Fredric D. Wolinsky Iowa City VAMC The University of Iowa Li Liu Suzanne E. Bentler The University of Iowa AcademyHealth June 30, 2009 Acknowledgements We appreciate the contributions to our prior papers in BMC Health Services Research (2006;6:131 and 2007;7:70) by: Thomas Miller Hyonggin An Paul Brezinski Thomas Vaughn Gary Rosenthal We gratefully acknowledge the National Institutes of Health for supporting this work with grants R01-AG022913 R03-AG027741 R21-AG030333 R21-AG031307 R21-AT004578 Disclaimer Neither the National Institutes of Health nor the Veterans Administration had any role in the design of the study, the analyses conducted, or the interpretation and presentation of the results. The opinions expressed here today are those of the authors and do not necessarily reflect those of any of the funding, academic or governmental institutions involved. Approvals Our research was approved by all applicable oversight groups, specifically including: the AHEAD restricted data access board (#2003-006; 2-20-2003), the University of Iowa IRB (#200303008; 03-24-2003; and all successful annual reviews), and the Center for Medicare and Medicaid Services (DUA #14807; 3-3-2005). Background About 10 million veterans are eligible to use both the Veterans Health Administration (VHA) and the private health care delivery systems due to their combined veteran and Medicare entitlements This phenomenon is known as dual use Dual use can have positive and negative implications: On the positive side, it provides veterans with access to more sources and sites of health care and to a greater diversity of health care product lines On the negative side, it may decrease the likelihood that veterans receive continuously coordinated care, because that care is received from multiple health professionals in two distinct and disarticulated delivery systems Etiologic Mechanism Older adults, whether veterans or not, have multiple chronic conditions that are best addressed with central management and coordination of services When older adults receive services from several different providers who are not centrally managed and coordinated, adverse processes may result in: needless replication of tests and procedures diagnostic and treatment delays decreased monitoring effectiveness increased likelihood of medical errors contraindicated and competing regimens Those processes begin a cascade that results in preventable hospitalizations (ACSCs), increased costs, and greater mortality risk How Big is the Dual Use Problem? Hard to say, because the estimates range from 10% to 68% One GAO report says that among Medicare-eligible veterans who used any health care services, 81% used Medicare only, 9% used only the VA, and 10% used both systems Fisher and Welch reported that 52% of all Medicare eligible VA patients filed at least one Medicare benefit claim within a single year Another GAO report suggests that 54% of Medicare eligible veterans were dual users VIReC has concluded that although 90% of older VA patients were enrolled in Medicare, 22% used only VA services, 30% used only Medicare services, and 43% used services from both sources Note that the lowest estimated prevalence rate is from the only population-based study, while the highest are from VA user samples Preliminary Studies Previously we estimated both the prevalence of dual use and its potential adversity among older men using the Survey on Assets and Health Dynamics among the Oldest Old (AHEAD) Baseline interviews were conducted in 1993-1994 with 7,447 respondents > 70 years old, with an 80.4% response rate Re-interviews were conducted in 1995, 1998, 2000, 2002, 2004, 2006, and 2008, but we only used the baseline interview data Of the 2,911 AHEAD men, our prior analyses were restricted to the 1,566 (53.8%) who were self-respondents and whose survey data could be linked to the UNICON LLC release of their Medicare claims We used an indirect measure of dual use based on the concordance of self-reported vs. claims-based hospitalizations in the 12 months prior to each man’s baseline interview Preliminary Studies… Over-reporting existed when the self-reported number of hospitalizations exceeded the claims-based number Veteran status was self-reported The cross-classification of over-reporting and veteran status was used to identify four distinct groups: Veterans who over-reported (the implied dual users group; N=96) Veterans who accurately reported (N=792) Non-veterans who over-reported (N=60) Non-veterans who accurately reported (the reference group; N=574) Using this approach, 10.8% of the veterans were “dual users”, which is remarkably equivalent to the only prior population-based estimate, which came from the GAO Preliminary Studies… We then used National Death Index (NDI) data to model the effect of dual use on mortality through 2002 (50.3% overall, but 64.9% of dual users vs. 49.3% of all other men) Covariates included in the proportional hazards regression models were age, race, education, income, wealth, self-rated health, ADLs, IADLs, cancer, diabetes, heart disease, lung disease, stroke, cognitive status, CES-D symptoms, prior self-reported hospitalization, a propensity score variable for selection bias, and a binary marker for having any ACSC hospital admission Model building and estimation followed standard guidelines The Four Comparison