Health care resource utilization and expenditures associated with the use of insulin glargine

Clinical Therapeutics/Volume 29, Number 3, 2007

Health Care Resource Utilization and Expenditures Associated with the Use of Insulin Glargine Donald R. Miller, ScD1,2;John A. Gardner, PhD2,3; Ann M. Hendricks, PhD2,3; Quanwu Zhang, PhDa; and Benjamin G. Fincke, MD 1,2 1Center for Health Quality, Outcomes, and Economic Research, Veterans Affairs Medical Center, Bedford, Massachusetts; 2Department of Health Policyand Management, Boston University School of Public Health, Boston, Massachusetts; 3Health Care Financingand Economics, VeteransAffairs Boston Health Care System, Jamaica Plain, Massachusetts; and 4sanofi-aventis USA, Bridgewater, NewJersey ABSTRACT Background: Newer insulins, such as long-acting analogues, offer promise of better glycemic control, reduced risk for diabetes complications, and moderation of health care use and costs. Objective: We studied initiation of insulin glargine to evaluate its association with subsequent health service utilization and estimated expenditures. Methods: Patients of the Veterans Health Administration, US Department of Veterans Affairs (VA) who initiated insulin glargine (n = 5064) in 2001-2002 were compared with patients receiving other insulin (n = 69,944), matched on prescription month (index date). Inpatient and outpatient VA care in the 12 months after a patient's index date was evaluated using Tobit regression, controlling for prior utilization, demographic characteristics, comorbidities, glycosylated hemoglobin (HbAlc) levels, and diabetes severity. National average utilization costs and medication acquisition costs were used to estimate the value of VA expenditures. Results: Compared with other insulin users, insulin glargine initiators had higher HbA k values (8.72% vs 8.16%) prior to the index date, but greater subsequent HbA k reduction (-0.50% vs -0.22%). After adjustment for age, prior utilization, HbA k levels, and other factors, insulin glargine initiation was associated with 2.4 (95% CI, 1.1-3.7) fewer inpatient days for patients with any hospital admission (US $820 lower costs per initiator), 1.6 (1.2-1.9) more outpatient encounters ($279 higher costs per initiator), and $374 ($362-$387) higher costs for diabetes medications. The net difference was an average lower VA cost of $166 (-$290 to $622) per patient. Conclusions: Insulin glargine use was associated with decreased inpatient days but increased outpatient care, and the value of the net change in utilization to VA 478

offset the additional medication expenditures. Initiation of insulin glargine improves glycemic control and may reduce time in hospital without additional use of health resources. (Clin Ther. 2007;29:478-487) Copyright © 2007 Excerpta Medica, Inc. Key words: diabetes, insulin, delivery of health care, health care costs.

INTRODUCTION Exogenous insulin is an effective means for managing blood glucose in most patients with diabetes and for reducing the risks for diabetes complications) -6 Despite this, insulin is underutilized in part because it requires careful management and there are concerns over hypoglycemia and other tolerability issues. 6,7 New long-acting insulin analogues may remedy some of these concerns and offer the promise of more effective and tolerable glycemic control. One such agent is insulin glargine, a long-acting, recombinant human insulin analogue that is given once daily, provides 24-hour basal insulin coverage with no pronounced activity peak, and has an activity pattern resembling that of endogenous insulin. 8,9 The lower and more stable glucose levels associated with insulin glargine Study findings were presented in poster form at the 65th Annual Meeting of the American Diabetes Association, San Diego, California, June 10-14, 2005; and the Centers for Disease Control and Prevention Diabetes and Obesity Conference, Denver, Colorado, May 16-19, 2006.

Accepted for publicationJanuary 30, 200Z Express Track online publication March 9, 2007. doi:l 0.1016/].clinthera.2007.03.003 0149-2918/$32.00 Printed in the USA. Reproduction in whole or part: is not: permitted.

Copyright © 2007 ExcerptaMedica, Inc.

Volume 29 Number 3

D.R. Miller et al.

use may be expected to result in lower risk for diabetes complications and better diabetes-related outcomes. These benefits are suggested by clinical trials that have found efficacy in glycemic control and reduced incidence of hypoglycemia. 1°-1s To further inform prescribing practices, health care providers and payers also need evaluations of medications in less controlled clinical settings with the more heterogeneous population of patients actually using the medication. 16 If these benefits are realized in actual practice, patients using insulin glargine would be expected to have improved glycemic control and its use would be expected to be associated with a lower incidence and milder severity of diabetes-related morbidity, less health service utilization, and lower health care costs. We examined this issue in an observational study in patients with diabetes receiving care from the Veterans Health Administration, US Department of Veterans Affairs (VA). Parameters in patients initiating treatment with insulin glargine were compared with those in patients continuing to use other insulin. Glycemic control and health care resource utilization in the year following initiation were compared in the 2 study groups, controlling for relevant differences in the 2 populations at the time of initiation. With this design, we assessed whether insulin glargine initiation was associated with improved glycemic control and lower utilization, and we estimated the dollar values of utilization differences from the provider's perspective. We also assessed whether differences in utilization associated with insulin glargine use would at least offset the higher cost of the diabetes medication. MATERIALS A N D METHODS

