300 vs first-generation BI, as were all-cause (111.9 vs 148.8 P100PY), diabetes-related (54.8 vs 74.2 P100PY), and hypoglycemia-related (2.9 vs 5.7 P100PY) ED visits (all P < 0.0001; Figure 3). Rate ratios (rate ratio [95% CI]) generated by generalized linear regression suggested that Gla-300 was associated with approximately 27% fewer all-cause hospitalizations (0.73 [0.68-0.78]; P < 0.0001), 23% fewer diabetes-related hospitalizations (0.77 [0.69-0.86]; P < 0.0001), 21% fewer all-cause ED visits (0.79 [0.75-0.82]; P < 0.0001), 23% fewer diabetes-related ED visits (0.77 [0.72-0.83]; P < 0.0001), and 46% fewer hypoglycemia-related ED visits (0.54 [0.41-0.70]; P < 0.0001). There was no significant difference in hypoglycemia-related hospitalizations (1.0 vs 1.5 P100PY; rate ratio: 0.71 [0.43-1.15]; P = 0.16).

FIGURE 2.

HbA1c Reduction From Baseline to 12 Months

FIGURE 3.

Health Care Resource Utilization

Costs for all-cause hospitalizations ($12,996 vs $16,375 PPPY; P = 0.033; Figure 4) and hypoglycemia-related ED visits ($46 vs $92 PPPY; P = 0.0068; Figure 4) were significantly lower for Gla-300 vs first-generation BI. There was no significant difference between Gla-300 and first-generation BI in diabetes-related ($5,626 vs $6,210; P = 0.31) or hypoglycemia-related ($199 vs $333; P = 0.43) hospitalization costs (Figure 4). Nor was there a significant difference in all-cause ($1,767 vs $2,437; P = 0.34) or diabetes-related ($1,087 vs $1,543; P = 0.10) ED costs (Figure 4). Although pharmacy costs were significantly higher for Gla-300 vs first-generation BIs (all medications: $13,688 vs $10,642; diabetes medications: $7,093 vs $5,178; both P < 0.0001; Figure 4), all-cause ($41,255 vs $45,316; P = 0.64), diabetes-related ($22,613 vs $25,165; P = 0.92), and hypoglycemia-related ($713 vs $1,326; P = 0.45) total health care costs were not significantly different (Figure 4).

FIGURE 4.

Health Care Costs

Discussion

The results of this retrospective, real-world observational study in people with T2D who switched from previous BI to either the second-generation BI analog, Gla-300, or an alternative first-generation BI analog (Gla-100 or IDet) showed that, in the 12-month follow-up period, switching to Gla-300 was associated with significantly better therapy persistence. Those who switched to Gla-300 were less likely to discontinue their medication, with a median time to discontinuation that was approximately 1 month longer than that for participants in the first-generation BI group. Furthermore, those who switched to Gla-300 vs first-generation BI had significantly improved adherence to therapy, being approximately 4% higher for Gla-300, which translated to 20% higher odds of adherence vs switching to a first-generation BI. Participants who switched to Gla-300 had a statistically greater reduction in HbA1c compared with those who switched to an alternative first-generation BI. Although standardized differences are similar, the absolute value of many variables (including the proportions of individuals with followup HbA1c measurements) suggest that glycemic control may be improved in those who received Gla-300 vs first-generation BI. Indeed, the beneficial effects of modest decreases in HbA1c have been reported in several publications.^22-24 Although significantly more participants who switched to Gla-300 vs first-generation BI achieved an HbA1c target of less than 8%, there was no significant difference in the proportions of participants achieving an HbA1c target of less than 7%. This may well be reflective of the fact that a glycemic control target of less than 7% is more difficult to achieve than one of less than 8%. Furthermore, guidelines for primary care suggest an HbA1c goal of between 7% and 8%²⁵ rather than less than 7% per the American Diabetes Association Standards of Care. Additionally, the older age of the demographic of this cohort, with a mean age of 68 years, suggests that many participants were likely to have a less stringent HbA1c target, thereby reducing the numbers of patients available for the analysis of HbA1c less than 7%.

Compared with those receiving first-generation BI analogs, those who switched to Gla-300 also had significantly lower all-cause and diabetes-related hospitalizations, as well as all-cause, diabetes-related, and hypoglycemia-related ED visits, with the rate of hypoglycemia-related ED visits being almost halved. The rates of hypoglycemia-related hospitalizations for Gla-300 and first-generation BIs were not significantly different, which could possibly be related to the treatment pathway, in that, although patients with severe hypoglycemia might be taken to the ED, most are treated and released without being hospitalized. However, related to this, the recently reported real-world iNPHORM study results suggest that over half of people experiencing severe hypoglycemia never visit an ED, suggesting that the ICD codes capturing these events are greatly underreported in the real-world setting.²⁶ It is notable that although glycemic control (HbA1c reduction and the proportion of participants with HbA1c < 8%) at 12 months was better for those who switched to Gla-300 vs first-generation BI, those in the Gla-300 group also experienced lower hypoglycemia rates (as suggested by the HRU data). This is consistent with the results of several randomized studies,^27-30 real-world studies,^13,31 and meta-analyses.^32,33

