• Center on Health Equity & Access
  • Clinical
  • Health Care Cost
  • Health Care Delivery
  • Insurance
  • Policy
  • Technology
  • Value-Based Care

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

Publication
Article
The American Journal of Managed CareNovember 2004 - Part 2
Volume 10
Issue 11 Pt 2

Objective: To determine the impact of dual-system utilizationby veterans on regional variation in lower-extremity amputationrates.

Study Design: Retrospective longitudinal cohort analysis.

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

P

Results: The addition of CMS data significantly increased theprevalence of amputations and risk factors for the 218 528 duallyeligible veterans (all values <.001). In FY 1999, we identified 3.1minor 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 conditionin FY 1997-1998 was 5.7% (VHA) versus 13.0% (VHA/CMS).The impact of including CMS data varied across regions for amputationoutcomes, ranging from an additional 34.3% to 150.7%.Using observed-to-expected amputation ratios and 99% confidenceintervals, the addition of CMS data changed the outlier statusfor 8 of 22 regions for both major and minor amputations.

Conclusion: Risk covariates and amputation outcomes weresubstantially underestimated using VHA data only. Our findingsdemonstrate the importance of evaluating dual-system utilizationwhen conducting program evaluations for healthcare systems witha substantial number of dual enrollees.

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

There is an increasing emphasis by the privatesector,1 accrediting organizations,2 and governmentagencies3 on measuring adherence to recommendedprocess and interim outcomes to decreasehospitalizations and patient morbidity and mortality atthe provider, plan, and population level. In the currenthealthcare environment, variation in health outcomeswithin payer and provider healthcare systems is likelyto be used for internal benchmarking and accreditation,and may be subject to public reporting.4 Thus, it is criticalto address issues such as data availability and riskadjustment to ensure the accuracy of the results andfairness in comparisons.

Recently, the Agency for Healthcare Research andQuality (AHRQ) included prevention quality indicatorsin its National Healthcare Quality Report Card.5 Theseindicators were chosen based on expert consensus thatthe provision of timely and effective outpatient carebefore hospitalization could prevent the hospitalization.Because the indicators were developed using hospitalizationrates rather than individual patient-level data,they primarily are intended to assess community-levelvariation; however, they could be adapted by healthcaresystems as well.

However, healthcare system accountability for preventablehospitalizations would be contingent on theability of these systems both to control and evaluate thequality of ambulatory preventive care provided to theirenrollees. Consequently, a significant methodologicalchallenge for both federal and private healthcare systemsis how to evaluate variation in health outcomes—that is, identification of best- and worst-performingregions or units—when beneficiary care may be fragmentedamong more than 1 healthcare system.This fragmentation of care is especially true withinthe Department of Veterans Affairs (VA), the nation'slargest integrated healthcare system. Based on a large,comprehensive survey of veteran clinical users in1999, approximately 62% of veterans self-reported utilizationof non-VA care during the prior year.6 Thelargest dual-coverage category was Medicare, which wasa source of care for 53% of all veteran enrollees reportingdual coverage.

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

Within the Veterans Health Administration (VHA),the Preservation Amputation Care and TreatmentProgram8 mandates VA medical centers to develop multidisciplinaryprograms to identify veterans at risk forlower-extremity complications and to refer them tofoot-care specialists for ongoing evaluation and care.However, prior epidemiologic studies and operationalreports of amputation rates and trends have exclusivelyutilized VHA administrative datasets, thus not capturingrelevant amputation data outside the VHA.9-11

Consequently, our specific objectives in this studywere to determine and evaluate the impact of the inclusionof Medicare utilization data (ie, Centers forMedicare and Medicaid Services [CMS] data) on theprevalence 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 regionalvariation in minor and major amputation outcomes inthe subsequent year. We defined the regions using theservice areas of the Veterans Integrated ServiceNetworks (VISNs).

METHODS

Data Sources and Cohort Assembly

International Classification of Diseases, NinthRevision, Clinical Modification (ICD-9-CM)

We utilized a cohort of veteran clinical users withdiabetes who also were enrolled in Medicare in fiscalyear (FY) 1997 or FY 1998 as previously described.12Potential subjects were identified from several inpatientand outpatient claims-data sources by the presence ofat least 1 diabetes-specific (250.xx, 357.2, 362.0,366.41) code orprescription of a diabetes-specific medicine.

