Currently Viewing:
The American Journal of Managed Care November 2004 - Part 2
Screening for Depression and Suicidality in a VA Primary Care Setting: 2 Items Are Better Than 1 Item
Kathryn Corson, PhD; Martha S. Gerrity, MD, MPH, PhD; and Steven K. Dobscha, MD
The Veterans Health Administration: Quality, Value, Accountability, and Information as Transforming Strategies for Patient-Centered Care
Jonathan B. Perlin, MD, PhD, MSHA; Robert M. Kolodner, MD; and Robert H. Roswell, MD
VA Health Services Research: Lessons for the World's Healthcare Organizations
Steven J. Bernstein, MD, MPH
Variation in Implementation and Use of Computerized Clinical Reminders in an Integrated Healthcare System
Constance H. Fung, MD, MSHS; Juliet N. Woods, MS; Steven M. Asch, MD, MPH; Peter Glassman, MBBS, MSc; and Bradley N. Doebbeling, MD, MSc
Currently Reading
Dual-system Utilization Affects Regional Variation in Prevention Quality Indicators: The Case of Amputations Among Veterans With Diabetes
Chin-Lin Tseng, DrPH; Jeffrey D. Greenberg, MD, MPH; Drew Helmer, MD, MS; Mangala Rajan, MBA; Anjali Tiwari, MD; Donald Miller, ScD; Stephen Crystal, PhD; Gerald Hawley, RN, MSN; and Leonard Pogach, M
The Relationship of System-Level Quality Improvement With Quality of Depression Care
Andrea Charbonneau, MD, MSc; Victoria Parker, DBA; Mark Meterko, PhD; Amy K. Rosen, PhD; Boris Kader, PhD; Richard R. Owen, MD; Arlene S. Ash, PhD; Jeffrey Whittle, MD, MPH; and Dan R. Berlowitz, MD,
Designing an Illustrated Patient Satisfaction Instrument for Low-literacy Populations
Janet Weiner, MPH; Abigail Aguirre, MPA; Karima Ravenell, MS; Kim Kovath, VMD; Lindsay McDevit, MD; John Murphy, MD; David A. Asch, MD, MBA; and Judy A. Shea, PhD
Problems Due to Medication Costs Among VA and Non-VA Patients With Chronic Illnesses
John D. Piette, PhD; and Michele Heisler, MD, MPA

Dual-system Utilization Affects Regional Variation in Prevention Quality Indicators: The Case of Amputations Among Veterans With Diabetes

Chin-Lin Tseng, DrPH; Jeffrey D. Greenberg, MD, MPH; Drew Helmer, MD, MS; Mangala Rajan, MBA; Anjali Tiwari, MD; Donald Miller, ScD; Stephen Crystal, PhD; Gerald Hawley, RN, MSN; and Leonard Pogach, M

Objective: To determine the impact of dual-system utilization by veterans on regional variation in lower-extremity amputation rates.

Study Design: Retrospective longitudinal cohort analysis.

Patients and Methods: Subjects were veterans with diabetes who used Veterans Health Administration (VHA) care and were dually enrolled in Medicare fee for service in fiscal years (FY) 1997-1999. We evaluated the impact of Centers for Medicare and Medicaid Services (CMS) data on prevalence of baseline foot risk factors, medical comorbidities, and amputations in FY 1997-1998, and ranking of 22 regions using risk-adjusted major and minor amputation rates in FY 1999.

Results: The addition of CMS data significantly increased the prevalence of amputations and risk factors for the 218 528 dually eligible veterans (all P values <.001). In FY 1999, we identified 3.1 minor and 4.5 major amputations per 1000 patients (VHA data) versus 5.5 minor and 8.6 major amputations per 1000 patients (VHA/CMS data); the prevalence of any peripheral vascular condition in FY 1997-1998 was 5.7% (VHA) versus 13.0% (VHA/CMS). The impact of including CMS data varied across regions for amputation outcomes, ranging from an additional 34.3% to 150.7%. Using observed-to-expected amputation ratios and 99% confidence intervals, the addition of CMS data changed the outlier status for 8 of 22 regions for both major and minor amputations.

Conclusion: Risk covariates and amputation outcomes were substantially underestimated using VHA data only. Our findings demonstrate the importance of evaluating dual-system utilization when conducting program evaluations for healthcare systems with a substantial number of dual enrollees.

