Comparison of utilization, cost, adherence, and hypoglycemia in patients with type 2 diabetes initiating rapid-acting insulin analog with prefilled pen versus vial/syringe

Abstract

Background:

Studies examining outcomes of different insulin delivery systems are limited. The objective of this study was to compare healthcare utilization, costs, adherence, and hypoglycemia rates in patients with type 2 diabetes mellitus (T2DM) initiating rapid-acting insulin analog (RAIA) using prefilled pen versus vial/syringe.

Methods:

A retrospective analysis was conducted using a US claims database (1/1/2007 to 12/31/2008). Inclusion criteria were: ≥18 years old, with T2DM, ≥12 months of continuous eligibility, and new to RAIA. Difference-in-difference analyses after propensity score matching were conducted to compare changes in outcomes from 6 months prior to and 6 months after initiating RAIA with a prefilled pen versus vial/syringe (Wilcoxon rank-sum test for costs and t-test for other outcomes). Categories of utilization and costs (2009 USD) included total and diabetes-related inpatient, outpatient, and emergency room. Adherence was measured by proportion of days covered (PDC). Hypoglycemia was identified using ICD-9-CM codes.

Results:

Baseline characteristics were similar between the prefilled pen (n = 239) and vial/syringe (n = 590) cohorts after matching. Adherence to RAIA was greater in the prefilled pen cohort than the vial/syringe cohort (PDC: 54.6 vs. 45.2%, p < 0.001). While the increase in diabetes-related pharmacy costs from before to after initiating RAIA was greater in the prefilled pen cohort than the vial/syringe cohort (+$900 vs. +$607, p < 0.001), the prefilled pen cohort was associated with greater reductions in the total diabetes-related costs (–$235 vs. +$61, p = 0.006) and the utilization of oral anti-hyperglycemic agents (–1.3 vs. –0.7, p = 0.016). There were no significant differences in other outcomes.

Limitations:

Claims databases do not provide optimal measures for adherence or T2DM severity, and only capture hypoglycemia events requiring clinical intervention.

Conclusion:

Initiating RAIA with a prefilled pen was associated with better adherence and greater reduction in total diabetes-related costs than a vial/syringe. There was no significant difference in total healthcare costs.

Key words::

• Adherence
• Insulin
• Pen device
• Type 2 diabetes
• Vial

Introduction

Diabetes mellitus is the most common metabolic disorder and the global prevalence has been increasing at an alarming rate1. In the United States (US), the estimates for diabetes incidence and prevalence in 2007 were 1.6 million and 23.6 million, respectively, with projected prevalence being about 30 million by 20302,3. Type 2 diabetes mellitus (T2DM) comprises 90–95% of all diagnosed diabetes cases in adults2. The economic impact of diabetes on the US healthcare system is substantial. In 2002, the total costs attributable to diabetes were estimated to be about $132 billion4 and the estimated costs increased to $174 billion in 2007, including $116 billion and $58 billion attributed to direct and indirect costs, respectively5. Hospitalization was the most expensive cost component of direct costs (50% of total costs)5.

Insulin is an effective and long-standing anti-hyperglycemic therapy (AHT). Over the years, several studies encouraged an earlier use of insulin therapy, including the latest consensus statement by the American Diabetes Association and the European Association for the Study of Diabetes6–8. However, approximately half of the patients diagnosed with T2DM have sub-optimal glycemic control9,10. Also, non-adherence to AHTs is widespread. According to Cramer et al., adherence to oral anti-hyperglycemic agents (OHAs) and insulin therapy was in the range 36–93% and 62–64%, respectively11. Several treatment-, patient-, and physician-related factors have been identified to be associated with non-adherence to AHTs including the complexity of drug regimen, incorrect dose titration, fear of side-effects, and concern that patients may not be able to use therapy properly11–14.

Various insulin delivery systems have been introduced since the mid-1980s to ease patient burden associated with self-injection. According to a literature review by Molife et al., patient preference and satisfaction were greater with insulin pen delivery systems than vial/syringe systems15. Patients also reported greater convenience, flexibility, ease of use, self-efficacy, and dosing accuracy with a pen than with a vial/syringe15.

To date, four retrospective database analyses have been conducted to examine the outcomes of patients with T2DM using a prefilled pen versus a vial/syringe in administering insulin therapy16–19. The pen delivery system of focus in all four studies was the insulin aspart and aspart mix (Novolog, Novolog Mix70/30 and Novolog Mix50/50) FlexPen. Study results, as well as study design and population, were different across the four studies. Lee et al. and Cobden et al. reported that pens were associated with improved medication adherence and healthcare cost reduction versus vials/syringes17,18. These studies examined outcomes prior to and after patients converted from a vial/syringe to a pen, without a control group. A study by Pawaskar et al. conducted two separate analyses: (1) an insulin-naive analysis: patients were naive to insulin therapy and its delivery systems, and (2) a pen-naive analysis: patients who switched from a vial/syringe to a pen were compared to those who remained on a vial/syringe19. The authors concluded that diabetes-related medication adherence was significantly greater with a pen than with a vial/syringe for the pen-naive analysis, but not for the insulin-naive analysis19. Both diabetes-related and total costs were significantly lower with a pen than with a vial/syringe for the insulin-naive analysis, but not for the pen-naive analysis19. Last, Baser et al. compared outcomes between patients who switched from a vial/syringe to a pen and those who remained on a vial/syringe16. The pen delivery system was associated with greater medication adherence than the vial/syringe, but no significant difference in healthcare costs or hypoglycemia rates was observed16.