Sample Selection

A total of 446 896 veterans with diabetes were identifiedand were alive as of September 30, 1998. Of these,338 028 patients were enrolled in both the VHA andMedicare programs, and had 1999 Medicare Part A andVHA claims data. Because outpatient utilization data forMedicare beneficiaries enrolled in HMOs were not capturedfor this period, we excluded cohort members withany HMO membership. Consequently, we had 218 689unique patients with inpatient and outpatient data fromboth VHA and CMS. Of these patients, 62 had missingrace information, and 99 had no face-to-face clinicalencounter in the claims data, resulting in a final sampleof 218 528 subjects for analysis.

Variables in the Study

Amputation Outcomes in FY 1999.

ICD-9-CM

ICD-9-CM

ICD-9-CM

We defined FY1999 minor amputations (toe [ 84.11] andtransmetatarsal [84.12]) and major amputations(transtibial [84.15] and transfemoral [84.17]) as thosewith an procedure code in any field in eitherVHA patient treatment files or Medicare Part A files.9-11Multiple procedures with the same code onthe same day were considered to be a single amputation,because there are no modifiers to enable identificationof bilateral amputations. Similarly, differentamputation codes from the same hospitalization wereassigned as a single procedure at the highest level.

Risk Factors Associated With Amputation Outcome.

We used administrative claims data to identify potentialrisk factors for lower-extremity ulcers and amputationsbased on prior epidemiologic studies,7,13 classifyingthem into demographic variables, foot risk factors, andmedical comorbidities. Demographic variables consistingof age, sex, and race were obtained from both VHAand Medicare files. Age and sex information from the 2data sources were comparable. However, because racewas self-reported in the CMS data, in contrast to theVHA data, we used CMS data as the primary source ofrace.14 As previously described,15 we created 5 categoricalbaseline (FY 1997 and FY 1998) foot risk factors:peripheral neuropathy, chronic infections, peripheralvascular diseases, foot deformity, and prior amputation(FY 1997 or FY 1998). We also included relevant majormedical comorbidities such as ischemic heart disease,congestive heart failure, stroke, and renal disease. Allvariables were binary (0/1) except for peripheral vasculardisease, which had 4 categories of severity (0 indicatingnot present).

Statistical Analysis

First, we described the population prevalence footrisk factors, medical comorbidities, and amputationoutcomes using VHA-only data versus VHA/CMS combineddata. We also used multiple logistic regression toexamine the association between demographic variables,foot risk factors and medical comorbidities, andthe healthcare system (VHA or Medicare) in which theamputation occurred. Second, we obtained the prevalenceusing VHA/CMS combined data at the VISNregional area level (VISN level) for these variables. Atthe VISN level, we also calculated the effect of CMS dataon the study variables as 100% multiplied by the ratio ofthe number of additional subjects diagnosed with a conditionusing CMS data alone to the number of subjectsdiagnosed with a condition using VHA data alone. Wesummarized 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 forVHA-only data versus combined VHA/CMS data for majorand minor amputations using a multinomial logisticregression model previously validated for total amputationswith combined VHA/CMS data.15 Because ourresponse outcome is dichotomized, we simultaneouslymodeled 2 sets of logits (eg, major amputation versus noamputation, minor amputation versus no amputation);this approach is more sophisticated and efficient than ifthe same 2 sets of logits were modeled separately (ie,binomial logistic regression) We then compared and notedthe magnitudes, signs, and significance levels of each variablefor VHA versus combined VHA/CMS data. The C-statisticwas used to evaluate the predictive ability of models,and the Hosmer-Lemeshow statistic was reported as aglobal index for evaluation of goodness-of-fit.16,17

We used the risk-adjustment model to calculateobserved-to-expected (O/E) ratios and to rank VISNsbased on these standardized ratios. The observed numberof total amputations was determined in each VISN.The expected number of amputations in each VISN wascalculated by summing the predicted probability ofamputations derived from the multinomial logistic modelfor all patients from each VISN. VISNs were ranked inorder of their O/E ratios and identified as outliers if the99% confidence interval (CI) for the O/E ratios did notinclude 1.0.18 VISNs with O/E ratios greater than 1 and99% CIs that did not cross 1 were classified as "high" outliers,implying a higher number of amputations than predictedby the model using baseline characteristics.Similarly, VISNs with O/E ratios less than 1 and 99% CIsthat did not cross 1 were classified as "low" outliers,implying a lower number of amputations than predictedby the model using baseline characteristics. We note thatthe outlier identification methodology is solely based onstatistical significance tests, and does not necessarilyimply clinically important differences. Finally, we evaluatedthe changes in ranking of the 22 VISNs when VHA-onlydata were compared with combined VHA/CMS data.