This research was conducted in the VA using nationally standardized electronic databases available for care of all VA patients. 17-19 These include: (1) VA inpatient and outpatient medical encounters obtained from the Austin Automation Center; (2) VA outpatient prescription records and shelf costs from the Pharmacy Benefits Management Strategic Healthcare Group; (3) laboratory test data from the Decision Support System; and (4) death records from the Beneficiary Identification and Record Locator file (a registry of all veterans whose families applied for VA death benefits2°), supplemented by Social Security records. For this study, we used patient data from these sources that had been linked and processed for longitudinal analysis as part of the Diabetes Epidemiology Cohorts, a registry of all March 2007

VA patients with diabetes since 1998. 21 This study was approved by the institutional review board at the VA Medical Center, Bedford, Massachusetts. Study Sample

We identified all patients prescribed insulin glargine in the VA from the time of its first use in April 2001 through September 2002, finding 5689 patients. We limited the sample to the 5317 patients whose insulin glargine use followed VA prescriptions for other insulin, as directed by VA guidelines. 22 We further restricted the sample to the 5064 veteran patients who survived at least 12 months after insulin glargine initiation and were still active users of the VA system, as indicated by an outpatient visit or inpatient stay at a VA facility. The study was designed to test the effects of initiating insulin glargine according to VA guidelines. Accordingly, the comparison group was drawn from other patients with diabetes who were not prescribed insulin glargine as of September 2003, but who were prescribed other insulin. Each insulin glargine initiator was matched to other patients randomly chosen from the pool of veteran patients who had received a prescription for insulin in the same month and who survived and were active users of the system over the subsequent year. We used a matching ratio of 14:1 (except for the last 2 months, when it was 13:1), and the final comparison group included 69,944 patients. An index date was assigned to each patient to define a period of characterization (the 12 months preceding the index date) and a period of observation (the 12 months following the index date). For insulin glargine initiators, this was the date of the first prescription for insulin glargine. For the comparison group, it was the date of the other insulin prescription in the month in which they were matched to an insulin glargine initiator. This resulted in similar distributions of index dates in the 2 study groups. Patient Measures

Glycosylated hemoglobin (HbAlc) test results from the years before and after the index date were summarized in various ways. For the primary analysis, the last HbAlc test result up to 180 days before the index date was compared with the first HbAI~ test result that was between 28 and 180 days after the index date (allowing some time for the new treatment to have an effect). Episodes of hypoglycemia during the charac479

Clinical Therapeutics

terization and observation years were identified in the patient record by the presence of International Classi-

fication of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes (251). 23,24 Prescription records from the 2 years were used to summarize use of diabetes medications and to monitor for continued use of glargine and other insulin. Insulin formulations were classified as basal only (neutral protamine Hagedorn [NPH], lente, or ultralente), prandial only (regular, semilente, aspart, or lispro), premixed combinations of the 2, or combinations of the 2 with multiple prescriptions. Oral diabetes medication use was summarized by the number of classes of medications prescribed (sulfonylureas, biguanides, thiazolidinediones, 0~-glucosidase inhibitors, meglitinides). Acute health service utilization in the VA was assessed for the 12-month periods before and after the index date. Level of outpatient care was expressed as the number of days of outpatient visits to a VA facility when the visit involved face-to-face meetings with a health care provider (eg, visits only to pick up medication were not included). Inpatient care in the year was quantified as the number of hospital discharges and the total number of days of stay at a VA hospital. Costs of differences in utilization were estimated using national aggregated encounter-level costs for utilization calculated by the VA Health Economics Resource Center. 25,26 For outpatient care, costs were based on the Medicare relative value scale for each Current Procedural Terminology (CPT)-4 code, scaled to the VA outpatient care budget allocation. Inpatient care costs came from a cost regression model for nonVA discharges using diagnosis-related group, intensive care unit days, and length of stay, scaled to the VA inpatient care budget allocation. To account for differences between patients prior to their assignment to the 2 study groups, additional measures were taken from the records preceding the index date. Demographic factors (age, sex, race) and eligibility for VA care came from entries of several years of medical records processed to improve accuracy and reduce the number of missing values. Medicare enrollment was identified in linked Medicare records from 2000. Prevalent diabetes complications and other comorbidities were identified using diagnosis and CPT codes 23,24 from medical records for the 24-month period preceding the index date. Code combinations used for disease identification have been assessed in previous studies, and their screening properties are 480

known. 21,27-3° As a measure of illness burden, comorbidities other than the diabetes complications were summarized using the Comorbidity Index. 31