In January 2008, a new, easy-to-use prefilled pen delivery system, the KwikPen, was launched prefilled with insulin lispro (Humalog) in the US^†. To date, the economic outcomes of the KwikPen have not been compared to those of the vial/syringe. The purpose of this study is to compare US healthcare resource utilization, costs, medication adherence, and hypoglycemia rates in patients with T2DM initiating rapid-acting insulin analog (RAIA) with the KwikPen versus vial/syringe.

There are several ways in which the current study is different from previous insulin pen delivery system outcomes studies. The current study is not only the first study to compare real-world outcomes between the KwikPen and vial/syringe, but also includes healthcare resource utilization as a key outcome measure, and the study population includes T2DM patients who are naive to index insulin therapy as well as to the index pen delivery system.

Methods

Data source and study design

A retrospective analysis was conducted using a large US claims database (MarketScan)^‡. This dataset contains medical and pharmacy claims of >20 million US residents who were insured with a variety of commercial health plans. Data from January 1, 2007 to December 31, 2008 were used to compare the changes in outcomes 6 months prior to (pre-index) and 6 months after (post-index) initiating a RAIA using the KwikPen (for insulin lispro) versus vial/syringe (for insulin lispro or aspart) in patients with T2DM. Another form of RAIA, insulin glulisine (Apidra)^§ was not included in the primary analysis because the prefilled pen delivery system for insulin glulisine was not available during the study period. However, the authors conducted a sensitivity analysis with insulin glulisine included in the vial/syringe cohort to test the robustness of the primary results. Approximately 5% of patients who initiated RAIA using vial/syringe initiated with insulin glulisine.

Sample selection

A previously validated algorithm20 was used to identify patients with diabetes, in which patients were identified as having diabetes if they had ≥2 outpatient claims with a diagnosis code of diabetes (International Classification of Disease, 9th Edition, Clinical Modification [ICD-9-CM] codes 250.xx) on two different days or ≥1 inpatient claim with a diagnosis code of diabetes during the study period. Patients with T2DM were further identified by excluding patients with diagnosis of type 1 diabetes mellitus (ICD-9-CM codes 250.x1 or 250.x3).

The period from January 1, 2008 to June 30, 2008 was used to capture the index RAIA and respective delivery system of KwikPen or vial/syringe (denoted as ‘index RAIA’ from here on). If patients had at least one pharmacy claim for a new RAIA and the KwikPen was the first pen device used for RAIA, then the patients were categorized as the KwikPen cohort. Otherwise, if the vial/syringe was the delivery system associated with the first RAIA prescription, then those patients were categorized as the vial/syringe cohort. Patients were required to be continuously enrolled in the health plan during both the pre- and post-index periods.

Patients with the following criteria were excluded from the study: <18 years of age at index date; evidence of pregnancy (ICD-9-CM codes 761.5x, V22.xx, V72.40, V2.32, V23.86) or gestational diabetes (ICD-9-CM codes 648.0x or 648.8x) during the entire study period; any pharmacy claims of insulin pump, pump supplies, or inhaled insulin during the entire study period. Because the analysis only included those who were RAIA naive, patients were excluded if they had pharmacy claims of any RAIA or any insulin therapy used with the KwikPen during the pre-index period. Patients with total costs during the study period that were >99% of the sample were also excluded to minimize cost outliers. Patients who switched RAIA or the delivery system during the post-index period were further excluded from the study.

Outcome measures

Baseline characteristics included patient age, gender, geographic region of residence, provider type and time period associated with the index RAIA, insurance information, complications, comorbidities, medication use, and adherence. Diabetes-related complications included microvascular diseases (neuropathy, nephropathy, and retinopathy) and macrovascular diseases (cardiovascular diseases and peripheral circulatory diseases). Comorbidities included hypertension, disorders of lipid metabolism, obesity, depression, dexterity-impacting diseases, and diseases of the esophagus, stomach, and duodenum. Both complications and comorbidities were identified using ICD-9-CM codes or Current Procedural Terminology (CPT) codes.

Resource utilization was largely divided into pharmacy and medical utilizations. Both were reported as number (frequency) per patient in the 6-month pre- and post-index periods, separately. The pharmacy utilization was measured by number of claims and the categories included OHAs (metformin, sulfonylureas [SUs], thiazolidinediones [TZDs], alpha-glucosidase inhibitors [AGIs], non-SU secretagogues, dipeptidyl peptidase-4 [DPP-4] inhibitors, and the fixed-dose combination therapies), insulin, exenatide/pramlintide (considered together), self-monitoring blood glucose (SMBG), and other drugs of interest (diuretics, angiotensin-converting enzyme inhibitors, angiotensin-receptor blockers, β-blockers, calcium-channel blockers, anticoagulants, antiplatelets, statins, and the fixed-dose combination therapies). Insulin use during the pre-index period was categorized as ‘insulin-naive’ if a patient did not have any pharmacy claims of insulin therapy; ‘basal pen’ if a patient used long-acting insulin (insulin glargine or insulin detemir) with a pen; ‘basal vial’ if a patient used long-acting insulin with a vial/syringe; ‘other pen’ if a patient used any other insulin (e.g., neutral protamine Hagedorn [NPH], premixed insulin, etc.) with a pen; or ‘other vial’ if a patient used any other insulin with a vial/syringe. A separate variable was created to identify analog versus human insulin during the pre-index period if a patient used any insulin therapy.

