Urvi Desai, PhD

Insulin, a hormone produced by the pancreas, is important for blood glucose management. For a person whose body cannot produce insulin (patients with type 1 diabetes mellitus) or use it adequately (patients with type 2 diabetes mellitus), insulin therapy is often required to maintain adequate blood glucose levels.

 More than 20 insulin products are available in the United States, and they can be categorized as synthetic analogues or biosynthetic human insulin based on the method used for production, and as rapid-acting, short-acting, intermediate-acting, or long-acting based on the time they take to start working in the body and the duration of action.

 These products come in the form of vials, pre-filled pens, or cartridges for reusable pens, containing insulin as a solution or suspension of certain strengths. Typically, for products available in the United States, 1 mL solution (or suspension) equals 100 units of insulin, except for regular U-500 (concentrated) human insulin (Humulin R U-500) for which 1 mL equals 500 units. The daily dose and type of insulin required by patients vary depending on patient characteristics such as lifestyle and body weight.

A majority of the insulin products in the United States are supplied by Eli Lilly and Company (LLY) (ie, insulin lispro and human insulin), Novo Nordisk (NN) (ie, insulin aspart, human insulin, and insulin detemir), and Sanofi-Aventis (ie, insulin glulisine and insulin glargine). Insulin lispro and insulin aspart have similar pharmacokinetic/pharmacodynamic properties

 and are respectively used in several comparable LLY and NN products. In addition, both LLY and NN make comparable human insulin products. Studies have indicated that insulin glulisine has potentially different absorption and action profiles than other rapid-acting insulins.

 Further, insulin detemir and insulin glargine, both long-acting insulins, have different dosing regimens and action profiles.

 Thus, these products are excluded from the current analysis.

However, even within the comparable products, the availability of delivery device options differs between the 2 manufacturers. For example, LLY’s human insulin isophane is available in both a pen and a vial, whereas NN’s is only available in a vial. Therefore, for all products combined, the per claim per day units and costs of comparable insulins may differ between LLY and NN, depending on the types of insulins used (eg, analogue vs human, pre-mixed vs not), their associated delivery devices (eg, pens vs vials), and characteristics of patients (eg, age, gender, comorbidity profile, and type of physician[s] seen) using these products. Having such quantitative information on the entire portfolio of comparable insulin products (see 

, for a complete list) offered by LLY and NN can help payers make formulary coverage decisions for this range of products. Therefore, the objective of this study was to use administrative claims data to compare units and costs of comparable LLY and NN insulin products supplied per claim per day in 2011. Because patient characteristics and product mix may affect the comparison, the analysis was conducted with and without accounting for such differences.

This was a retrospective analysis of de-identified employer-based commercial insurance claims data obtained from OptumHealth Reporting and Insights. The database contains enrollment history, patient demographics, medical claims, and drug claims for the years 1999 to 2011 for approximately 16 million beneficiaries from 60 large, self-insured companies with locations across the United States. The medical claims contain dates of service, up to 2 

International Classification of Diseases, Ninth Revision, Clinical Modification

) diagnosis codes, procedure codes based on Current Procedural Terminology 4, Healthcare Common Procedure Coding System, and 

 procedure codes, place of service (eg, inpatient, outpatient, physician’s office), and physician specialty (eg, primary care, endocrinology, psychiatry), which were all used to estimate baseline characteristics. The prescription drug claims provide information on date of fill, days supplied, quantity dispensed, and National Drug Codes, which were used to estimate the units of insulin supplied per claim per day. In addition, the prescription drug claims included costs paid by insurers which were used to estimate the costs of insulin per claim per day from a payer’s perspective.