(Am J Manag Care. 2004;10(part 2):886-892)

There is an increasing emphasis by the private sector,1 accrediting organizations,2 and government agencies3 on measuring adherence to recommended process and interim outcomes to decrease hospitalizations and patient morbidity and mortality at the provider, plan, and population level. In the current healthcare environment, variation in health outcomes within payer and provider healthcare systems is likely to be used for internal benchmarking and accreditation, and may be subject to public reporting.4 Thus, it is critical to address issues such as data availability and risk adjustment to ensure the accuracy of the results and fairness in comparisons.

Recently, the Agency for Healthcare Research and Quality (AHRQ) included prevention quality indicators in its National Healthcare Quality Report Card.5 These indicators were chosen based on expert consensus that the provision of timely and effective outpatient care before hospitalization could prevent the hospitalization. Because the indicators were developed using hospitalization rates rather than individual patient-level data, they primarily are intended to assess community-level variation; however, they could be adapted by healthcare systems as well.

However, healthcare system accountability for preventable hospitalizations would be contingent on the ability of these systems both to control and evaluate the quality of ambulatory preventive care provided to their enrollees. Consequently, a significant methodological challenge for both federal and private healthcare systems is how to evaluate variation in health outcomes—that is, identification of best- and worst-performing regions or units—when beneficiary care may be fragmented among more than 1 healthcare system. This fragmentation of care is especially true within the Department of Veterans Affairs (VA), the nation's largest integrated healthcare system. Based on a large, comprehensive survey of veteran clinical users in 1999, approximately 62% of veterans self-reported utilization of non-VA care during the prior year.6 The largest dual-coverage category was Medicare, which was a source of care for 53% of all veteran enrollees reporting dual coverage.

To evaluate the impact of dual Medicare enrollment patterns on evaluation of regional variation in a prevention quality indicator among veterans, we chose lower-extremity amputations. In addition to being 1 of the 3 diabetes-specific AHRQ prevention quality indicators,5 amputations result in substantial morbidity, disability, and costs for persons with diabetes.7 Consequently, a reduction in amputation rates is the focus of multiple governmental programs.3,5,8

Within the Veterans Health Administration (VHA), the Preservation Amputation Care and Treatment Program8 mandates VA medical centers to develop multidisciplinary programs to identify veterans at risk for lower-extremity complications and to refer them to foot-care specialists for ongoing evaluation and care. However, prior epidemiologic studies and operational reports of amputation rates and trends have exclusively utilized VHA administrative datasets, thus not capturing relevant amputation data outside the VHA.9-11

Consequently, our specific objectives in this study were to determine and evaluate the impact of the inclusion of Medicare utilization data (ie, Centers for Medicare and Medicaid Services [CMS] data) on the prevalence of foot risk factors, medical comorbidities, and prior amputations during a 2-year baseline period; and to determine the impact of these data on the regional variation in minor and major amputation outcomes in the subsequent year. We defined the regions using the service areas of the Veterans Integrated Service Networks (VISNs).


Data Sources and Cohort Assembly

We utilized a cohort of veteran clinical users with diabetes who also were enrolled in Medicare in fiscal year (FY) 1997 or FY 1998 as previously described.12 Potential subjects were identified from several inpatient and outpatient claims-data sources by the presence of at least 1 diabetes-specific (250.xx, 357.2, 362.0, 366.41) International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code or prescription of a diabetes-specific medicine.

Sample Selection

A total of 446 896 veterans with diabetes were identified and were alive as of September 30, 1998. Of these, 338 028 patients were enrolled in both the VHA and Medicare programs, and had 1999 Medicare Part A and VHA claims data. Because outpatient utilization data for Medicare beneficiaries enrolled in HMOs were not captured for this period, we excluded cohort members with any HMO membership. Consequently, we had 218 689 unique patients with inpatient and outpatient data from both VHA and CMS. Of these patients, 62 had missing race information, and 99 had no face-to-face clinical encounter in the claims data, resulting in a final sample of 218 528 subjects for analysis.

Variables in the Study

Amputation Outcomes in FY 1999. We defined FY 1999 minor amputations (toe [ICD-9-CM 84.11] and transmetatarsal [84.12]) and major amputations (transtibial [84.15] and transfemoral [84.17]) as those with an ICD-9-CM procedure code in any field in either VHA patient treatment files or Medicare Part A files.9-11 Multiple procedures with the same ICD-9-CM code on the same day were considered to be a single amputation, because there are no modifiers to enable identification of bilateral amputations. Similarly, different amputation codes from the same hospitalization were assigned as a single procedure at the highest level.