RESULTS

The study cohort of 218 528 veterans with diabeteswas 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 outcomevariable (amputations) and independent risk-adjustmentvariables for VHA-only data and combinedVHA/CMS data. For independent variables, as shown inTable 1, the addition of CMS data to VHA data resultedin various degrees of increase in their prevalence rates.The prevalence of key major disease conditionsincreased from 0.6% to 1.2% for stroke, 15.5% to 26.1%for congestive heart failure, 42.0% to 54.4% for ischemicheart disease, and 6.9% to 11.0% for chronic renal disease.Similarly, we found an increased prevalence offoot-specific conditions. The prevalence of any peripheralvascular condition (regardless of severity levels)based on VHA data more than doubled when CMS datawere added (5.7% vs. 13.0%). The prevalence rate of anychronic deep skin infection also rose from 5.3% to 8.9%.

Using only VHA datasets, we identified 3.1 minoramputations and 4.5 major amputations per 1000patients in the cohort. The addition of CMS dataincreased the crude amputation rate to 5.5 per 1000patients for minor amputations (75% increase) and 8.6per 1000 patients for major amputations (90%increase), respectively.

P

P

P

The percentage of veterans with diabetes incurringan amputation in the CMS data ranged from 26.4% to61.8% among the VISNs. Ten VISNs were in the range of40% to 50%, while the remaining 12 VISNs were equallysplit in the ranges of less than 40% and more than 50%.We examined whether the healthcare system in whichthe amputation occurred (Medicare vs VHA) was associatedwith the independent variables under study, aswell as the VISN region. Because the results from themultiple logistic regression models were similar formajor and minor amputations, we combined the 2amputation categories. We found that VISN variation inthe healthcare system site of amputations was significant( < .001). We found that the site of the amputations(VHA or the private sector) was significantlyassociated with VISN region. In addition, we found thatveterans who incurred an amputation in the private sectorcompared with the VHA were more likely to be older( < .01), Caucasian, and to have more severe peripheralvascular conditions and medical comorbiditiessuch as congestive heart failure. On the other hand,they were less likely to have preexisting chronic skininfections ( < .01) and prior amputations.

We also evaluated VISN-level variation in the prevalenceof baseline (FY 1997-1998) covariates and theoutcome variable, amputations in FY 1999, using VA-onlyor combined datasets (Table 2). The variation inCMS contribution across VISNs ranged from 34.3% to150.7% for FY 1999 amputation outcomes. For the independent variables, thevariation in CMS contributionacross VISNs ranged from 32.9% to106.0% for prior (FY 1997-1998) amputationand from 24.9% to 257.6% for any peripheral vascularcondition.

P

We contrasted the parameter estimates andmodel performance usingVHA-only data versusVHA/CMS combined dataseparately for major (versusno) amputation andminor (versus no) amputation(data not shown).Although small differencesin the parameterestimates were noted forthe VHA and VHA/CMSminor-amputation risk-adjustmentmodels (with the exception of patientsex), the results were similar;no parameter estimateschanged directionof association (positive tonegative or vice versa)and all variables remainedsignificant. The additionof CMS data had noimpact on the C-statisticfor the major-amputationmodel (C = 0.82 forboth VHA-only and VHA/CMS data). Similarly, the additionof CMS data for the minor-amputation model minimallydecreased the C-statistic from 0.79 to 0.78. Allmodels demonstrated adequate goodness-of-fit, basedon the Hosmer-Lemeshow test ( > .05).

Table 3 shows our evaluation of individual VISN performanceas determined by O/E ratios of major-amputationrates versus minor-amputation rates and outlierstatus. The range of O/E ratios across VISNs for majoramputations was greater for VHA-only data (0.58 to1.45) than for VHA/CMS data (0.72 to 1.31). Similarly,the range of O/E ratios for minor amputations wasgreater for VHA-only data (0.56 to 1.42) than forVHA/CMS data (0.75 to 1.46).