Patients with at least 1 claim for comparable LLY or NN insulin between January 1 and September 30, 2011, were included in the analysis. Claims with missing or nonpositive values for days supplied, quantity supplied, or cost were excluded. In addition, claims with quantities dispensed (in mL) equal to or exceeding the package size (in mL) multiplied by the strength (1 mL = 100 units for all products) were deleted due to uncertainty regarding the dispensing unit. Patients were required to have continuous coverage for ≥6 months prior to their first insulin claim in 2011 through health plans that had ≥100 LLY and NN insulin claims, ≥5% of which were attributable to each manufacturer (

). The 6-month period before the first insulin claim in 2011 (baseline) was used to evaluate patients’ demographic profile (age and gender), comorbidity profile (select comorbidities and Charlson comorbidity index [CCI],

 excluding diabetes), type of health insurance (preferred provider organization [PPO], health maintenance organization [HMO], indemnity, point-of-service [POS], or other), antidiabetic medication use (ie, insulin use, insulin pump use, oral antidiabetic medications alone or in combinations and other antidiabetic injectables), and proportions of patients with visits to an endocrinologist. Patients were further divided into 2 subgroups: aged <65 years and ≥65 years. Age was evaluated at the time of each patient’s first insulin claim in 2011.

The primary end point was the difference in mean units/claim/day between comparable LLY and NN insulin products, adjusting for differences in underlying patient characteristics. Because the quantity of insulin supplied was recorded in the database in mLs, units/claim/day were calculated by multiplying total quantity per claim (in mL) by strength and dividing by total days supplied. The units/claim/day were estimated for all patients, as well as separately for patients aged <65 years old and patients aged ≥65 years, as dosage requirements for those aged ≥65 years may be different from those aged <65 years.

 Among those aged <65 years, regression-adjusted units/claim/day were also assessed separately for patients aged <18 years and those aged 18 to 64 years.

In addition, the difference in mean costs/claim/day for comparable LLY and NN insulins was estimated adjusting for differences in underlying patient characteristics. Costs represented the paid amount from insurers. Cost/claim/day was calculated by dividing the cost per claim by total days supplied. Cost/claim/day was only estimated separately for patients aged <65 years and those aged ≥65 years, because the database contains information on costs to commercial insurers, which may often be equal to $0 for patients aged ≥65 years (due to Medicare coverage), and would lead to inaccurate cost estimates for all patients combined.

Chi-square tests were used to descriptively compare categorical variables (eg, comorbidity rates and proportions of patients using antidiabetic medications), and Student’s 

 tests were used to compare baseline age and CCI. The distribution of units/claim/day and cost/claim/day for LLY vs NN insulins was descriptively compared between the 2 groups for all comparable products combined, by package type (pens vs vials), as well as by each comparable product category using Wilcoxon rank-sum test. The mean units/claim/day and cost/claim/day were estimated using individual claims for each patient. While the descriptive analyses did not account for within-patient correlation, it was adjusted for in the multivariate analyses as described below.

For the primary analyses, mean units/claim/day adjusting for baseline differences was estimated using generalized estimating equation (GEE) models with an identity link function and normal distribution of the error term. Mean cost/claim/day adjusting for baseline differences were estimated using a log link function and gamma distribution of the error term. All models controlled for age on first LLY or NN insulin claim in 2011, gender, type of health insurance, endocrinologist visits, CCI (excluding diabetes), insulin pump use, select comorbidities with diagnosed prevalence ≥5%, and prescription drug classes used by ≥5% of the sample. The models also accounted for repeated patient claims, and an exchangeable working correlation structure was used for the within-patient correlations. 

 values <.05 were defined as being statistically significant.

To assess the sensitivity of results to the statistical method used for adjusting for patient characteristics, models using matched subjects were estimated. Propensity score matching was used to identify populations of patients with similar baseline characteristics who were treated with comparable LLY and NN insulins. First, the propensity of being prescribed LLY insulin versus comparable NN insulin was estimated using logistic regression models that accounted for the same baseline patient characteristics as the multivariate models. Following this, for each comparable product category, patients were matched 1:1 using greedy matching (which identifies the matched pairs such that the difference within each matched pair, independent of the effect on the entire cohort, is minimized)

 on the propensity score (within a caliper of a quarter of the standard deviation) and on the number of claims for the respective comparable products in 2011. Similar to the primary analyses, GEE models were used to estimate mean units/claim/day adjusting for differences remaining after matching.

All analyses were conducted using SAS version 9.2 (SAS Institute, Cary, North Carolina).