Risk Factors Associated With Amputation Outcome. We used administrative claims data to identify potential risk factors for lower-extremity ulcers and amputations based on prior epidemiologic studies,7,13 classifying them into demographic variables, foot risk factors, and medical comorbidities. Demographic variables consisting of age, sex, and race were obtained from both VHA and Medicare files. Age and sex information from the 2 data sources were comparable. However, because race was self-reported in the CMS data, in contrast to the VHA data, we used CMS data as the primary source of race.14 As previously described,15 we created 5 categorical baseline (FY 1997 and FY 1998) foot risk factors: peripheral neuropathy, chronic infections, peripheral vascular diseases, foot deformity, and prior amputation (FY 1997 or FY 1998). We also included relevant major medical comorbidities such as ischemic heart disease, congestive heart failure, stroke, and renal disease. All variables were binary (0/1) except for peripheral vascular disease, which had 4 categories of severity (0 indicating not present).

Statistical Analysis

First, we described the population prevalence foot risk factors, medical comorbidities, and amputation outcomes using VHA-only data versus VHA/CMS combined data. We also used multiple logistic regression to examine the association between demographic variables, foot risk factors and medical comorbidities, and the healthcare system (VHA or Medicare) in which the amputation occurred. Second, we obtained the prevalence using VHA/CMS combined data at the VISN regional area level (VISN level) for these variables. At the VISN level, we also calculated the effect of CMS data on the study variables as 100% multiplied by the ratio of the number of additional subjects diagnosed with a condition using CMS data alone to the number of subjects diagnosed with a condition using VHA data alone. We summarized the distribution of these VISN-level percentages (the prevalence and CMS impact) using median, minimum, and maximum values.

Third, we separately fit a risk-adjustment model for VHA-only data versus combined VHA/CMS data for major and minor amputations using a multinomial logistic regression model previously validated for total amputations with combined VHA/CMS data.15 Because our response outcome is dichotomized, we simultaneously modeled 2 sets of logits (eg, major amputation versus no amputation, minor amputation versus no amputation); this approach is more sophisticated and efficient than if the same 2 sets of logits were modeled separately (ie, binomial logistic regression) We then compared and noted the magnitudes, signs, and significance levels of each variable for VHA versus combined VHA/CMS data. The C-statistic was used to evaluate the predictive ability of models, and the Hosmer-Lemeshow statistic was reported as a global index for evaluation of goodness-of-fit.16,17

We used the risk-adjustment model to calculate observed-to-expected (O/E) ratios and to rank VISNs based on these standardized ratios. The observed number of total amputations was determined in each VISN. The expected number of amputations in each VISN was calculated by summing the predicted probability of amputations derived from the multinomial logistic model for all patients from each VISN. VISNs were ranked in order of their O/E ratios and identified as outliers if the 99% confidence interval (CI) for the O/E ratios did not include 1.0.18 VISNs with O/E ratios greater than 1 and 99% CIs that did not cross 1 were classified as "high" outliers, implying a higher number of amputations than predicted by the model using baseline characteristics. Similarly, VISNs with O/E ratios less than 1 and 99% CIs that did not cross 1 were classified as "low" outliers, implying a lower number of amputations than predicted by the model using baseline characteristics. We note that the outlier identification methodology is solely based on statistical significance tests, and does not necessarily imply clinically important differences. Finally, we evaluated the changes in ranking of the 22 VISNs when VHA-only data were compared with combined VHA/CMS data.


The study cohort of 218 528 veterans with diabetes was predominantly male (98.5%) and white (79.9%); 33.1% of the study subjects were aged 75 years or older. In Table 1 we present the prevalence of our outcome variable (amputations) and independent risk-adjustment variables for VHA-only data and combined VHA/CMS data. For independent variables, as shown in Table 1, the addition of CMS data to VHA data resulted in various degrees of increase in their prevalence rates. The prevalence of key major disease conditions increased from 0.6% to 1.2% for stroke, 15.5% to 26.1% for congestive heart failure, 42.0% to 54.4% for ischemic heart disease, and 6.9% to 11.0% for chronic renal disease. Similarly, we found an increased prevalence of foot-specific conditions. The prevalence of any peripheral vascular condition (regardless of severity levels) based on VHA data more than doubled when CMS data were added (5.7% vs. 13.0%). The prevalence rate of any chronic deep skin infection also rose from 5.3% to 8.9%.


Copyright AJMC 2006-2019 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