For major amputations, the addition of CMS datachanged the ranking of 17 of 22 VISNs and the outlierstatus of 8 VISNs. Specifically, the CMS data changed 3VISNs (H, M, T) that were rated as high outliers by VHA-onlydata to nonoutlier status with VHA/CMS data.Furthermore, 2 VISNs (A, S) that were initially rated asnonoutliers were reclassified as low outliers after additionof the CMS data. Two nonoutlying VISNs (G, O)changed to high outliers, and one VISN (R) changedfrom a low outlier to a nonoutlier. For minor amputations,the addition of CMS data changed the ranking of19 of 22 VISNs and the outlier status of 8 VISNs. These8 VISNs were not the same 8 VISNs whose outlier statuschanged for major amputations. The VISNs thatchanged from high outliers to nonoutliers for eithermajor or minor amputations were those that had thelowest percentages (<40%) of veterans who had amputationsin the private sector. On the other hand, thosewhose status changed from nonoutlier/low outlier tohigh outlier in either amputation category were amongthose that had the highest percentage (>50%) of veteranswho had amputations in the private sector (data notshown).

DISCUSSION

Our results demonstrate that the inclusion ofMedicare administrative data yielded a more comprehensivedataset for foot risk factors, comorbidities, andamputations. In some instances (eg, peripheral vascularconditions), the addition of CMS data doubled theprevalence. Most importantly, the addition of Medicaredata to VHA data changed the outlier designation for 8of 22 VISN regions for major amputations and a differentsubset of 8 of 22 VISN regions for minor amputations.There were marked demographic differencesamong dually enrolled veterans incurring amputationsin the VHA compared with the private sector, and theVISN high-outlier designation based on VHA data alonewas associated with higher proportions of amputationsperformed within the VHA. These 2 findings suggestthat selection biases in patient choice of healthcare siteare significant factors in the evaluation of health outcomes.

Our findings are consistent with prior researchdemonstrating the difficulties in evaluating care providedto individuals with chronic diseases across multiplehealth systems. In addition to underestimating hospitalizations,19,20 it is increasingly appreciated that use ofadministrative data underestimates performance measurementadherence in both VHA and Medicare facilities.21,22Rather than being unique to the VHA, the challengesof dual-system enrollment for health outcomeassessment are pertinent to other governmental agencies,including the Medicare and Medicaid programs23,24and the Department of Defense Tricare program.25 Ourfindings raise legitimate policy questions regarding howdecision makers can assess the effectiveness of diseasemanagement programs if payment and eligibility issuespresent both patients and providers with conflictingrules and incentives that may lead to discoordination ofcare.26 CMS recently initiated a number of demonstrationprojects on care coordination and disease managementin beneficiaries with chronic disease; in some, asingle disease management program will pay for allmedications, even if enrollees are dually eligible.27However, because fair and accurate comparison of specifichospitalization rates requires a totality of data toascertain and risk-adjust outcomes, comparisons amongnational systems may not be possible until there is asingle electronic medical record.28

Consequently, as an interim proposal for implementationof prevention quality indicators within largenational healthcare systems, we suggest that single-payeradministrative data could be used to identifyregional variation for the purpose of programmatic evaluationrather than benchmarking. This proposal is consistentwith recent disease-specific certificationstandards established by the Joint Commission onAccreditation of Healthcare Organizations.29

We previously demonstrated that although system-levelcare coordination for patients with feet at high riskof amputation may be an important process in reducingamputation,30 opportunities remain for improvement inthe organizational coordination of foot-care programs.31Consistent with the chronic-care model,32 we proposethat healthcare systems could utilize electronic medicalrecords to create disease-specific registries to identifyindividuals who receive care in more than 1 healthcaresystem. Healthcare systems would then be able toproactively coordinate care between, as well as within,systems of care. Such efforts would respect patientchoice, inform local quality improvement efforts toaddress unmet needs and improve access, and mayimprove cost and health outcomes.

This study has a number of strengths. First, ournational cohort of dually eligible veterans with diabeteswas analyzed on an individual patient level. We wereable to reliably ascertain the diagnosis of diabetes forour cohort, as well as identify both major-amputationand minor-amputation outcomes and comorbiditiesusing both VHA and Medicare datasets. Second, ourstudy of VHA and Medicare fee-for-service users explicitlyexcluded users of Medicare HMOs. Thus, there wasmore certainty about the totality of medical information.We were able to identify the relative contributionof Medicare data to comorbidities and amputation outcomes,as well as the impact of these data on networkrank and outlier status.

