Author + information
- Received March 28, 2014
- Revision received October 2, 2014
- Accepted October 30, 2014
- Published online January 27, 2015.
- P. Michael Ho, MD, PhD∗,†∗ (, )
- Colin I. O’Donnell, MS∗,†,
- Steven M. Bradley, MD, MPH∗,†,
- Gary K. Grunwald, PhD∗,†,
- Christian Helfrich, PhD‡,
- Michael Chapko, PhD‡,
- Chuan-Fen Liu, PhD‡,
- Thomas M. Maddox, MD, MSc∗,†,
- Thomas T. Tsai, MD, MSc∗,†,
- Robert L. Jesse, MD, PhD§,
- Stephan D. Fihn, MD, MPH‡,‖ and
- John S. Rumsfeld, MD, PhD∗,†
- ∗Veterans Affairs Eastern Colorado Health Care System, Denver, Colorado
- †Division of Cardiology, University of Colorado Denver, Denver, Colorado
- ‡Veterans Affairs Puget Sound Health Care System, Seattle, Washington
- §Department of Veteran Affairs, Veterans Health Administration, Washington, DC
- ‖Veterans Affairs Office of Analytics and Business Intelligence, Washington, DC
- ↵∗Reprint requests and correspondence:
Dr. P. Michael Ho, Denver Veteran Affairs Medical Center, University of Colorado, Denver, Section of Cardiology, 1055 Clermont Street (111B), Denver, Colorado 80220.
Background There is significant interest in measuring health care value, but this concept has not been operationalized in specific patient cohorts. The longitudinal outcomes and costs for patients after percutaneous coronary intervention (PCI) provide an opportunity to measure an aspect of health care value.
Objectives This study evaluated variations in 1-year outcomes (risk-adjusted mortality) and risk-standardized costs of care for all patients undergoing PCI in the Veterans Affairs (VA) system from 2007 to 2010.
Methods This retrospective cohort study evaluated all veterans undergoing PCI at any of 60 hospitals in the VA health care system, using data from the national VA Clinical Assessment, Reporting, and Tracking (CART) program. Primary outcomes were 1-year mortality and costs following PCI. Risk-standardized mortality and cost ratios were calculated, adjusting for cardiac and noncardiac comorbidities.
Results A median of 261 PCIs were performed in the 60 hospitals during the study period. Median 1-year unadjusted hospital mortality rate was 6.13%. Four hospitals were significantly above the 1-year risk-standardized median mortality rate, with median mortality ratios ranging from 1.23 to 1.28. No hospitals were significantly below median mortality. Median 1-year total unadjusted hospital costs were $46,302 per patient. There were 16 hospitals above and 19 hospitals below the risk-standardized median cost, with risk-standardized ratios ranging from 0.45 to 2.09, reflecting a much larger magnitude of variability in costs than in mortality.
Conclusions There is much smaller variation in 1-year risk adjusted mortality than in risk-standardized costs after PCI in the VA. These findings suggest that there are opportunities to improve PCI value by reducing costs without compromising outcomes. This approach to evaluating outcomes and costs together may be a model for other health systems and accountable care organizations interested in operationalizing value measurement.
There is increasing interest in measuring health care value, particularly as the health care system moves toward accountable care (1,2). Within accountable care organizations (ACOs), groups of providers assume responsibility for cost and quality of care for a patient population. Value in health care focuses on similar concepts by measuring outcomes achieved relative to costs for a cycle of care (1). The value numerator is outcomes of care, such as mortality or morbidity; the value denominator is total cost of care over time, not just a single episode of care such as hospitalization (1). Attaining high-value care, such as positive clinical outcomes at low costs, is of interest to patients, providers, health systems, and payers. To date, value assessments have not been operationalized and applied to specific patient populations.
Percutaneous coronary intervention (PCI) is an important component of care for patients with ischemic heart disease (3). In 2010, 492,000 PCIs were performed in the United States at an mean charge of $67,000 per procedure (4). Although quality of care for the PCI procedure itself has improved, 1-year rehospitalization and mortality rates following the procedure remain high (5–7). These findings highlight the need to take a longitudinal approach toward evaluating PCI care, including costs for this care. Accordingly, PCI and the longitudinal care provided to patients following the procedure provide an opportune clinical scenario with which to measure 1 aspect of health care value.
