Currently Viewing:
The American Journal of Managed Care March 2015
Evaluation of Care Management Intensity and Bariatric Surgical Weight Loss
Sarit Polsky, MD, MPH; William T. Donahoo, MD; Ella E. Lyons, MS; Kristine L. Funk, MS, RD; Thomas E. Elliott, MD; Rebecca Williams, DrPh, MPH; David Arterburn, MD, MPH; Jennifer D. Portz, PhD, MSW; and Elizabeth Bayliss, MD, MSPH
Potential Savings From Increasing Adherence to Inhaled Corticosteroid Therapy in Medicaid-Enrolled Children
George Rust, MD, MPH, FAAFP, FACPM; Shun Zhang, MD, MPH; Luceta McRoy, PhD; and Maria Pisu, PhD
Innovation in Plain Sight
Karen Ignagni, MBA, President and Chief Executive Officer, America's Health Insurance Plans
Early Changes in VA Medical Home Components and Utilization
Jean Yoon, PhD, MHS; Chuan-Fen Liu, PhD, MPH; Jeanie Lo, MPH; Gordon Schectman, MD; Richard Stark, MD; Lisa V. Rubenstein, MD, MSPH; and Elizabeth M. Yano, PhD, MSPH
Are Healthcare Quality "Report Cards" Reaching Consumers? Awareness in the Chronically Ill Population
Dennis P. Scanlon, PhD; Yunfeng Shi, PhD; Neeraj Bhandari, MD; and Jon B. Christianson, PhD
Developing a Composite Weighted Quality Metric to Reflect the Total Benefit Conferred by a Health Plan
Glen B. Taksler, PhD; and R. Scott Braithwaite, MD, MSc, FACP
Insurance Impact on Nonurgent and Primary Care-Sensitive Emergency Department Use
Weiwei Chen, PhD; Teresa M. Waters, PhD; and Cyril F. Chang, PhD
Cost Differential by Site of Service for Cancer Patients Receiving Chemotherapy
Jad Hayes, MS, ASA, MAAA; J. Russell Hoverman, MD, PhD; Matthew E. Brow, BA; Dana C. Dilbeck, BA; Diana K. Verrilli, MS; Jody Garey, PharmD; Janet L. Espirito, PharmD; Jorge Cardona, BS; and Roy Beveridge, MD
The Combined Effect of the Electronic Health Record and Hospitalist Care on Length of Stay
Jinhyung Lee, PhD; Yong-Fang Kuo, PhD; Yu-Li Lin, MS; and James S. Goodwin, MD
Strategy for a Transparent, Accessible, and Sustainable National Claims Database
Robin Gelburd, JD, BA
Treatment Patterns, Healthcare Utilization, and Costs of Chronic Opioid Treatment for Non-Cancer Pain in the United States
David M. Kern, MS; Siting Zhou, PhD; Soheil Chavoshi, MS; Ozgur Tunceli, PhD; Mark Sostek, MD; Joseph Singer, MD; and Robert J. LoCasale, PhD
Trends in Mortality Following Hip Fracture in Older Women
Joan C. Lo, MD; Sowmya Srinivasan, MD; Malini Chandra, MS, MBA; Mary Patton, MD; Amer Budayr, MD; Lucy H. Liu, MD; Gene Lau, MD; and Christopher D. Grimsrud, MD, PhD
Currently Reading
Long-Term Outcomes of Analogue Insulin Compared With NPH for Patients With Type 2 Diabetes Mellitus
Julia C. Prentice, PhD; Paul R. Conlin, MD; Walid F. Gellad, MD, MPH; David Edelman, MD; Todd A. Lee, PharmD, PhD; and Steven D. Pizer, PhD

Long-Term Outcomes of Analogue Insulin Compared With NPH for Patients With Type 2 Diabetes Mellitus

Julia C. Prentice, PhD; Paul R. Conlin, MD; Walid F. Gellad, MD, MPH; David Edelman, MD; Todd A. Lee, PharmD, PhD; and Steven D. Pizer, PhD
There were no significant differences in the risk of ambulatory care—sensitive condition hospitalization or mortality between patients who initiated analogue insulin compared with the neutral protamine Hagedorn.

Long-acting insulin analogues (eg, insulin glargine and insulin detemir) are an alternative to neutral protamine Hage-dorn (NPH) insulin for maintaining glycemic control in patients with diabetes. Clinical trials comparing analogue insulin and NPH have neither been adequately powered nor had sufficient follow-up to examine long-term health outcomes.

Objectives: To compare the effects of NPH and long-acting insulin analogues on long-term outcomes.

Study Design: This retrospective observational study relied on administrative data from the Veterans Health Administration and Medicare from 2000 to 2010. Local variations in analogue insulin prescribing rates were used in instrumental variable models to control for confounding. Outcomes were assessed using survival models.

Methods: The study population included US veterans dually enrolled in Medicare who received at least 1 prescription for oral diabetes medication and then initiated long-acting insulin be-tween 2001 and 2009. Outcomes included ambulatory care–sensi-tive condition (ACSC) hospitalizations and mortality.