Overall, 42,173 patients met the sample selection criteria (Table 1). Of the final sample, 24,616 and 20,705 respectively had at least 1 claim for comparable LLY or NN insulin products in 2011. On average, patients in both groups were aged 58.8 years at the time of their first insulin claim in 2011, and 53.1% of both samples were male. Patients in both groups were most likely to have PPO-type health insurance coverage, followed by indemnity, HMO, and POS, although the proportions of patients enrolled in each health insurance type were substantially different between the groups (

). During baseline, lower proportions of LLY insulin users were diagnosed with most comorbidities evaluated and had lower mean CCI versus comparable NN insulin users (0.9 ± 1.7 vs 1.1 ± 1.8; 

<.0001). Fewer LLY insulin users had at least 1 visit to an endocrinologist (24.7% vs 25.6%), but greater proportions used insulin—particularly insulin pump&mdash;in the baseline period, compared with NN insulin users. In addition, the proportions of patients using oral antidiabetic medications or antidiabetic injectables during baseline were lower for the LLY group versus the NN group (Table 2).

Unadjusted and Adjusted Units per Claim per Day

Before adjusting for differences in baseline characteristics, the overall mean units/claim/day were significantly lower for all comparable LLY insulins combined than those of all comparable NN insulins combined (73.9 vs 74.7; 

= .0028) among all patients. The mean units/claim/day were also lower for all comparable LLY vials combined; however, for comparable pens, the difference in mean units/claim/day between the 2 groups was not statistically significant (

). Within the comparable products, the units/claim/day were similar for LLY and NN insulins, with the exception of significantly lower units/claim/day of LLY human insulin isophane vials versus NN human insulin isophane vials, LLY regular human insulin vials vs NN regular human insulin vials, and LLY human insulin 70/30 vials vs NN human insulin 70/30 vials (Table 3). The difference in units/claim/day for human insulin isophane and regular human insulin vials remained after accounting for differences in baseline patient characteristics using multivariate regression. However, after adjusting for baseline differences, the mean units/claim/day were similar for LLY human insulin 70/30 vials and NN human insulin 70/30 vials (Table 3). In addition, the overall mean units/claim/day for all comparable products combined were not statistically different after adjusting for differences in baseline patient characteristics (LLY: 72.3 vs NN: 72.8; 

Among patients aged <65 years, after adjusting for baseline characteristics, the mean units/claim/day were significantly lower for LLY vials combined than comparable NN vials (LLY: 80.4 vs NN: 82.9; 

= .0240), although the overall units/claim/day for all comparable products combined were similar between the 2 groups (

). The results for those aged ≥65 years were similar to the results for all patients (except the overall mean units/claim/day for all products combined were also significantly lower for LLY insulins: 66.6 vs 68.7 for NN; 

= .0224) (Table 4). In addition, the results for regression-adjusted units/claim/day among patients aged <18 years and those aged 18 to 64 years (

) were similar to the results for patients aged <65 years, except that among patients aged 18 to 64 years, the difference in units/claim/day for all vials with comparable products combined was not significant (Table 4).

Unadjusted and Adjusted Costs per Claim per Day

For patients aged <65 years, the mean cost/claim/day for pens with comparable products was similar for the 2 groups (LLY: $8.9 vs NN: $8.8; 

= .6670), while the mean cost/claim/day for LLY insulin vials was significantly lower than that of NN insulin vials ($6.4 vs $7.0; 

<.0001). This resulted in lower overall cost/claim/day for comparable LLY insulins ($7.2 vs $7.9; 

<.0001). Within each comparable product category, the cost/claim/day for both groups was similar except for significantly higher cost/claim/day for insulin lispro vials, LLY human insulin isophane vials, and LLY human insulin 70/30 mix vials versus comparable NN vials (

). The results were similar after adjusting for baseline differences (Table 5).

Among patients aged ≥65 years, before adjusting for baseline differences, mean cost/claim/day was lower for comparable LLY insulin versus NN insulin ($5.9 vs $6.9; 

<.0001), again due to lower mean cost/claim/day for vials. In addition, mean cost/claim/day of LLY insulin was significantly lower than that of NN insulin within all but 3 comparable product categories. Mean cost/claim/day was higher for LLY insulin lispro pens than for NN insulin aspart pens and similar between LLY and NN insulin analogue mixture products in both pens and vials (Table 5). However, after adjusting for differences in baseline characteristics, mean cost/claim/day for each comparable product category was similar except for LLY regular human insulin vials and all LLY insulin vials combined, for which the mean cost/claim/day was significantly lower than for comparable NN products. This resulted in significantly lower cost/claim/day for all comparable products combined (Table 5).