The medical utilization was categorized as outpatient utilization (counts of provider and facility visits on different days), inpatient utilization (count of hospital admissions), ER visits (count of ER visits on different days), and number of HbA_1c tests on different days. The medical utilization was further identified as diabetes-related if it had ≥1 medical claim with a diagnosis of diabetes (ICD-9-CM codes 250.xx).

Cost variables were created using the amount reimbursed in the claims database during the 6-month pre- and post-index periods, separately. Costs were analyzed as total direct costs, total diabetes-related costs, and the following subgroups: pharmacy, outpatient, inpatient, and emergency room costs. All costs were adjusted to 2009 US Dollars (USD) using the Consumer Price Index. The diabetes-related costs for outpatient service, inpatient service, and ER were defined as costs associated with a claim with diabetes diagnosis codes (ICD-9-CM 250.xx). Diabetes-related pharmacy costs included costs of AHTs (OHAs, insulin, exenatide/pramlintide) and SMBG.

Proportion of days covered (PDC) was used to measure adherence to the index RAIA during the post-index period. PDC was calculated by dividing the number of days with the index RAIA on hand (using the ‘days supplied’ field in pharmacy claims) in the post-index period by 6 months of the post-index period21. If there was an overlap in days supplied of the index RAIA, the patients were assumed to finish the current prescription before starting the next prescription22. When days supplied was missing, average days supplied was imputed using days supplied of RAIA with the same NDC code and quantity dispensed in the study sample. PDC is a flexible adherence measurement, and PDC for a single drug, such as the index RAIA, can be considered as an adjusted medication procession ratio (adjusted MPR, total days supplied divided by number of days in the analysis period, capped at 1)23. PDC is also recommended to measure adherence to a class of drugs24,25, and it was used to calculate the baseline adherence to AHTs and SMBG in this study.

The rate of hypoglycemia was counted during the pre- and post-index periods, separately. A hypoglycemia event was identified using ICD-9-CM codes 251.0x, 251.1x, 251.2x, and 250.3x for outpatient or inpatient services on different days. A hypoglycemia event during an ER visit followed a validated coding algorithm26 as an ER visit with claims of the aforementioned ICD-9-CM codes and with claims of ICD-9-CM codes 251.8x, but excluding ICD-9-CM codes 259.8x, 272.7, 681.xx, 682.xx, 686.9x, 707.1x–707.9x, 709.3x, 730.0x–730.2x, and 731.8x.

Statistical analysis

The propensity score matching method was used to control for potential sample selection bias across the two study cohorts. The logistic regression model determined the propensity score of initiating vial/syringe versus the KwikPen. All patient characteristics in Table 1 and baseline outcome variables (costs, medical and pharmacy utilization, rate of utilization, number of HbA_1c tests, adherence to AHT and SMBG) were used in the logistic regression. To balance the variance and the bias in the sample, the caliper matching strategy was implemented. The matches were kept if the propensity score in the vial/syringe cohort was within a pre-defined radius (≤0.01) of the propensity score in the KwikPen cohort. Patients in the KwikPen and vial/syringe cohorts were matched on a 1:3 ratio. For each patient in the KwikPen cohort, at least one and up to three patients in the vial/syringe cohort were selected as long as the patients had a propensity score ≤0.01 that of the patients in the KwikPen cohort. The patient characteristics were compared across study cohorts using a t-test for continuous variables and the Pearson chi-square test for categorical variables, both before and after the matching process. The changes in outcomes were calculated by subtracting pre-index outcome from post-index outcome for each cohort. Then, the changes in outcomes were compared between the two cohorts using a difference-in-difference (DID) analysis that subtracted the difference in the KwikPen cohort from the difference in the vial/syringe cohort. Because cost data were skewed, the Wilcoxon rank-sum test was used to conduct the analysis on costs. Other outcomes (adherence, hypoglycemia rate, and utilization) and the corresponding changes were compared using t-tests.

Table 1.  Patient characteristics before and after propensity score matching.