Accordingly, we evaluated 1-year outcomes (risk-adjusted mortality) and 1-year risk-standardized costs of care for all patients who underwent PCI in the U.S. Department of Veterans Affairs (VA) health care system from 2007 to 2010. We compared risk-standardized mortality and costs for VA hospitals that performed PCI and evaluated the proportion of costs attributable to the index procedure compared with the follow-up period. It is hoped that the findings will inform the discussion of how to measure and move toward achieving high-value PCI care and also serve as a potential model for similar assessments in non-VA settings.
The Veterans Health Administration is the largest integrated health care system in the United States (8). The VA has a financial management system (Decision Support System [DSS]) that uses a cost accounting method to track costs rather than charges and payments. This system is the primary financial data source for VA health care operations and the source of cost data for these analyses (8,9).
The VA Clinical Assessment, Reporting, and Tracking (CART) program is the clinical quality program for 76 cardiac catheterization laboratories, 64 of which perform PCI. The CART program uses a clinical software application integrated with the VA electronic health record to collect patient and procedural data at the point-of-care for all cardiac catheterization procedures and PCIs performed in the VA system (10,11).
Patient and hospital inclusion criteria
Hospitals had to perform a minimum of 20 PCIs between October 1, 2007, to September 30, 2010 to be included in the study. During these study years, patients without a history of PCI and who underwent a PCI in the VA system were included. Following the index PCI, all subsequent health care utilizations during the year were aggregated for that patient, regardless of where care occurred, either at the PCI hospital or at another VA hospital. We attributed all patient costs and mortality to the index PCI hospital. We also included costs of care occurring outside of VA where VA paid for care using the fee basis files. As noted below, we also had Medicare utilization data for a subset of patients. Of the 64 hospitals that performed PCI, we excluded 4 hospitals that performed fewer than 20 PCIs during the study period and 57 patients (0.03%) for whom cost data were not captured. The final dataset consisted of 60 hospitals and 19,148 patients. The Colorado Multiple Institutional Review Board approved the current analysis.
The primary clinical outcome was all-cause mortality during the year following the index PCI procedure. Death was ascertained from the VA Information Resource Center (VIReC) Vital Status File, which compiles data from the BIRLS (Beneficiary Identification Records Locator Subsystem) Death file, the VA Medicare Vital Status File, and the Social Security Administration (SSA) Death Master File.
The primary cost outcome was total cost during the year after PCI. Costs for each patient were collected from DSS and adjusted for regional variations, using the Medicare wage index. Total costs were divided into 2 periods: 1) index procedural costs, defined as all costs within 7 days of PCI; and 2) subsequent costs, defined as all costs from days 8 to 365. Within these categories, we further divided costs into inpatient, outpatient, and/or fee basis costs. All inpatient costs were further broken down into 6 unique DSS-defined categories: nursing, pharmacy, radiology, surgery, laboratory, and all other costs. All outpatient costs were aggregated for each patient and then to the PCI hospital. Outpatient costs were matched to internal VA codes that identified the purpose of visit.
Statistical analyses were performed using SAS version 9.3 software (SAS Institute Inc., Cary, North Carolina), R version 2.15.1 software (R Foundation for Statistical Computing, Vienna, Austria), and WinBUGS version 1.4.3 software (Imperial College and MRC, Cambridge, United Kingdom).
Bayesian profiling of standardized mortality risk (SMR) and standardized cost risk (SCR) was modeled following Markov chain Monte Carlo (MCMC) methods (14). Logistic regression was used for mortality, and cost was logged prior to modeling. To adjust for both cardiac and noncardiac comorbidities, we included variables from the National Cardiovascular Data Registry risk prediction mortality model and from the Elixhauser comorbidities scale (12,13). The resulting dataset contained more than 35 adjustment covariates. To reduce the number of covariates used in MCMC, a single backward elimination using maximal likelihood was used to obtain the most significant 24 covariates, respectively (Table 1).
We used the univariate method of Timbie and Normand (14) in WinBUGS with a single chain of 10,000 iterations with 2,000 burn-in iterations to obtain an SMR and an SCR and their associated credibility intervals for each hospital. These adjusted ratios are Bayesian analogs to observed to expected ratios. SMRs and SCRs for each hospital were used to quantify variation in mortality and cost across hospitals and to identify significantly unusual hospitals as those whose 95% credibility interval did not contain 1. To examine the association between mortality and cost at the hospital level, each hospital’s SCR was plotted against the SMR, and Pearson correlation was used.