Results: There was no significant relationship between type of insulin and ACSC hospitalization or mortality. The hazard ratio for mortality of individuals starting a long-acting analogue insulin was 0.97 (95% CI, 0.85-1.11), and was 1.05 (95% CI, 0.95-1.16) for ACSC hospitalization. Differences in risk remained insignificant when predicting diabetes-specific ACSC hospitalizations, but starting on long-acting analogue insulin significantly increased the risk of a cardiovascular-specific ACSC hospitalization.

Conclusions: We found no consistent difference in long-term health outcomes when comparing use of long-acting insulin analogues and NPH insulin. The higher cost of analogue insulin without demonstrable clinical benefit raises questions of its cost-effectiveness in the treatment of patients with diabetes.

Am J Manag Care. 2015;21(3):e235-e243

The progressive nature of type 2 diabetes mellitus requires many patients to initi-ate insulin. This research compares the effects of the neutral protamine Hagedorn (NPH) and analogue insulin on long-term outcomes. n Analogue insulin is significantly more expensive than NPH.

  • Clinical trials comparing analogue insulin with NPH have not been adequately powered nor had sufficient follow-up to examine long-term health outcomes.
  • There were no significant differences in the risk of ambulatory care–sensitive condition hospitalization or mortality for patients who initiated analogue insulin compared with NPH.
  • The higher cost of analogue insulin without demonstrable clinical benefit raises questions of its cost-effectiveness in the treatment of patients with diabetes.

The progressive nature of type 2 diabetes mellitus requires many patients to use insulin to maintain glycemic control.1,2 Neutral protamine Hagedorn (NPH) insulin was the most commonly used intermediate- to long-acting insulin until the introduction of the long-act-ing insulin analogues: insulin glargine in 20002,3 and insulin detemir in 2005.4 The long-acting insulin analogues were designed with properties that prolonged absorption, and with a flattened activity peak that extended duration of ef-fect and also resulted in a reduced risk of hypoglycemia.5-7

The significantly higher cost of analogue insulin compared with NPH8,9 led researchers and policy makers to compare the efficacy and cost of these medications. Several reviews found no significant difference between analogue insulin and NPH on glycemic control or severe hypoglycemia (ie, low glucose level requiring assistance from another person) but did show reduced likelihood of nocturnal hypoglycemia for patients using analogue insulin.6,7,10,11

Despite these short-term differences, potential long-term benefits have been difficult to test due to the short time frame of studies comparing analogue insulin with NPH.2,7 Some cost-effectiveness studies were based on modeling that used clinical trial results as predictors of long-term treatment ef-fects (eg, Center for Outcomes Research Diabetes Model).1 Other cost-effectiveness studies used retrospective claims that may better reflect clinical settings outside of trials. Results from these studies are mixed, with some concluding analogue insulin was cost-effective1,5 and others concluding analogue insulin was not an efficient use of healthcare resources.12,13 However, an important limitation of these claims-based stud-ies was the assumption that no systematic, unobserved differ-ences existed between the groups being compared (eg, those that started on NPH compared with analogue insulin).

To bridge this gap in the evidence, we used national Vet-erans Health Administration (VA) records to systematically compare long-term outcomes for patients using NPH or ana-logue insulin. We used a novel approach to an established observational comparative effectiveness method, instru-mental variables, to obtain unbiased estimates despite the possibility of unobserved differences between groups.


Data Sources

Patient-level national data from the VA were used and supplemented with data from Medicare to ensure com-pleteness in the measures of outcomes, as VA patients often use non-VA facilities for hospital care.14 The study was reviewed and approved by the Institutional Review Board at the VA Boston Healthcare System.

Study Population

All prescription claims for metformin, sulfonylurea, thiazolidinedione, and long-acting insulin (ie, NPH, insu-lin glargine and insulin detemir, and mixtures of short- and long-acting insulins) between 2000 and 2007 were extracted from VA pharmacy files (Figure). Eligible indi-viduals had a history of being prescribed at least 1 oral diabetes medication, and initiated long-acting insulin between February 1, 2001, and December 31, 2009. The initiation date for the long-acting insulin was the start of the study period (“index date”) for each patient and the prior 12 months was the baseline period. To ensure that all hospitalizations were recorded, we further limited the cohort to include only those enrolled in Medicare as well as the VA. The last index date permitted was at the end of 2009, to allow a minimum 12-month outcome period. This resulted in a cohort of 142,940 patients, including 118,878 who initiated NPH and 24,062 who initiated ana-logue insulin.

Insulin Treatment

The main objective of the study was to compare the long-term effects of NPH and analogue insulin. Patients who initiated NPH or an NPH/regular in-sulin mixture were compared with patients who started glargine, detemir, or short-act-ing and long-acting mixtures that included analogue insulin. To improve comparabil-ity among patients, we attempted to isolate patients at a similar point in the progres-sion of diabetes. Consequently, patients were required to have a prescription for an oral diabetes medication during the base-line period before initiating insulin. Eighty percent of the individuals remained on the same insulin they started (NPH or analogue insulin). Among those who started on analogue insulin, 99% re-ceived glargine.