After adjusting for differences in patient characteristics remaining post propensity score matching, the mean units/claim/day of LLY insulins were similar to those of NN insulins within all comparable product categories but 1 (units/claim/day were significantly lower for LLY regular human insulin vial). This is similar to the results from the multivariate analyses, suggesting that the results are robust to the method used for adjusting for baseline differences (

More than 20,000 patients each had 1 or more claims for a comparable LLY or NN insulin product in 2011. At baseline, the 2 groups were aged approximately 59 years, predominantly male (approximately 53%), and had similar enrollment patterns for different types of health insurance plans. However, the 2 groups were different with regard to the rates of diagnosed comorbid conditions and antidiabetic medication use at baseline. Fewer patients using LLY insulin were diagnosed with comorbid conditions such as hypertension and depression at baseline, and fewer LLY patients visited an endocrinologist or used oral antidiabetic medications. However, LLY insulin users were more likely to have used insulin in the baseline period, which may be an implication of greater shares of these patients having diagnoses for type 1 diabetes mellitus during baseline.

The unadjusted overall mean units/claim/day for all comparable products were significantly lower for LLY insulins than NN insulins but were similar upon adjustment for differences in baseline patient characteristics. This is consistent with the hypothesis that patients with similar characteristics require similar amounts of comparable insulin products, independent of the manufacturer. In addition, although the cost/claim/day of insulin products were different between the 2 groups in the descriptive analyses, regression-adjusted cost/claim/day for most comparable product categories was similar. The difference in overall cost/claim/day (lower costs for LLY vs comparable NN products combined) may be attributable to the difference in product mix for the 2 manufacturers (greater shares of LLY claims were for vials and human insulins, which had lower cost/claim/day than pens and analogue insulins, respectively).

To the best of our knowledge, this study is the first to use administrative claims data to compare the units/claim/day and cost/claim/day of comparable insulin products manufactured by 2 different companies from a payer’s perspective. Because recent administrative claims data were used for this analysis, the results may aid payers in making formulary decisions regarding the comparable insulin products manufactured by LLY and NN. While the results may not be useful when making decisions about the use or cost of insulin products by other manufacturers (eg, Sanofi), they underscore that patient characteristics and product mix should be accounted for by payers when comparing the overall use and costs of insulin between different manufacturers. However, the results are limited to private insurers with frequent insulin use (ie, ≥100 LLY and NN insulin claims, ≥5% of which are attributable to LLY and NN) only and may not be generalizable to other payers (eg, Medicare and Medicaid, payers with infrequent insulin use).

The study explored the potential reasons for differences in quantities and costs per claim per day of comparable insulin products manufactured by LLY and NN after accounting for differences in baseline patient characteristics as well as product mix. However, other variables that may affect the use and costs of insulin from a payer perspective (eg, formulary and tier status) were not available in the database and therefore the effect of these variables on the study results is unknown. In addition, because the data lack clinical detail, the study relied on the accuracy of diagnosis codes recorded in the 6 months prior to the first insulin claim in 2011 to estimate the baseline profiles, which might have underestimated the actual prevalence of the comorbid conditions. Similarly, any inaccuracies in the data may have affected our estimates of antidiabetic medication use and costs in the study samples. Furthermore, because patients aged ≥65 years are dually eligible for Medicare, pharmacy claims may be incomplete. However, the impact of any inaccuracy in coding or incompleteness of the claims is believed to be the same for both LLY and NN groups, and therefore, the overall results are still reliable.

In conclusion, comparisons between the overall use and costs of comparable insulin products supplied by different manufacturers should account for differences in both the product mix offered by each manufacturer as well as characteristics of patients using these products.

Articles

Patient characteristics and product mix play an important role in determining the overall use and costs of comparable insulins across different manufacturers.