We performed several sensitivity analyses. First, despite a national cost accounting system, we evaluated potential variations in cost coding across hospitals by assessing Pearson's correlation coefficient between costs and utilization after PCI discharge. We derived a mean utilization per hospital on the basis of number of unique outpatient visits and inpatient admissions by number of patients undergoing PCI. A positive correlation would suggest that higher costs at hospitals reflect higher utilization and not necessarily variations in cost coding. Second, VA patients who are 65 years of age or older may obtain care outside of VA by using Medicare benefits. To address this, we evaluated the correlation between hospital readmissions in Medicare and VA hospital costs. A negative correlation would suggest that VA hospitals are low cost because their patients are rehospitalized outside of VA and that these readmissions are not accounted for in VA costs. Medicare readmission data were available for the cohort through December 31, 2010. Third, we also adjusted for patient costs in the 6 months prior to the PCI in the assessment of variation in hospital costs to account for differences in high-cost patients prior to the PCI between hospitals. Fourth, we stratified our cohort by acute coronary syndrome (ACS) versus non-ACS indication for the PCI procedure and assessed variations in 1-year mortality and costs across the hospitals.
The median number of PCIs performed in the 60 hospitals was 261, and the mean was 319 (interquartile range [IQR]: 195 to 357) (Table 1). Almost one-half (43.0%) of PCIs were performed for ACS. Comorbidities were common: 21.6% heart failure, 17.5% peripheral vascular disease, 29.5% chronic pulmonary disease, 11.0% renal failure, 42.6% diabetes, and 39.7% obesity.
At the hospital level, the 30-day unadjusted median mortality rate after PCI was 1.29% (IQR: 0.85% to 2.22%). One-year unadjusted median mortality was 6.13% (IQR: 4.51% to 7.34%). With risk-standardized median 1-year mortality ratio of 1.00, 4 hospitals had credibility intervals above 1.0 (Figure 1). These hospitals had mortality rates approximately 23.2% to 28.5% higher than the median. There were no hospitals with credibility intervals below the risk-standardized median ratio.
There was substantially greater variation in total costs. The median unadjusted total 1-year costs were $46,302 (IQR: $37,291 to $57,886) per patient (Figure 2). Risk-standardized 1-year total costs demonstrated that 16 hospitals had credibility intervals above the risk-standardized median, up to 2.09 (95% confidence interval [C]I: 1.84 to 2.36) or 209% higher costs than median, and 19 hospitals were below, with the lowest being 0.45 (95% CI: 0.39 to 0.51) or 55% lower costs than median (Figure 3). Adjusting for patient costs in the 6 months prior to PCI did not change the number of outlier hospitals. The index procedural costs accounted for 41.7% of total 1-year costs. Of these index costs, 66.2% were inpatient, 32.6% were outpatient, and 1.2% were fee basis costs. Index inpatient costs were attributed to the following categories: nursing, 31.6%; surgery, 2.8%; laboratory, 2.6%; radiology, 2.2%; pharmacy, 6.6%; and all others, 54.2%.
Following the index procedure, subsequent costs accounted for 58.3% of total 1-year costs and were attributable to inpatient (28.2%), outpatient (64.3%), and fee basis costs (7.5%). The top 3 primary diagnoses for inpatient costs were: 1) other forms of chronic ischemic heart disease; 2) heart failure; and 3) care involving rehabilitation procedures (Table 2 lists the top 10 diagnoses). For outpatient costs, cardiac-related care was the largest cost category (18.0% of all outpatient costs), followed by pharmacy (15.2%), prosthetics (14.9%), outpatient procedures (7.8%), screening and laboratory (7.2%), primary care (6.5%), and radiology (3.3%). The remaining outpatient costs were attributed to a variety of reasons (26.7%).
Next, we plotted standardized mortality and total cost ratios for each hospital (Central Illustration). There were 4 hospitals with risk-standardized credibility intervals above 1.0, (Central Illustration, blue dots and lines). There were 35 hospitals with risk-standardized 1-year total costs either above or below 1.0 (Central Illustration, red dots and lines). Overall, there was no correlation between hospital standardized mortality and costs (r = 0.004; 95% CI: −0.24 to 0.26; p = 0.97). One hospital had both mortality and cost ratios that were above the standardized risks, suggesting that this hospital was “low” value.