	Before propensity score matching			After propensity score matching
	KwikPen	Vial/syringe	p-value*	KwikPen	Vial/syringe	p-value*
Sample size (n)	268	3,820		239	590
Mean age (years)	58	62	<0.001	59	60	0.224
Gender (%)
Male	52.2	51.2	0.731	53.1	51.5	0.674
Female	47.8	48.8	0.731	46.9	48.5	0.674
Geographic region (%)
Northeast	10.4	7.2	0.048	10.0	9.2	0.691
North central	40.3	32.6	0.010	40.6	40.8	0.945
South	37.3	39.3	0.516	38.1	39.5	0.705
West	11.9	20.9	0.000	11.3	10.5	0.740
Population density (%)
Rural	16.0	18.4	0.339	16.3	16.4	0.966
Urban	84.0	81.6	0.339	83.7	83.6	0.966
Provider type associated with the index RAIA (%)
Primary care physician	46.3	38.5	0.011	47.7	48.6	0.805
Endocrinologists	22.0	5.9	<0.001	18.8	19.3	0.870
Other	31.7	55.6	<0.001	33.5	32.0	0.870
Time of index RAIA (%)
First quarter of 2008	28.0	50.8	<0.001	30.1	29.8	0.933
Second quarter of 2008	72.0	49.2	<0.001	69.9	70.2	0.933
Insurance plan source (%)
Commercial	72.0	57.8	<0.001	70.3	66.8	0.327
Medicare^†	28.0	42.2	<0.001	29.7	33.2	0.327
Insurance plan type (%)
Indemnity plan	18.7	20.8	0.411	47	20.0	0.913
Other	81.3	79.2	0.411	192	80.0	0.913
Copayment amount ($, mean)
Insulin	27.0	22.1	0.000	26.9	26.7	0.880
OHAs	15.6	13.7	0.032	15.3	14.8	0.683
CCI (mean)	2.1	2.2	0.295	2.1	2.2	0.520
Diabetes-related complications^‡ (%)
Neuropathy	17.2	22.1	0.060	17.2	18.5	0.655
Nephropathy	14.9	18.1	0.195	15.5	16.8	0.648
Retinopathy	8.2	10.7	0.202	8.4	8.1	0.912
Ischemic heart diseases	18.7	21.7	0.237	18.8	19.7	0.784
Cerebrovascular diseases	5.6	10.7	0.008	6.3	6.4	0.930
Congestive heart failure	4.9	12.4	0.000	5.4	5.1	0.835
Peripheral vascular diseases	5.2	9.0	0.035	5.4	6.9	0.425
Comorbidities^‡ (%)
Hypertension	30.6	36.5	0.052	30.5	31.0	0.894
Disorders of lipid metabolism	27.6	20.2	0.004	25.5	23.9	0.622
Obesity	1.9	2.2	0.698	2.1	2.7	0.607
Depression	3.4	6.6	0.035	3.3	3.1	0.825
Diseases of esophagus, stomach, and duodenum	7.1	7.0	0.977	6.3	9.5	0.134
Dexterity-impacting diseases	6.0	9.4	0.059	5.4	5.9	0.783
Pre-index insulin use^§ (%)
Insulin naive	29.5	39.1	0.002	33.1	38.5	0.143
Basal pen	32.1	3.3	<0.001	25.5	15.8	0.001
Basal vial	27.2	41.8	<0.001	29.7	33.1	0.350
Other pen	9.0	1.2	<0.001	6.7	4.6	0.213
Other vial	11.9	21.4	0.000	13.4	16.1	0.326
Use of insulin: analog versus human (%)
Analog	63.8	48.2	<0.001	59.8	52.9	0.069
Human	10.4	17.4	0.004	10.9	12.2	0.593
Use of OHAs (%)
No OHAs	23.5	44.5	<0.001	25.1	25.3	0.964
1 OHA	26.9	29.0	0.455	28.5	30.3	0.591
>1 OHAs	49.6	26.5	<0.001	46.4	44.4	0.593
Use of exenatide/pramlintide (%)	14.2	5.6	<0.001	12.1	11.0	0.646
Use of SMBG (%)	30.2	28.9	0.638	29.7	25.9	0.268
Adherence to AHTs and SMBG^¶ (%)
PDC <50%	45.5	58.1	<0.001	46.0	44.6	0.704
PDC ≥50% and <80%	41.0	31.2	0.001	40.6	39.7	0.806
PDC ≥80%	13.4	10.7	0.162	13.4	15.8	0.387

*T-test was used for continuous variables and chi-squared test was used for categorical variables.

^†Indicates Medicare Supplemental Plan only. Information on basic Medicare plan is not included.

^‡ICD-9-CM and CPT codes were used to identify diabetes-related complications and comorbidities.

^§The variable, ‘pre-index insulin use’ was categorized as: ‘insulin naive’ if a patient did not have any pharmacy claims of insulin therapy, ‘basal pen’ if a patient used long-acting insulin (e.g., insulin glargine) with a pen, ‘basal vial’ if a patient used long-acting insulin with vial/syringe, ‘other pen’ if a patient used any other insulin (e.g., premixed insulin) with a pen, or ‘other vial’ if a patient used any other insulin with vial/syringe.

^¶AHTs are inclusive of OHAs (metformin, sulfonylureas (SUs), thiazolidinediones (TZDs), alpha-glucosidase inhibitors (AGIs), non-SU secretagogues, dipeptidyl peptidase-4 (DPP-4) inhibitors), insulin, and exenatide/pramlintide.

OHAs, oral antihyperglycemic agents; RAIA, rapid-acting insulin analog; CCI, Charlson Comorbidity Index; SMBG, self-monitoring blood glucose; AHTs, anti-hyperglycemic agents; PDC, proportion of days covered.