However, we also acknowledge several limitations.We were unable to ascertain Medicaid or private-sectorutilization. Although the accuracy of amputation coding has been validated in the VHA,33 the accuracy ofthe outpatient coding for covariates in both the VHAand Medicare datasets has not been systematicallyanalyzed.

CONCLUSION

Our study demonstrates that addition of CMSadministration data to VHA databases not only addsimportant comorbidity data and amputation outcomes,but also alters statistical comparisons of regional variationsdefined by VHA network service areas. Ourresults demonstrate the potential limitations of the useof amputation rates in particular, and perhaps all preventionquality indicators, within systems of carewhere fragmentation of healthcare occurs. Furtherresearch is necessary to understand the various factors,including plan benefits, access, and system-level quality-of-care issues, that impact patient care preferencesand their choice of care.

From the Center for Healthcare Knowledge Management, Veterans Affairs–New JerseyHealthcare System, East Orange, NJ (CLT, JDG, DH, MR, AT, LP); the University ofMedicine and Dentistry of New Jersey–New Jersey Medical School, Newark, NJ (CLT, DH,LP); New York University School of Medicine, New York, NY (JDG); the New Jersey Instituteof Technology, Newark, NJ (AT); the Bedford Center for Health Quality, Outcomes &Economic Research, Bedford, Mass (DM); Boston University School of Public Health,Boston, Mass (DM); Rutgers University, New Brunswick, NJ (SC); and the Veterans HealthAdministration Office of Patient Care Services, Washington, DC (GH).

This study was funded by grant 02-079-1 from the Health Services Research andDevelopment Service, Department of Veterans Affairs (Dr Pogach), and a Department ofVeterans Affairs Epidemiology (Medical Service) Merit Review grant (Drs Pogach andMiller).

Address correspondence to: Leonard Pogach, MD, MBA, Center for HealthcareKnowledge Management, Veterans Affairs–New Jersey Healthcare System, 385 TremontAve, East Orange, NJ 07018. E-mail: leonard.pogach@med.va.gov.

Surgery.

1.Birkmeyer JD, Dimick JB. Potential benefits of the new Leapfrog standards:effect of process and outcomes measures. 2004;135:569-575.

JAMA.

2. Schneider EC, Riehl V, Courte-Wienecke S, Eddy DM, Sennett C. Enhancingperformance measurement: NCQA's road map for a health information framework.National Committee for Quality Assurance. 1999;282:1184-1190.

3. Healthy People 2010. Available at: http://www.healthypeople.gov/. AccessedSeptember 1, 2004.

NEngl J Med.

4. Lee TH, Meyer GS, Brennan TA. A middle ground on public accountability. 2004;350:2409-2412.

AHRQ Quality Indicators. Guideto Prevention Quality Indicators: Hospital Admission for Ambulatory CareSensitive Conditions.

5. Agency for Healthcare Research and Quality. Rockville, Md: Agency for Healthcare Research and Quality;2001. AHRQ publication 02-R0203.

Med Care Res Rev.

6. Shen Y, Hendricks A, Zhang S, Kazis LE. VHA enrollees' health care coverageand use of care. 2003;60:253-267.

Diabetes in America.

7. Reiber GE, Boyko EJ, Smith DG. Lower extremity foot ulcers and amputationsin diabetes. In: 2nd ed. Bethesda, Md: National Institute ofDiabetes and Digestive and Kidney Diseases; 1995. NIH publication 95-1468.

Preservation Amputation Care and Treatment.

8. Department of Veterans Affairs. Washington, DC: Department of Veterans Affairs; 2001. VHA directive 2001-030.

J Rehabil Res Dev.

9. Mayfield JA, Reiber GE, Maynard C, Czerniecki JM, Caps MT, Sangeorzan BJ.Trends in lower limb amputation in the Veterans Health Administration, 1989-1998. 2000;37(1):23-30.

Diabetes Care.

10. Young BA, Maynard C, Reiber G, Boyko EJ. Effects of ethnicity and nephropathyon lower-extremity amputation risk among diabetic veterans. 2003;26:495-501.

11. Healthcare Analysis and Information Group. Lower extremity rates, complicationsand outcomes in VHA (FY 1998-99). Available at: http://www.va.gov/haig/.Accessed October 22, 2004.

Diabetes Care.