Statistical Analysis In describing the 2 study groups, unadjusted comparisons are presented except for prevalent disease rates, which were standardized to the age and sex distribution of the combined population of the 2 groups using the direct method. 32 HbAlc test results in the observation period were compared using ordinary leastsquares means regression to adjust for demographic characteristics, prevalent comorbidities, insulin formulation and oral diabetes medication prescriptions, VA and Medicare eligibility, VA health service utilization in the prior year, and HbA k test results in the prior year. These analyses were conducted using SAS version 8.2 (SAS Institute Inc., Cary, North Carolina). Group differences in VA utilization during the observation period were analyzed using a Tobit or censored regression model. 33 This 2-step model takes into account the problem of censoring at zero (common because most patients have no hospitalization), providing valid estimates of the differences in number of days of care among those with any care while accounting for the likelihood of receiving any care. All Tobit models included terms for preinitiation health status, personal characteristics and preinitiation VA utilization (same terms as in the HbAI~ models listed previously). The parameters of the model were estimated by maximum likelihood 34 using LIMDEP version 8.0 (Econometric Software, Inc., Plainview, New York). As an alternative method for adjusting differences between groups, we also conducted a matched propensity score analysis. 35 Propensity scores, computed as the predicted probability of being prescribed glargine, were obtained from a probit analysis that included all of the other independent variables from the Tobit models discussed previously. Based on propensity scores, patients from the 2 insulin groups were assigned to 1 of 190 categories, with the range of predicted probability spanning -0.2 percentage point for most categories. Using a bootstrapping technique, a series of 30 regression analyses were performed with samples consisting of patients prescribed glargine and 1-to-1 matched random samples of patients continuing to use other insulin. The difference between values of the utilization variables were calculated for all patients in each of the 30 samples, and averaged over the samples. Volume 29 Number 3

D.R. M i l l e r et al.

Aggregate cost implications of observed differences in utilization associated with insulin glargine use were calculated to illustrate how a provider could use the utilization differences to estimate the impact on its costs. The utilization difference for those patients with a given service obtained from the Tobit models was multiplied by the VA service-specific cost estimate, and the product was multiplied by the proportion of patients with the service. 95% CIs were calculated by applying the cost estimates to the SEs for the utilization differences. For inpatient care, cost differences were not assigned to differences in discharges because differences in days comprise the contributions of both discharges and length of stay. Medication cost differences between the study groups were calculated as the total VA acquisition (or lowest publicly available) costs 36 of all diabetes medications dispensed during the 12-month observation period. Net cost difference and CIs were estimated by summing individual cost differences and variances (inpatient, outpatient, and medications). Although this calculation assumes that the utilization and medication cost differences are uncorrelated, variation from this assumption is likely to have little impact on the estimates. RES U LTS

There were differences between the insulin glargine initiators and the comparison group of other insulin users (Table I). Most patients in both groups were male, reflecting the veteran patient population, but those who initiated insulin glargine were younger (mean age, 59 vs 65 years), less likely to have been enrolled in Medicare, and more likely to have been eligible for VA care due to disability related to military service. The 2 groups were statistically similar with regard to VA utilization in the prior year, with insulin glargine initiators having slightly higher proportions with any hospitalization and slightly larger numbers of inpatient days and outpatient visits. More insulin glargine initiators had indicators of severe, difficult-to-manage diabetes. Many more of them had been treated with both basal and prandial insulin (64.9% vs 38.8%). Although fewer took any oral diabetes medication in addition to their insulin, when they did take it, their prescriptions tended to include more classes of oral agents (17.6% vs 12.1% received >2 classes). Insulin glargine initiators had higher HbAI~ levels in the test preceding the index date (8.72% vs 8.16%), and many more of them had epiMarch 2007