Post hoc and sensitivity analyses were conducted to examine the impact of the KwikPen and vial/syringe on sub-groups of cost outcomes and to test the robustness of the results. In the post hoc analyses, diabetes-related pharmacy and medical costs were further broken down into the following sub-groups: pharmacy (OHAs and the subgroups of metformin, SUs, TZDs, DPP-4 inhibitors, other single OHAs, and fixed-dose combination; insulin and the subgroups of long-acting analog, NPH, RAIA, short-acting human insulin [SAHI], premixed analog, and premixed human; exenatide/pramlintide; and SMBG), outpatient costs (healthcare provider costs and the subgroups of primary care physicians [PCPs], endocrinologists, cardiologists, and other providers; facility costs; and other outpatient costs), and inpatient costs (primary reason for hospitalization, including microvascular complications, macrovascular complications, and other hospitalization). Sensitivity analysis was conducted by including patients who initiated insulin glulisine in the vial/syringe cohort.

Results

Patient characteristics

The flow of study sample selection is illustrated in Figure 1. A total of 1,518,441 patients were identified as having diabetes, of which 1,229,723 patients were determine to have T2DM. Among patients with T2DM, 14,865 patients had ≥1 pharmacy claim for index RAIA, including insulin lispro KwikPen (n = 590), insulin lispro vial/syringe (n = 8,424), and insulin aspart vial/syringe (n = 5,851). The remaining sample selection criteria led to a final sample of 268 patients initiating insulin lispro KwikPen and 3,820 patients initiating insulin lispro or insulin aspart vial/syringe. Before propensity score matching, multiple patient characteristics were different between the two study cohorts (Table 1). Most notably, patients who initiated a RAIA using the KwikPen were younger than those who initiated using a vial/syringe (mean: 58 vs. 62 years, p < 0.001). A greater proportion of patients in the KwikPen cohort had endocrinologists as their index physician type and commercial health plan as their insurance source than those in the vial/syringe cohort (p < 0.001 for both comparisons). Pre-index insulin use was significantly different between the two cohorts. For example, fewer patients in the KwikPen cohort were insulin-naive during the pre-index period than the vial/syringe cohort (p = 0.002). Among patients who used insulin during the pre-index period, a greater proportion of patients in the KwikPen cohort used insulin with a pen (‘basal pen’ and ‘other pen’) than the vial/syringe cohort (p < 0.001 for both comparisons). These differences corroborate a previous study that examined the key predictors of pen versus vial initiation in patients with T2DM27.

Figure 1.  Study sample flow diagram.

The propensity score matching method was applied successfully, with a success rate of 89%. A total of 590 vial/syringe cohort patients were matched to 239 patients in the KwikPen cohort. After matching, the two cohorts were balanced in terms of baseline characteristics. There were no significant differences across the matched cohorts across approximately 20 different variables (Table 1), except for the variable ‘basal pen’. The use of ‘basal pen’ during the pre-index period was greater in the KwikPen cohort than the vial/syringe cohort, although the magnitude of difference was reduced after matching (post-match: 25.5 vs. 15.8%, p = 0.001; pre-match: 32.1 vs. 3.3%, p < 0.001).

Resource utilization and costs

After matching, there were no significant differences in resource utilization across the two cohorts during the pre-index period (Table 2). In the post-index period, the KwikPen cohort had significantly lower utilization of OHAs compared to the vial/syringe cohort (mean count: 2.2 vs. 2.7, p = 0.021). From pre- to post-index periods, the OHA utilization decreased in both cohorts, but the reduction was greater in the KwikPen cohort than in the vial/syringe cohort (mean count: –1.3 vs. –0.7, p = 0.016, negative value representing reduction: DID = 0.6). The KwikPen and vial/syringe cohorts were similar in other post-index utilizations and changes in utilization from pre- to post-index periods.

Table 2.  Results of adherence and difference-in-difference (DID) analysis on resource utilization and hypoglycemia rate.

	Pre-index			Post-index			Difference from pre-index to post-index^†
	KwikPen (n = 239)	Vial/syringe (n = 590)	p-value*	KwikPen (n = 239)	Vial/syringe (n = 590)	p-value*	KwikPen (n = 239)	Vial/syringe (n = 590)	p-value*	DID^‡
	[A]	[B]		[C]	[D]		[E] = [C] – [A]	[F] = [D] – [B]		[G] = [F] – [E]
Pharmacy utilization (count, mean)
OHAs	3.4	3.4	0.896	2.2	2.7	0.021	–1.3	–0.7	0.016	0.6
All AHTs and SMBG	6.0	5.9	0.824	8.3	8.3	0.936	2.3	2.4	0.731	0.1
Other drugs of interest^§	6.8	6.9	0.907	7.4	7.6	0.734	0.6	0.7	0.721	0.1
Diabetes-related medical utilization  (count, mean)
Outpatient service	3.5	3.6	0.649	3.5	3.5	0.946	0.0	–0.1	0.717	–0.1
Inpatient admission	0.1	0.1	0.324	0.1	0.1	0.624	–0.0	–0.0	0.602	–0.0
Emergency room visits	0.1	0.1	0.845	0.1	0.1	0.245	0.1	0.0	0.312	–0.0
Total medical utilization  (count, mean)
Outpatient service	14.1	15.5	0.300	15.4	17.6	0.148	1.3	2.1	0.459	0.8
Inpatient admission	0.2	0.3	0.214	0.2	0.2	0.586	–0.1	–0.1	0.507	–0.0
Emergency room visits	0.4	0.4	0.580	0.4	0.5	0.517	0.0	0.0	0.900	0.0
HbA1c tests (count, mean)	0.9	0.9	0.262	0.7	0.7	0.679	–0.2	–0.1	0.461	0.1
Hypoglycemia rate  (episodes per patient over 6 months)	0.03	0.03	0.905	0.03	0.02	0.539	0.004	–0.003	0.707	–0.01
Adherence to index RAIA	—	—	—	54.6%	45.2%	<0.001	—	—	—	—

*T-test was used to compare the difference.