12. Miller DR, Safford MM, Pogach L. Who has diabetes? Best estimates of diabetesprevalence in the Veterans Health Administration based on computerizedpatient data. 2004;27(suppl 2):B10-B21.

Diabetes Care.

13. Adler AI, Boyko EJ, Ahroni JH, Smith DG. Lower-extremity amputation indiabetes. The independent effects of peripheral vascular disease, sensory neuropathy,and foot ulcers. 1999;7:1029-1035.

Health Care Financ Rev.

14. Arday SL, Arday DR, Monroe S, Zhang J. HCFA's racial and ethnic data: currentaccuracy and recent improvements. 2000;21(4):107-116.

Med Care.

15. Tseng CL, Mangala R, Miller D, et al. Use of administrative data to risk adjustamputation rates in a national cohort of Medicare-enrolled veterans with diabetes. In press.

Risk Adjustment for MeasuringHealthcare Outcomes.

16. Ash AS, Shwartz M. Evaluating the performance of risk-adjustment methods:dichotomous outcomes. In: Iezzoni LI, ed. 2nd ed. Chicago, Ill: Health Administration Press;1997:427-470.

Applied Logistic Regression.

17. Hosmer D, Lemeshow S. New York, NY: JohnWiley & Sons, Inc; 2000.

IARC Sci Publ.

18. Breslow NE, Day NE. Statistical methods in cancer research, volume II: thedesign and analysis of cohort studies. 1987;82:1-406.

Med Care.

19. Fleming C, Fisher ES, Chang CH, Bubolz TA, Malenka DJ. Studying outcomesand hospital utilization in the elderly. The advantages of a merged data base forMedicare and Veterans Affairs hospitals. 1992;30:377-391.

Health Serv Res.

20. Wright SM, Daley J, Fisher ES, Thibault GE. Where do elderly veterans obtaincare for acute myocardial infarction: Department of Veterans Affairs or Medicare? 1997;31:739-754.

Int J Qual Health Care.

21. Jones D, Hendricks A, Comstock C, et al. Eye examinations for VA patientswith diabetes: standardizing performance measures. 2000;12:97-104.

Health Serv Res.

22. Keating NL, Landrum MB, Landon BE, Ayanian JZ, Borbas C, Guadagnoli E.Measuring the quality of diabetes care using administrative data: is there bias? 2003;38(6 pt 1):1529-1545.

Health Care Financ Rev.

23. Walsh EG, Clark WD. Managed care and dually eligible beneficiaries: challengesin coordination. 2002;24:63-82.

Issue Brief Natl Health Policy Forum.

24. Ryan J, Super N. Dually eligible for Medicare and Medicaid: two for one ordouble jeopardy? 2003;(794):1-24.

Manag Care Interface.

25. Chargois CA, Sumter RL. Solving the military's health care dilemma forMedicare eligibles. 2001;14(2):44-49.

Healthcare FinancManage.

26. Ripley J. Coordinating care for the dual-eligible population. 2001;55(6):39-43.

27. Centers for Medicare and Medicaid Services. Demonstrations serving thosedually-eligible for Medicare and Medicare. Available at:http://63.240.208.147/researchers/demos/DESM.asp. Accessed September 10,2004.

J Am Med InformAssoc.

28. Yasnoff WA, Humphreys BL, Overhage JM, et al. A consensus action agendafor achieving the national health information infrastructure. 2004;11:332-338.

29. Joint Commission on Accreditation of Healthcare Organizations. Disease-specificcare. Available at: http://www.jcaho.org/dscc/dsc/index.htm. AccessedSeptember 10, 2004.

Diabetes Care.

30. Wrobel JS, Charns MP, Diehr P, et al. The relationship between providercoordination and diabetes-related foot outcomes. 2003;26:3042-3047.

Am J Manag Care.

31. Pogach L, Charns MP, Wrobel JS, et al. The impact of policies and performancemeasurement upon the development of organizational coordinating strategiesfor chronic care delivery. 2004;10(2 pt 2):171-180.

Milbank Q.

32. Wagner EH, Austin BT, Von Korff M. Organizing care for patients with chronicillness. 1996;74:51-544.

Med Care.

33. Kashner TM. Agreement between administrative files and written medicalrecords: a case of the Department of Veterans Affairs. 1998;36:1324-1336.

Related Videos
Related Content
© 2024 MJH Life Sciences
AJMC®
All rights reserved.