Provider Practice Pattern as Instrumental Variable

The principal threat to a simple comparison of out-comes is that the selection of treatment by patients and providers could be influenced by unmeasured differences in patient risk (selection bias or confounding by indica-tion). To address this, we developed instrumental vari-ables models, which identify a factor (the instrumental variable [IV]) that influences treatment but is effectively random with respect to patient risk and other potential confounders.15,16 The statistical model isolates the com-ponent of treatment variation attributable to the IV and measures the relationship between this component and outcomes.17 The success of the approach depends on the effective randomness of the IV (controlling for all the other variables in the model) as well as the strength of the IV’s influence on treatment status.16,18

VA patients are assigned to their primary care physi-cian by variable and often arbitrary methods, so in this study, the IV was the proportion of long-acting insulin prescriptions written for analogue insulin by each pro-vider during an individual’s baseline period19—the 12 months before the index date on which analogue insulin was prescribed. The instrument assigned to each individ-ual was the proportion of analogue insulin prescriptions written by the provider who prescribed the initial insulin prescription on the index date.

For example, if an individual started on long-acting in-sulin on January 1, 2003, their baseline period was January 1, 2002, to December 31, 2002 (see eAppendix Figure A1). We identified the provider that prescribed insulin on Jan-uary 1, 2003, and calculated their proportion of analogue insulin prescriptions in that baseline period (January to December 2002). An individual’s insulin prescription does not contribute to their instrumental variable calcu-lation, eliminating concerns about selection bias. Provid-ers and patients were aligned based on the index date to minimize confounding that could occur if patients later switched providers.

If a provider prescribed insulin to less than 10 unique patients during the baseline period (53% of the time), the rate at the community-based outpatient clinic (CBOC) or VA medical center (VAMC) at which the provider prac-ticed was used.

Provider Quality Controls

The IV will not be effective if correlated with other provider characteristics that might also affect outcomes, causing biased estimates of the treatment-outcome rela-tionship. Therefore, we included 3 provider-level process quality variables: percent of glycated hemoglobin (A1C) labs ≥9%,20,21 percent of blood pressure readings ≥140/90 mm Hg,22 and percent of low-density lipoprotein choles-terol labs >100 mg/dL.22 These variables were computed at the same provider, CBOC, or VAMC level, and time periods as the IV prescribing rate.


Additional control variables computed at baseline included patient age, sex, race, A1C, serum creatinine, urine microalbumin, body mass index (see Table 1 for cat-egorizations), 29 indicator variables for comorbidities,23 8 indicator variables for the components of the Young dia-betes severity index,24 and indicator variables for calen-dar years corresponding to index dates. Individuals can be diagnosed with multiple comorbidities (eg, obesity and congestive heart failure [CHF]).


Outcomes included mortality and hospital admission (VA or Medicare) for any of 13 ambulatory care–sensitive conditions (ACSCs) as defined by the Agency for Health-care Research and Quality.25,26 These hospitalizations are hypothesized to be preventable with high-quality outpa-tient care and include several diabetes and cardiovascular complications such as uncontrolled diabetes, short- and long-term complications of diabetes, CHF, and chronic obstructive pulmonary disease. Sensitivity analyses es-timated models for specific ACSC hospitalization types most closely related to diabetes and cardiovascular dis-ease. The VA Vital Status File, which determines the date of death from VA, Medicare, and Social Security Admin-istration data, was used to determine mortality.27 Patients were censored either at the end of 2010, or when they ex-perienced either outcome (mortality or ACSC hospitaliza-tion), or the date they first switched between long-acting insulin types (NPH and analogue insulin). Consequently, the modeled outcome was the amount of time between the index date and each individual’s censoring date.

Statistical Models

We used Stata version 10 (StataCorp LP, College Sta-tion, Texas) to estimate the effects of analogue insulin use on the risks of outcomes using Cox proportional hazards models. This model was chosen for 2 reasons: first, a time-to-event analysis has more statistical power than a logistic regression, because more information is used. In a logistic regression, estimates are based on dif-ferences between individuals who had events and those who did not. In a time-to-event model, estimates are based on differences in time to event as well. The second reason to use a Cox model was to mimic a clinical trial study design. In this study, individuals started analogue or NPH insulin and we then pre-dicted time until death or an ACSC hospitaliza-tion, using the provider’s prescribing pattern as an instrumental variable to control for selection bias. This is analogous to individuals starting on the treatment or placebo in a randomized clinical trial.

Copyright AJMC 2006-2018 Clinical Care Targeted Communications Group, LLC. All Rights Reserved.
Welcome the the new and improved, the premier managed market network. Tell us about yourself so that we can serve you better.
Sign Up