Compared with those switching to a first-generation BI, participants switching to Gla-300 had significantly lower all-cause hospitalization and hypoglycemia-related ED costs. Conversely, pharmacy costs were significantly higher for Gla-300 vs first-generation BI. Although this may reflect the increased cost of Gla-300 vs first-generation BI, it is also likely a direct result of more prescriptions being filled because both persistence and adherence were higher for Gla-300.

The results of this study are similar to those of a previous study that also used data from Optum Clinformatics, in which the authors reported that after 6 months of follow-up, 20.4% of people who switched to Gla-300 discontinued their medication compared with 36.4% of those who switched to other BIs.³⁴ In the current study, the 6-month difference between the 2 treatment groups was not as pronounced; however, the previous study used only 1 year of data for patient identification and had just 1,820 participants, whereas the current study used more recent data, had a final sample size of over 6,000 patients after PSM, and adjusted for confounding via PSM, thereby likely providing more robust results. Another retrospective study, using data from the Predictive Health Intelligence Environment database, studied PSM cohorts who switched from a BI to Gla-300 or another BI between March 2015 and May 2016. Participants who switched to Gla-300 were significantly less likely to experience hospitalizations, ED visits, and outpatient visits related to hypoglycemia, as well as all-cause and diabetes-related ED visits, but not hospitalizations or outpatient visits. Total-cost-savings (accounting for hospitalizations, ED visits, and outpatient visits) were $2,181, $1,838, and $1,439 PPPY for all-cause, diabetes-related, and hypoglycemia-related HRU, respectively.¹⁵

STRENGTHS

Strengths of this study were that using the Optum claims database permitted use of a large population with long-term follow-up and breadth of coverage, which provided considerable statistical power. However, the lack of randomization and subsequent biases, for example, preferential prescription of certain therapies to patients who are perceived as having more severe disease or to those who are in poor overall health, suggests caution is required when interpreting results of comparative observational studies. In this study, however, this should have been in part mitigated by PSM of cohorts.

LIMITATIONS

Limitations of this study include those common to administrative claims database studies, for example, the lack of information on daily doses of medication, with data limited to the quantity of medication and the number of prescriptions filled. Thus, it is not possible to say with certainty if a medication was actually used. Furthermore, study participants may have had other sources of insulin such as samples, which would not be captured in the database. It is also possible that medication provided through hospitalization and rehabilitation facilities could affect the findings. The assessment of persistence with injectable therapies such as insulin using claims data is challenging as the dosing schedules are not fixed, and as such, “30-day supply” rules, which are routinely used to measure persistence with oral therapies, cannot be applied. Although the imbalance in the proportions of participants who had experienced hospitalization at baseline had the potential to affect outcomes, this was adjusted for as a covariate in the models used in the analysis. The T2D cohorts were identified using ICD-9-CM and ICD-10-CM codes. Limitations of how hypoglycemia is defined per ICD codes, and how this information was inputted into the database, could have resulted in misclassification. It is also probable that severe hypoglycemia was underreported because of the ICD diagnosis codes associated with ED visits or hospitalizations. Indeed, results of the iNPHORM study²⁶ showed that the odds of hypoglycemia being treated in the hospital setting are reduced for those who are of older age and female, and those with lower incomes who live in suburban areas that are further away from hospitals, corroborating that hypoglycemia rates are often underreported. Limitations specific to the current analysis include a lack of baseline adherence data, and that the assessment did not account for titration or hospitalization, which causes a delay in fulfilling prescriptions. Individuals included in this analysis had to have a baseline HbA1c test, which may have selected individuals with greater adherence to diabetes management recommendations. Only the time to the first event of goal attainment was analyzed. Finally, the generalizability of the study may be limited to the populations represented in the Optum Clinformatics database; therefore, although these findings may not be generalized to the overall US T2D population, they can be considered generalizable to those with commercial insurance or Medicare Advantage plans.

Conclusions

In people with T2D previously treated with BI therapy, switching to the second-generation BI (longer-acting) analog, Gla-300, was associated with significantly better persistence, adherence, and HbA1c reduction vs switching to an alternative first-generation (long-acting) BI. All-cause and diabetes-related hospitalizations and all-cause, diabetes-related, and hypoglycemia-related ED visits were significantly lower for Gla-300. Costs for all-cause hospitalizations and hypoglycemia-related ED visits were significantly lower for Gla-300 vs first-generation BI. Despite significantly higher pharmacy costs for Gla-300, total health care costs were similar to those for first-generation BIs.