sodes of hypoglycemia (11.5 % vs 4.4%) and other diabetes complications (7.4% vs 3.7%) in the prior year. Insulin glargine initiators also had more prevalent lower-leg ulcers, diabetic eye disease, and renal disease. At the index date, most of the patients initiating insulin glargine were prescribed a prandial insulin as well (85%), and this proportion remained >80% over the course of the subsequent year. Fifty-four percent of the glargine plus prandial insulin users were receiving rapid insulin (lispro or aspart) compared with 6% of the NPH plus prandial insulin users. Among insulin glargine initiators, 94% were still taking insulin and 81% were still prescribed insulin glargine by the end of the year. This compares with year-end insulin use of 91% and 71% persistence on the initial formulation in the group prescribed other insulin. HbAlc levels declined rapidly in patients prescribed insulin glargine and remained lower throughout the observation year. The average decline was greater than that observed in the patients using other insulin (-0.50% vs -0.22%), and the difference remained statistically significant (P < 0.001) after adjusting for other factors. There was some variation in response according to whether prandial insulin was also prescribed. In those prescribed insulin glargine alone, HbAlc decreased by 0.80%, from 8.73% to 7.93%, compared with a decrease of 0.23%, from 8.10% to 7.87%, in patients receiving NPH only (P < 0.001). In those prescribed insulin glargine with prandial insulin, HbAI~ decreased by 0.50%, from 8.72% to 8.22%, compared with a decrease of 0.22%, from 8.18% to 7.96%, in patients receiving NPH plus a prandial insulin (P < 0.01). HbAI~ patterns were nearly identical between premix and combined prescriptions for NPH and prandial insulin. ICD-9-CM codes for hypoglycemia were found in the medical records during the observation period more often in insulin glargine initiators than in continuing users of other insulin, even after adjustment for other factors (8.3% vs 3.5%). Compared with the prior year, however, this represents a reduction in hypoglycemia rates in both groups, with a proportionally greater decline in the insulin glargine initiators (-3.2% vs -0.9%; P < 0.001). Differences between study groups in VA health service utilization during the observation period are presented in Table II. Insulin glargine initiators had relatively more outpatient visits, and this remained 481

Clinical Therapeutics

Table I. Characteristics of patients initiating insulin glargine or continuing other insulin.

Characteristic Sex, no. (%) Male Female

Insulin Glargine (n = 5064)

Other Insulin (n = 69,944)

4847 (95.7) 217 (4.3)

68,489 (97.9) 1455 (2.1)

59 (13.0)

65 (11.6)

Age Mean (SD), y Age group, no. (%) <45 y 45-<55 y 55-<65 y 65-<75 y 75-<85 y _>85 y

678 1294 1326 1155 591 20

(13.4) (25.6) (26.2) (22.8) (11.7) (0.4)

2946 11,887 16,277 22,728 15,236 870

(4.2) (17.0) (23.3) (32.5) (21.8) (1.2)

Race, no. (%) White Black Hispanic Native American Asian Unknown

3602 540 197 21 16 688

(82.3) (12.3) (4.5) (0.5) (0.4) (13.6)

47,669 12,829 4946 218 502 3780

(72.0) (19.4) (7.5) (0.3) (0.8) (5.4)

VA eligibility, no. (%) Service-related disability Low income Special categories

2243 (44.3) 2395 (47.3) 425 (8.4)

22,732 (32.5) 39,658 (56.7) 7554 (10.8)

Medicare enrolled, no. (%)1

2259 (44.6)

47,282 (67.6)

VA use in past year2 Any hospitalization, no. (%) Hospital discharges, mean (SD) Inpatient days, mean (SD) 3 Outpatient visits, mean (SD) Comorbidity index score, mean (SD) 4

942 1.8 14.8 17.3

(18.6) (0.05) (0.9) (0.2)

2.2 (1.9)

12,450 1.7 14.1 17.0

(17.8) (0.01) (0.2) (0.1)

2.3 (2.0)

Insulin in past year, no. (%) Basal only Prandial only Basal and prandial

1293 (25.5) 486 (9.6) 3285 (64.9)

35,907 (51.3) 6899 (9.9) 27,138 (38.8)

Oral medication s in past, no. (%) None 1 Class 2 Classes _>3 Classes

3056 1118 148 33

38,575 22,861 2632 108

HbAlc tests in past year, mean (SD) Last HbAlc level, % Days since HbAI~ Mean HbAI~ level in past year, %

8.72 (2.13) 78 (121) 8.67 (1.85)

(60.4) (22.1) (12.6) (5.0)

(55.2) (32.7) (10.8) (1.3)

8.16 (1.97) 149 (1 66) 8.09 (1.78) (continued)

482

Volume 29 Number 3

D.R. M i l l e r et al.

Table I. (Continued)

Characteristic Diabetes morbidity, no. (adj. %)6 Hypertension I-lyperlipidemia Diabetic eye disease Peripheral vascular disease Other heart disease7 Lower limb ulcer/gangrene Renal disease Hypoglycemia Other infectionss Myocardial infarction Cerebrovascu lar disease 9 Other complications 1° Cellulitis and abscess Mental disorders, no. (%)11 Alcohol abuse, no. (%)

Insulin Glargine (n = 5064)

3614 3014 2425 1778 1593 989 828 578 511

507 436 405 221

(75.4) (61.5) (49.5) (38.7) (35.6) (21.5) (18.0) (11.5) (9.6) (10.6) (10.7) (7.4) (4.5)

Other Insulin (n = 69,944)

59,243 43,792 30,161 26,777 27,997 13,257 10,334 3064 8344

8528 8531 2578 3253

(84.4) (62.5) (43.0) (38.0) (38.3) (18.8) (14.7) (4.4) (11.9) (10.8) (12.0) (3.7) (4.6)