^†Negative value indicates reduction in outcomes. Positive value indicates increase in outcomes.

^‡DID = (difference from pre- to post-index in the vial/syringe cohort) – (difference from pre- to post-index in the KwikPen cohort). Values are rounded after subtraction.

^§Inclusive of diuretics, angiotensin-converting enzyme inhibitors, angiotensin-receptor blockers, beta blockers, calcium-channel blockers, anticoagulants, antiplatelets, and statins.

OHAs (oral antihyperglycemic agents), AHTs (antihyperglycemic agents; inclusive of OHAs, insulin, and exenatide/pramlintide), and SMBG (self-monitoring blood glucose).

There were no significant differences in costs across the two cohorts during the pre-index period (Table 3). The diabetes-related pharmacy costs increased from pre- to post-index periods in both cohorts, and the increase was greater in the KwikPen cohort by $294 than the vial/syringe cohort (+$900 vs. +$607, p < 0.001; DID = –$294). The total pharmacy costs also increased for both cohorts, but the difference in the increased amounts was not significant (+$1,171 vs. +$874, p = 0.058; DID = –$297). The KwikPen cohort showed a decrease in total diabetes-related costs of $235, while the vial/syringe cohort showed an increase in total diabetes-related costs of $61 (DID = $297, p = 0.006). There were no significant differences in other costs and changes in costs from pre- to post-index periods between the two study cohorts.

Table 3.  Results of difference-in-difference (DID) analysis on cost.

	Pre-index			Post-index			Difference from pre-index to post-index^†
	KwikPen (n = 239)	Vial/syringe (n = 590)	p-value*	KwikPen (n = 239)	Vial/syringe (n = 590)	p-value*	KwikPen (n = 239)	Vial/syringe (n = 590)	p-value*	DID^‡
	[A]	[B]		[C]	[D]		[E] = [C] – [A]	[F] = [D] – [B]		[G] = [F] – [E]
Diabetes-related costs ($, mean)
Pharmacy costs	963	880	0.203	1,864	1,486	<0.001	900	607	<0.001	–294
Outpatient costs	1,095	1,054	0.504	501	751	0.457	–594	–304	0.196	290
Inpatient costs	931	919	0.455	355	685	0.284	–576	–235	0.796	341
Emergency room costs	60	60	0.995	94	53	0.187	34	–7	0.120	–41
Total diabetes-related costs	3,049	2,914	0.623	2,814	2,975	0.008	–235	61	0.006	297
Total costs ($, mean)
Pharmacy costs	2,867	2,870	0.146	4,038	3,745	0.018	1,171	874	0.058	–297
Outpatient costs	5,023	5,150	0.787	4,836	5,440	0.112	–187	290	0.791	477
Inpatient costs	4,543	5,505	0.359	2,535	2,539	0.078	–2,008	–2,967	0.579	–958
Emergency room costs	220	230	0.849	297	230	0.551	77	0	0.828	–77
Total direct costs	12,652	13,755	0.500	11,706	11,954	0.461	–946	–1,802	0.814	–855

*Wilcoxon rank-sum test was used to compare the difference.

^†Negative value indicates reduction in costs. Positive value indicates increase in costs.

^‡DID = (difference from pre- to post-index in the vial/syringe cohort) – (difference from pre- to post-index in the KwikPen cohort). Values are rounded after subtraction.

Adherence and hypoglycemia rate

In the matched sample, adherence to the index RAIA during the post-index period (Table 2) was significantly greater in the KwikPen cohort than the vial/syringe cohort (PDC: 54.6 vs. 45.2%, p < 0.001). There were no significant differences in hypoglycemia rate between the two cohorts (episodes per patient over the 6-month pre-index period: 0.03 vs. 0.03, p = 0.905; episodes per patient over the 6-month post-index period: 0.03 vs. 0.02, p = 0.539; change from pre- to post-index period: +0.004 vs. –0.003, p = 0.707).

Post hoc and sensitivity analyses

In the post hoc analyses, significant differences were observed in several diabetes-related cost components between the KwikPen and vial/syringe cohorts (Table 4). Greater reduction in OHA costs was observed in the KwikPen than in the vial/syringe cohort (–$152 vs. –$60, p = 0.006), including greater reduction in the costs of SUs and DPP-4 inhibitors (–$22 vs. –$7, p = 0.003 and –$53 vs. –$10, p = 0.008, respectively). The KwikPen was also associated with a greater increase in insulin costs than the vial/syringe (+$982 vs. +$643, p < 0.001), including a greater increase in the costs of long-acting analog and RAIA (+$375 vs. +$201, p < 0.001 and +$669 vs. +$438, p < 0.001, respectively), with a slightly greater reduction in the cost of SAHI (−$9 vs. $0, p = 0.002). The increase in the cost of SMBG was significantly greater in the KwikPen cohort than the vial/syringe cohort (+$91 vs. +$41, p = 0.004). Among sub-categories of diabetes-related outpatient costs, the KwikPen was associated with a greater reduction in PCP costs (–$23 vs. –$1, p = 0.008) and outpatient facility costs (–$550 vs. –$314, p = 0.045). No significant differences were observed among sub-groups of diabetes-related inpatient costs across the cohorts.