PreliminaryAdjusted Studies…95% Confidence Groups, and Any ACSC Hazards Intervals Admission Ratios (Mortality) Veterans Who Over-Reported (Dual Users) 1.561** 1.120 – 2.177 Veterans Who Accurately Reported 1.168 0.982 – 1.389 Non-Veterans Who OverReported 0.775 0.529 – 1.136 Non-Veterans Who Accurately Reported (Reference Group) 1.000 Any ACSC Admission 1.874*** 1.552 – 2.262 Limitations of Our Preliminary Studies Although these results showed that veterans who were dual users of the VA and Medicare systems had a 56% increased risk of mortality, net of the intervening (mediating) risk associated with subsequent hospitalizations for ACSCs, that analysis had several key limitations: Limited linkage to Medicare claims NDI vital status determination only through 2002 Relatively small sample size with selection bias Inability to adjust for managed care participation New Claims Data Linkage In late 2007, we received a much improved Medicare claims linkage from 1991-2005 from ACUMEN LLC, which included all of the SAF files, including the denominator file, which contains death dates, and allowed us to exclude men in managed care at baseline, or to censor them at entry into managed care post-baseline Effective sample size of self-respondents linked to their Medicare claims increased to 2,133 (74%) vs. 1,522 (53%) previously We were able to exclude those in managed care at baseline, and to censor those who entered managed care post-baseline Prevalence of dual use remains the same, around 10-11% Three more years of vital status determination using Medicare denominator file rather than the NDI Ability to determine actual Medicare payments Mortality Determination and Analysis Vital status was taken from the denominator files The denominator files indicate on a monthly basis whether the Medicare beneficiary is alive or dead Because the exact day of the month of death is not given, we coded the number of days from each participant’s baseline AHEAD interview to their death as the actual number of days up to the first day of the month in which decedent status was indicated in the denominator file, plus 15 days We used multivariable proportional hazards regression to model time to death Proportional hazards model development and evaluation followed standard procedures ACSC Determination and Analysis ACSC status was taken from the inpatient files We used the AHRQ Guide to Prevention Quality Indicators: Ambulatory Care Sensitive Conditions codes for the 16 PQIs, including: Chronic Obstructive Pulmonary Disease Dehydration Bacterial Pneumonia Urinary Tract Infection Uncontrolled Diabetes Hypertension Once again, we used multivariable proportional hazards regression to model time to the first post-baseline ACSC hospitalization for each AHEAD man Medicare Payment Determination and Analysis Medicare payment amounts were taken from the inpatient, outpatient, and carrier SAF files We first summed across all three SAF files to determine total actual post-baseline Medicare payments For norming purposes, we then calculated the number of years of post-baseline surveillance for each man Within each strata of truncated years of exposure, we calculated the 90th percentile of total actual Medicare payments, and created status and date markers for men who within their surveillance strata reached the 90th percentile Finally, we used multivariable hazards regression to model time to reaching the 90th percentile for each AHEAD man The Four Comparison Groups Adjusted Adjusted Preliminary Studies… Hazards Hazards Adjusted Hazards Ratios for the 90th Payment Percentile Ratios for Mortality Ratios for ACSCs Veterans Who OverReported (Dual Users) 1.34+ 1.38+ 0.83 Veterans Who Accurately Reported 1.15 0.98 0.81 Non-Veterans Who Over-Reported (Reference Group) 1.00 1.00 1.00 Non-Veterans Who Accurately Reported 1.08 1.15 0.94 Discussion Why is there a difference (only “trend” [p < .10] evidence for mortality and ACSC hospitalizations) between our prior results and those that we have shown here today? One issue is the UNICON vs. ACUMEN Medicare claims data linkage Differences in sample size (one-third larger) and selection bias (less) Ability to exclude those in managed care at the outset, and upon entrance to it Another issue is the NDI vs. Medicare vital status determination All Medicare claims have vital status information (less selection bias), but the comparative validity is not yet well established It could also be the longer length of follow-up and the reduced value of more distal dual use information (baseline only) But there is no effect in the ACUMEN claims on Medical payments, although further modeling work is needed here Implications Although not statistically significant (only trend or p < .10 evidence), our indirect measure of dual use suggests increased mortality risk We are, however, still relying on an indirect measure of dual use based on discordance between self-reports and claims-reports of hospitalizations in the year prior to baseline Therefore, we urge the VA to facilitate and support linkage of the AHEAD survey and Medicare claims data to the VHA electronic health record In addition to allowing a direct assessment of dual use, this would allow us to calculate a time dependent dual use marker And the resulting data set would be, by far, the best ever to address the dual use issue from a population perspective