1432 (23.7)

19,449 (28.1)

277 (3.1)

4244 (6.2)

VA = Veterans Affairs; HbAlc = glycosylated hemoglobin. 1In 2000. 2Limited to non-long-term care. 3Among those with any non-long-term inpatient care. 4Count of other comorbiclities (see text). Slncluded sulfonylureas, biguanides, thiazolidinediones, 0~-glucosidase inhibitors, and meglitinides. 6Standardized to combined population sex and age distribution. 7Included ischemic heart disease, congestive heart failure, coronary artery bypass, percutaneous transangiography, and angina pectoris. 8Included pneumonia, septicemia, pyelonephritis, necrotizing facilities, renal abcess, bacteremia, endocarditis, osteomyelitis, and emphysematous cholecystitis. 9Included stroke and transient ischemic attacks. 1°Included diabetic ketoacidosis, hyperosmolar coma, and other diabetic coma. 11Included major depressive disorder, anxiety disorders, bipolar disorder, posttraumatic stress disorder, and schizophrenia.

after adjustment for potential confounding factors in the Tobit model (+1.6 [95% CI, +1.2 to +1.9]). Unadjusted numbers of hospital discharges and days were roughly comparable in the 2 groups, but insulin glargine initiation was associated with significantly fewer hospital discharges (-0.14 [-0.30 to -0.02]) and inpatient days (-2.4 [-3.7 to -1.1]) after adjustment for confounding. Most of the confounding effects controlled by the model (Table III) are related to age (which is lower in insulin glargine initiators and inversely related to health service utilization), and to prior utilization and HbAlc levels (both of which are higher in insulin glargine initiators and directly related to health service utilization). Similar results were March 2 0 0 7

found for diabetes-related care specifically, and when the comparison was made of specific formulations for insulin glargine versus NPH with or without prandial insulin (Table II). The adjusted group differences in inpatient days also persisted after the number of outpatient visits following initiation was included in the models (data not shown). The results from the matched sample propensity analysis were quite similar. For example, among patients with inpatient stays, there were fewer inpatient days for insulin glargine users in all of the 30 bootstrap samples, ranging from -4.18 to -0.51 days, and the mean of the average differences was -2.3 fewer days (-2.6 to -1.7). 483

Clinical Therapeutics

Table II. Differences in Veterans Affairs utilization and costs in observation year (N = 75,008).

Parameter

Hospital Discharges

Inpatient Days

Outpatient Visits

1.8 (0.04) 1.7 (0.01)

14.7 (0.7) 14.2 (0.2)

19.3 (0.2) 17.5 (0.3)

Unadjusted counts, mean (SD)* Insulin glargine Other insulin Adjusted group differences, ~ (95% CI)f Glargine vs other insulin Glargine vs NPH (basal only) Glargine vs NPH (basal + prandial) Premixvs NPH + prandial insulint

-0.14 -0.20 -0.12 +0.05

(-0.30 (-0.50 (-0.18 (-0.02

Cost estimatest Cost per day, mean, US $ Cost per initiator with care, mean (95% CI), US $

No. (%) of initiators with care Cost per initiator, mean (95% CI), US $

to -0.02) to -0.10) to -0.06) to +0.12)

-2.4 (-3.7 to -1.1) -3.5 (-7.1 to -0.1 ) -1.4 (-2.7 to -0.1)

+0.3 (-0..5 to +0.9)

+1.6 +1.3 +1.2 -0.08

(+1.2 to +1.9) (+0.9 to +1.7) (+1.0 to +1.4) (-0.17 to +0.01 )

$1856

$176

-$4454 (-$6867 to -$2042) 932/5064 (18.4) -$820 (-$1264 to -$376)

+$282 (+$211 to +$334) 5018/5064 (99.1) $279 (+$209 to +$331)

NPH = neutral protein Hagedorn. *Non-long-term care. fTobit model adjusts for demographics, prevalent comorbidities, insulin and oral diabetes medications, Veterans Affairs (VA) and Medicare eligibility, use of VA health services in prior year, and glycosylated hemoglobin test results in prior year. t Premix = preformulated NPH + prandial insulin.