Table 4.  Results of post hoc analyses on diabetes-related pharmacy, outpatient, and inpatient cost breakdowns.

Diabetes-related costs ($, mean)	Pre-index			Post-index			Difference from pre-index to post-index^†
	KwikPen (n = 239)	Vial/syringe (n = 590)	p-value*	KwikPen (n = 239)	Vial/syringe (n = 590)	p-value*	Kwiken (n = 239)	Vial/syringe (n = 590)	p-value*	DID^‡
	[A]	[B]		[C]	[D]		[E] = [C] – [A]	[F] = [D] – [B]		[G] = [F]–[E]
Pharmacy costs	963	880	0.203	1,864	1,486	<0.001	900	607	<0.001	–294
OHAs	414	358	0.373	262	298	0.172	–152	–60	0.006	92
Metformin	45	38	0.946	40	37	0.767	–5	–1	0.983	4
SU	38	29	0.113	16	22	0.025	–22	–7	0.003	15
TZD	161	173	0.589	105	142	0.324	–56	–32	0.404	24
Inhibitor DPP-4	96	53	0.002	43	43	0.451	–53	–10	0.008	43
Other single OHA	29	28	0.513	4	21	0.118	–25	–8	0.058	17
Fixed-dose combination	44	36	0.265	53	33	0.855	9	–3	0.709	–12
Insulin	379	356	0.328	1,362	999	<0.001	982	643	<0.001	–339
Long-acting analog	259	241	0.354	634	442	<0.001	375	201	<0.001	–173
NPH	12	21	0.075	9	23	0.002	–3	2	0.055	5
RAIA	1	1	0.866	671	439	<0.001	669	438	<0.001	–232
SAHI	10	3	0.012	0	3	0.070	–9	0	0.002	10
Premixed analog	77	72	0.828	46	84	0.231	–31	12	0.058	43
Premixed human	20	18	0.897	2	8	0.181	–18	–10	0.352	8
Exenatide/pramlintide	91	93	0.661	71	76	0.619	–20	–17	0.902	3
SMBG	79	73	0.308	170	114	0.001	91	41	0.004	–50
Outpatient costs	1,095	1,054	0.504	501	751	0.457	–594	–304	0.196	290
Health care provider costs	221	212	0.415	205	212	0.633	–15	0	0.392	15
Primary care physician^§	136	132	0.594	113	130	0.103	–23	–1	0.008	22
Endocrinologist	71	63	0.777	70	55	0.181	0	–7	0.446	–7
Cardiologist	2	3	0.877	2	3	0.732	0	1	0.415	0
Others^¶	12	15	0.778	181	359	0.415	8	8	0.816	0
Facility costs**	731	673	0.183	172	310	0.522	–550	–314	0.045	236
Others	143	169	0.307	114	179	0.274	–29	10	0.488	39
Inpatient costs	931	919	0.455	355	685	0.284	–576	–235	0.796	341
Microvascular complication related	1	57	0.303	0	80	0.091	–1	23	0.940	24
Macrovascular complication related	508	117	0.996	171	244	0.648	–336	127	0.512	463
Others	422	745	0.428	184	361	0.534	–238	–384	0.557	–146

*Wilcoxon rank-sum test was used to compare the difference.

^†Negative value indicates reduction in costs. Positive value indicates increase in costs.

^‡DID = (difference from pre- to post-index in the vial/syringe cohort) – (difference from pre- to post-index in the KwikPen cohort). Values are rounded after subtraction.

^§Inclusive of physician assistant and nurse practitioner costs.

^¶Inclusive of ophthalmologist, podiatrist, diabetes educator and dietitian costs.

**Inclusive of ambulatory surgery center, rehabilitation facilities, intermediate care, etc.

SUs, sulfonylureas; TZDs, thiazolidinediones; DPP-4, dipeptidyl peptidase-4 inhibitors; NPH, neutral protamine Hagedorn; SAHI, short-acting human insulin.

In the insulin glulisine sensitivity analysis, after matching, the final sample included 236 patients initiating insulin lispro using the KwikPen and 601 patients initiating insulin lispro, aspart or glulisine using vial/syringe. The results were consistent with the original results in this study. The KwikPen cohort was associated with a greater adherence than the vial/syringe cohort (PDC: 53.9 vs. 46.3%, p = 0.001). The diabetes-related pharmacy costs increased for both cohorts but the increase was greater in the KwikPen cohort than the vial/syringe cohort by $330 (+$942 vs. +$612, p < 0.001; DID = –$330). However, the reduction in the total diabetes-related costs was greater in the KwikPen cohort than the vial/syringe cohort by $110 (–$246 vs. –$136, p = 0.004; DID = $110) as well as the reduction in the OHA utilization (–1.2 vs. –0.7, p = 0.024; DID = 0.5).