When we assigned costs to the observed differences in utilization, we estimated the average reduction in costs per patients associated with the 2.4 fewer inpatient days in the observation year to be $820/y. This is offset partly by an estimated increase of $279/y associated with the 1.6 more outpatient visits in the year. We also considered the difference in diabetes medication costs during the observation period, which we estimated to be $374 (+$362 to +$387) higher in the insulin glargine initiators. Taken together, we estimate an average lower VA cost of $166 (-$290 to $622) per patient per year associated with insulin glargine initiation, which was not statistically different from zero. DISCUSSION

In general, new pharmaceutical agents are given initially to the subset of patients who are most likely to benefit from them to offset their higher costs. For insulin glargine, this is reflected by the VA guidelines that limit its use to "patients unable to achieve glycemic control targets because of recurrent episodes of symptomatic hypoglycemia, ...despite multiple at484

tempts with various insulin dosing regimens. ''22 The implementation of this guideline was evident in our study in that, compared with other patients using insulin, VA initiators of insulin glargine had more severe disease with higher HbAlc levels, more hypoglycemic reactions, and more diabetes complications despite intensive management with multiple insulins and multiple oral agents. In this group of difficult-to-treat diabetes patients, we found that use of insulin glargine resulted in improvements in HbAlc levels, improved patient persistence with insulin treatment, and a reduction in hypoglycemic reactions. In terms of subsequent utilization, there were increases in outpatient visits and medication costs but decreases in hospitalizations and number of days in the hospital. Cost reductions from inpatient care offset the increased outpatient and medication costs so that VA health care in patients initiating insulin glargine was approximately cost neutral. These findings are broadly consistent with those reported from 2 other studies of insulin glargine, 1 in a managed care system 37 and 1 in a population covered by Volume 29 Number 3

D.R. Miller e t al.

Table Ill. Tobit model building of utilization differences (N = 75,008). Parameter Inpatient days Group only Univariate adjustment + Age/sex + Diabetes comorbidities + Other comorbidities (CMI) + Prior utilization + Medicare enrolled + VA eligibility + Oral diabetes medication + HbAlc level + time lag to test Stepwise adjustment + Age/sex + prior utilization + Age/sex + prior utilization + HbAlc Full model O u t p a t i e n t visits Group only + HbAI~ level + time lag to test Full model

13"

Z2

P

-0.05

0

0.948

-1.35

-1.17 -0.02 -0.48 -0.08 -2.36

3.67 0.27 2.29 3.32 0.12 0.47 0.01

0.055 0.607 0.130 0.068 0.031 0.492 0.907

11.02

<0.001

-1.72 -2.80 -2.40

7.02 18.2,5 13.19

0.008 <0.001 <0.001

+2.24 +1.02 +1.56

63.83 13.47 69.46

<0.001

-0.35 +1.04

<0.001 <0.001

CMI = comorbidity index; VA = Veterans Affairs; HbAlc = glycosylated hemoglobin. *[[3 Coefficient For utilization (inpatient days or outpatient visits) From various models; 1 model per row.

Medicaid fee for service. 38 In both studies, insulin glargine use was associated with slightly lower net health care costs. The differential effect of insulin glargine on outpatient and inpatient care has potentially important implications for the interpretation of our findings. The VA guidelines require that a "prescriber must document improvement in either glucose control or hypoglycemia during the first 6 months of treatment. ''22 Therefore, it seems likely that the increase in outpatient visits is at least partly related to closer monitoring in the course of initiating insulin glargine. If so, then this could be a 1-time expense that is not expected to be recapitulated in subsequent years. In contrast, if the reduction in hospital days results from intrinsic advantages of insulin glargine, such as lower HbAlc levels and less extreme fluctuations in blood glucose levels, then the economic benefit may continue to accrue over the long term. Data limitations did not permit us to examine these possibilities, which should be explored in future studies. Our study has important limitations. First, our resuits only apply to the selected group of difficult-toMarch 2 0 0 7

control diabetic patients whom we studied, and we did not address the economic consequences of using insulin glargine more widely. This includes use that does not follow VA guidelines, such as initiation in patients without prior use of insulin, or use in patients who are near death, since the sample was restricted to those who survived the year after initiation. Second, the analytic models included adjustment for many potential differences between the study groups, including HbAlc levels and diabetes complications, but residual confounding by other, unmeasured factors remains a possibility. Third, our comparison of insulin glargine initiators with ongoing users of other insulin replicated the most common clinical scenario for starting the drug, but it could have introduced bias because it compares patients initiating a new therapy with those continuing on an existing one. If such bias does exist, it is unlikely to persist over the 12 months of followup. Fourth, we studied only VA health service utilization, and many patients could have received care from non-VA providers. It is reassuring that our models included terms for Medicare enrollment, which had only 485

Clinical Therapeutics

a slight effect on the estimates of utilization differences (Table III). Fifth, there are many assumptions and decisions underlying the VA cost data, and a different system of cost accounting might have yielded different results. For example, from a payer's perspective, there might be little or no actual savings from the decline in inpatient utilization if the insurer pays per discharge rather than per day. Sixth, we did not have information about which providers were giving care, and the observed effects may have been partly due to the provider rather than the insulin preparation itself. Also, the lower inpatient utilization of insulin glargine initiators could have been a consequence of the increased outpatient care. Tempering this concern is the fact that the glargine initiators were already more intensively managed at baseline, leaving less room for improvement due to practice style alone. Furthermore, including the number of outpatient visits in the model did not diminish the inpatient benefit. Finally, with only 1 year of follow-up, we had a limited view of the economic effects of insulin glargine, and we did not know whether the observed patterns would be sustained. CONCLUSIONS

Insulin glargine use was associated with decreased inpatient days but increased outpatient care, and the value of the net change in utilization to VA offset the additional medication expenditures. We conclude that initiation of insulin glargine improves glycemic control and may result in less need for health care resources to care for patients with diabetes. ACKNOWLEDGMENTS

This study was financially supported by the Department of Veterans Affairs, Boston University School of Public Health, and sanofi-aventis USA. We are grateful to Francesca Cunningham for assistance with data acquisition and technical support and to Qing Shao and Nien-Chen Li for assistance with data processing and statistical analysis. Dr. Miller has received research grant support from sanofi-aventis USA, GlaxoSmithKline, and Merck Pharmaceuticals. REFERENCES 1. UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of com-

486

plications in patients with type 2 diabetes (UKPDS 33) [published correction appears in Lancet. 1999;354:602]. Lancet. 1998;352:837-853. 2. The Diabetes Control and Complications Trial Research Group. The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. N EnglJ Med. 1993;329:977-986. 3. Abraira C, Colwell JA, Nuttall FQ, et al. Veterans Affairs Cooperative Study on Glycemic Control and Complications in Type 2 Diabetes (VA CSDM). Results of the feasibili W trial. Diabetes Care. 1995;18:1113-1123. 4. Ohkubo Y, Kishikawa H, Araki E, et al. Intensive insulin therapy prevents the progression of diabetic microvascular complications in Japanese patients with non-insulindependent diabetes mellitus: A randomized prospective

6-year study. Diabetes Res Clin Pract. 1995;28:103-11 7. 5. Abraira C, McGuire DK. Intensive insulin therapy in patients with Wpe 2 diabetes: Implications of the Veterans Affairs (VA CSDM) feasibility trial. Am Hearty. 1999;138: 5360-5365. 6. Genuth S. Exogenous insulin administration and cardiovascular risk in non-insulin-dependent and insulindependent diabetes mellitus. Ann Intern Med. 1996;124: 104-109. 7. Hayward IRA, Manning WG, Kaplan SH, et al. Starting insulin therapy in patients with type 2 diabetes: Effectiveness, complications, and resource utilization [published correction appears inJAMA. 1999;281:1989]. JAMA. 1997; 278:1663-1669. 8. Levien TL, Baker DE, White JR Jr, Campbell RK. Insulin

glargine: A new basal insulin. Ann Pharmacother. 2002;36: 1019-1027. 9. Gerich JE. Novel insulins: Expanding options in diabetes management. Amy Med. 2002;113:308-31 6. 10. Raskin P, Klaff L, Bergenstal R, et al. A 16-week comparison of the novel insulin analog insulin glargine (HOE 901 ) and NPH human insulin used with insulin lispro in patients with type 1 diabetes. Diabetes Care. 2000;23:16661671. 11. Yki-Jarvinen H, Dressier A, Ziemen M, The HOE 901/3002 Study Group. Less nocturnal hypoglycemia and better post-dinner glucose control with bedtime insulin glargine compared with bedtime NPH insulin during insulin combination therapy in type 2 diabetes. Diabetes Care. 2000;

23:1130-1136. 12. RosenstockJ, Park G, Zimmerman J, for the US Insulin Glargine (HOE 901 ) Type 1 Diabetes Investigator Group. Basal insulin glargine (HOE 901) versus NPH insulin in patientswith type 1 diabetes on multiple daily insulin regimens. Diabetes Care. 2000;23:1137-1142. 13. Riddle MC, RosenstockJ, GerichJ, for the Insulin Glargine 4002 Study Investigators. The treat-to-target trial: Randomized addition of glargine or human NPH insulin to

Volume 29 Number 3

D.R. Miller et al.

14.

15.

16.

17.

18.

19.

20.

21.