Discussion

Using a large healthcare claims database, the current study evaluated the impact of differing insulin delivery systems on outcomes in patients with T2DM who newly initiated RAIA. Despite the greater increase in diabetes-related pharmacy costs associated with initiating RAIA with the KwikPen versus the vial/syringe, the use of the KwikPen was associated with several key positive outcomes: (1) there was a greater reduction in the total diabetes-related costs (potential cost offset) with the KwikPen than with the vial/syringe (DID: $297 per patient); (2) there was a greater reduction in the utilization of OHAs with the KwikPen than with the vial/syringe (DID in mean count: 0.6; mean cost: $92); and (3) medication adherence was also greater with the KwikPen than with the vial/syringe (difference in PDC: 9.5%). There were no significant differences in the rate of hypoglycemia; total direct costs; or diabetes-related outpatient, inpatient, and ER costs between the two cohorts.

Post hoc analyses provided further differences among various cost sub-groups. In the post hoc analyses, the KwikPen was associated with greater reductions in OHA costs, specifically in the costs of SUs and DPP-4 inhibitors and increases in the cost of insulin therapies, including long-acting insulin and RAIA, when compared to the vial/syringe cohort. A potential hypothesis behind these changes (reduction in OHA costs and increase in insulin costs) is that the increased utilization (i.e., prescription fills) of insulin and intensification of insulin therapy would be associated with reduction in OHA use. When outpatient costs were broken down further into several sub-groups, KwikPen was associated with significant reductions in PCP visit costs and outpatient facility costs than the vial/syringe cohort. There are several potential hypotheses behind this study finding. For example, PCP visits may not be as frequent as expected since 70% of the KwikPen and 60% of the vial/syringe cohorts in this study (before propensity score matching) had experience in using insulin prior to initiating RAIA.

Selective positive outcomes in the current study corroborated findings from several previous studies comparing prefilled pen versus vial/syringe. A significant reduction in the diabetes-related healthcare costs was observed after patients either switched from vial/syringe to a prefilled pen or initiated the prefilled pen in studies by Lee et al., Cobden et al., and Pawaskar et al.17–19. Adherence was also greater with the prefilled pen than the vial/syringe in these studies and in a study by Baser et al., except for the insulin-naive analysis in the Pawaskar et al. study16–19. Similar to the current study, Baser et al. estimated the hypoglycemia rate by counting the number of hypoglycemia events and found no significant differences between the prefilled pen and vial/syringe cohorts16. In studies by Lee et al. and Cobden et al., there was significant reduction in the percentage of patients experiencing hypoglycemia events when patients converted from vial/syringe to prefilled pen17,18. Hypoglycemia was not reported by Pawaskar et al.19.

There are several limitations to this study, the majority of which are inherent to all retrospective database analyses involving patients with T2DM. First, adherence measure using pharmacy refill records is not optimal since medication use during a hospitalization, potential medication sharing or wastage cannot be assessed. However, PDC has been endorsed by the National Quality Forum to indicate the medication management quality25 and can measure adherence to one medication or a class of medications24. In addition, PDC is commonly used in the literature21,23,25. Second, the degree of glycemic control (e.g., HbA_1c) which may impact healthcare resource utilization is not identifiable in claims databases. The current study used surrogate markers such as diabetes-related complications, comorbidities, and concomitant medications to adjust for T2DM severity. Additional variables are not present in the claims database (e.g., smoking status). Applying advanced methods to control the unobserved as well as observed sample selection bias could provide an important contribution to treatment evaluation in patients with T2DM. Third, since hypoglycemia rates were calculated using ICD-9-CM codes, which require some clinical intervention, mild to moderate events may not be captured. Fourth, the results may not represent un-insured population or elderly patients with only basic Medicare coverage. The MarketScan database does not include all Medicare claims for those who are ≥65 years old.

Conclusion

The study findings suggest that patient outcomes vary by different insulin delivery systems (prefilled pen versus vial/syringe) when patients with T2DM initiate RAIA. Initiating RAIA with the KwikPen was associated with a greater increase in diabetes-related pharmacy costs but better RAIA adherence, greater reduction in total diabetes-related costs, and OHA utilization than vial/syringe. There were no differences in the changes to the total healthcare costs.

Declaration of interest

This study was financially supported by Eli Lilly and Company (Indianapolis, IN, USA). L.J.L. and M.D.P. are employees and stock owners of Eli Lilly and Company. S.M.C. is an employee of Lilly USA, LLC and a stock owner of Eli Lilly and Company. Q.L. and M.W.R. are employees of United BioSource Corporation (Lexington, MA, USA). Eli Lilly and Company contracted with United BioSource Corporation for the conduct of this study.

Transparency

Acknowledgments

The authors would like to thank Khaled Sarsour, PhD, an employee of Eli Lilly and Company, for his input on the development of the study hypothesis.

Notes

* Novolog, Novolog Mix70/30, Novolog Mix50/50, and FlexPen are registered trademarks of Novo Nordisk Pharmaceuticals Inc. (Princeton, NJ, USA)

^† Humalog and the KwikPen are registered trademarks of Eli Lilly and Company (Indianapolis, IN, USA)

^‡ MarketScan is a registered trademark of Thomson Reuters (Ann Arbor, MI, USA)

^§ Apidra is a registered trademark of Sanofi-Aventis US LLC (Bridgewater, NJ, USA)