22.

oral therapy of type 2 diabetic patients. Diabetes Care. 2003;26:30803086. Janka HU, Plewe G, Riddle MC, et al. Comparison of basal insulin added to oral agents versus twicedaily premixed insulin as initial insulin therapy for type 2 diabetes. Diabetes Care. 2005;28:254-259. Rosenstock J, Dailey G, MassiBenedetti M, et al. Reduced hypoglycemia risk with insulin glargine: A meta-analysis comparing insulin glargine with human NPH insulin in type 2 diabetes. Diabetes Care. 2005; 28:950-955. Rosenheck R, Leslie D, Sernyak M. From clinical trials to real-world practice: Use of atypical antipsychotic medication nationally in the Department of Veterans Affairs. Med Care. 2001 ;39:302-308. goyko EJ, Koepsell TD, GazianoJM, et al. US Department of Veterans Affairs medical care system as a resource to epidemiologists. Am J Epidemio[. 2000;151:307-314. Cowper DC, Hynes DM, Kubal JD, Murphy PA. Using administrative databases for outcomes research: Select examples from VA Health Services Research and Development. J Med Syst. 1999;23:249-259. Smith MW, Joseph GJ. Pharmacy data in the VA health care system. Med Care Res Rev. 2003;60(Suppl 3): 925-1235. Fisher SG, Weber L, Goldberg J, Davis F. Mortality ascertainment in the veteran population: Alternatives to the National Death Index. A m J Epidemio[. 1995;141:242-250. Miller DR, Safford MM, Pogach LM. Who has diabetes? Best estimates of diabetes prevalence in the Department of Veterans Affairs based on computerized patient data. Diabetes Care. 2004;27(Suppl 2): gl 0-B21. VA Pharmacy Benefits Management Strategic Healthcare Group, Medical Advisory Panel, Deborah Khachikian. Criteria for non-formulary use of insulin glargine (Lantus). Available

March 2007

23.

24.

25.

26.

27.

28.

29.

at: http :l lwww. vapbm, org/ criteria/ insulinglarginecriteria, pd f. Accessed January 2002. World Health International Classification of Diseases, Ninth Revision, Clinical Modification, 2003. Available at: ftp:// ftp.cd c.gov/p u b/H ealth_Statistics/ NCHS/Publications/ICD9-CM/ 2003/. Accessed July 10, 2004. American MedicalAssociation (AMA). Current Procedural Terminology." 2001. Chicago, II1:AMA; 2000. HERC's Inpatient Average Cost Datasets for VA Care. Version4: FixedYears 19982002. Menlo Park, Calif." Health Economics Resource Center; July 2003. Available at: http://www. herc.research.med.va.gov. Accessed October 2004. HERC's Outpatient Average Cost Datasets for VA Care. Fixed Years 1999-2000. Menlo Park, Calif." Health Economics Resource Center; August 2003. Available at: http://www.herc.research.med. va.gov. Accessed October 2004. Miller DR, Rogers WH, Kazis L, et al. Evaluation of disease status based on patient self-report in the Medicare Health Outcomes Survey using linked data from surveys and computerized medical data from the Veterans Health Administration. J Clin Epidemiol. Submitted. Borzecki AM, Wong AT, Hickey EC, et al. Identifying hypertension-related comorbidities from administrative data: What's the optimal approach? AmJ Med Qual. 2004;19:201-206. Petersen LA, Wright S, Normand SL, Daley J. Positive predictive value of the diagnosis of acute myocardial infarction in an administrative database..] Gen Intern Med. 1999;14:555558.

30. Kashner TM. Agreement between administrative files and written medical records: A case of the Department of Veterans Affairs. Med Care. 1998;36:1324-1336. 31. Selim A, Fincke G, Ren X, et al. The comorbidity index. In: Goldfield N,

32.

33.

34.

35.

36.

37.

38.

Pine M, Pine J, eds. Measuring and Managmg Healtk Care Quality. 2nd ed. New York, NY: Aspen Publishers; 2002:91-94. Rothman KJ. Modern Epidemiology. Boston, Mass: Little Brown and Co; 1986. Maddalla GS. Econometrics. New York, NY: McGraw-Hill; 1997:314317. Greene WH. EconometricAna~sis. 4th ed. Upper Saddle River, NJ: Prentice Hall; 2000. D'Agostino RB Jr. Propensity score methods for bias reduction in the comparison of a treatment to a nonrandomized control group. StatMed. 1998;17:2265-2281. Render ML, Nowak J, Hammond EK, Roselle G. Methods for estimating and comparing VA outpatient drug benefits with the private sector. Med Care. 2003;41 (Suppl 6):1161-1169. Bullano MF, AI-Zakwani IS, Fisher MD, et al. Differences in hypoglycemia event rates and associated costconsequence in patients initiated on long-acting and intermediate-acting insulin products. Curr Med Res Opin. 2005;21:291-298. Zhang Q, Menditto L. Incremental cost savings 6 months following initiation of insulin glargine in a Medicaid fee-for-service sample. AmJ Tker. 2005;12:337-343.

Address correspondence to: D o n a l d R. Miller, ScD, C e n t e r for H e a l t h Q u a l i t y , O u t c o m e s , a n d E c o n o m i c R e s e a r c h , VA M e d i c a l Center, 2 0 0 Springs R o a d (152), B e d f o r d , M A 0 1 7 3 0 . E - m a i l : d r m i l l e r @ b u . e d u

487