In sensitivity analyses, there was significant and strong correlation between utilization on the basis of outpatient visits and inpatient admissions and costs (r = 0.87; 95% CI: 0.79 to 0.92; p < 0.001), supporting the fact that higher costs at hospitals reflect higher utilization. Next, the correlation between hospital readmissions in Medicare and VA hospitals costs was small and not statistically significant (r = 0.11; p = 0.40), suggesting that the reason for lower-cost VA hospitals was not related to greater Medicare use by their patients. Finally, similar to the primary analysis, variations at the hospital level in 1-year mortality for ACS were similar to those for non-ACS indications for the procedure. For 1-year costs, significant variations remained, with 14 hospitals above and 16 hospitals below the risk-standardized median for ACS indications for the procedure and 16 above and 20 hospitals below the risk-standardized median for non-ACS indications for the procedure.
We evaluated hospital variations in 1-year risk-standardized mortality and 1-year costs of care following PCI at 60 VA hospitals (Central Illustration). Variations in 1-year mortality were low overall, and there were only 4 outliers, which had median standardized mortality ratios up to 28% above the median. In contrast, there was substantially greater variability in 1-year costs following PCI. Following risk adjustment, costs ranged from 55% below to 209% above the mean standardized costs. These findings highlight the fact that although there may be selective opportunities to improve patient outcomes following PCI, there appear to be greater opportunities to reduce costs for longitudinal PCI care. The results of this study suggest that the value of PCI care for the VA system may be enhanced through reducing unnecessary variation in costs of care while continuing to achieve good patient outcomes.
To our knowledge, this is one of the first studies to operationalize and measure both longitudinal outcomes and costs for a specific patient cohort and from a system perspective. The study findings may inform current discussions of health care value measurement and may be a model for other systems or ACOs attempting to measure and improve value. It has been recommended that in a health care system structured around value for patients, care should be organized around longitudinal care and outcomes rather than single episodes of care like hospitalization (1,15). Accordingly, we focused our assessment on 1-year outcomes and costs of care for patients who had undergone PCI, consistent with the approach advocated by Porter (1) and Porter et al. (15). Future studies should also capture patient health status outcomes as these are also critical outcomes of PCI, particularly for patients undergoing PCI in the setting of chronic stable angina. Future studies should also expand beyond PCI care to more broadly include the outcomes and costs for patients with ischemic heart disease.
Overall, patients receiving PCI in the VA achieved good 1-year outcomes, as shown by mortality rates comparable to randomized controlled clinical trials and statewide registries of PCI (7,16–19). The VA outcomes may be related to improvements in cardiovascular care delivery. Beginning in 2004, VA invested heavily in improving cardiac care by upgrading infrastructure, developed national performance metrics to assess cardiac care quality, and reorganized cardiac care into a hub-and-spoke model. Furthermore, the VA CART program, started in 2004, is a national quality improvement program that proactively monitors care and outcomes for all VA cardiac catheterization laboratories (10).
In contrast to mortality, there was significantly greater variation in health care costs in the year following PCI. Although absolute costs are important in any discussion of health care value, our main purpose was to compare relative costs spent by individual hospitals. These variations suggest there may be significant differences in processes and structures of care during the index procedure and in the year following PCI. A previous VA study found modest variations in efficiency as defined by quality of care indicators for inpatient care, 30-day mortality, and costs, without focusing on a specific condition (8). They found the least efficient VA hospitals had nearly 28% higher risk-adjusted costs than the most efficient hospitals, and one-half of hospitals were 5% to 11% less efficient than ideal. Furthermore, a previous study of the Medicare population found that complications following coronary artery bypass procedure, total hip replacement surgery, abdominal aortic aneurysm repair, or colectomy procedures were associated with higher costs but not with mortality (20). In contrast to the previous study in which there was an association between clinical outcomes and costs, we did not demonstrate such an association between 1-year risk-standardized mortality and costs.
Subsequent evaluations will be needed to identify specific reasons for the cost variations identified, 1 of which may be regional differences in practice patterns. Research is needed to understand the magnitude of variation for each inpatient cost category as well as what types of costs are included in the “all other” cost category. There may be structural factors related to staffing of cardiac catheterization laboratories and/or supply costs (e.g., coronary stents), which are negotiated by individual hospitals. For outpatient care, there will be opportunities to explore differences in follow-up care, some of which may be related to the intensity of care provided, frequency of cardiac testing, and/or a need for noncardiac-related care. Qualitative research, such as structured interviews, may help identify reasons for cost variations. Following these results, we will work with VA operational partners toward implementing interventions to reduce unnecessary variations. For example, in previous work, we have found wide variations in costs of stents across VA cardiac catheterization laboratories and are now pursuing a national purchasing contract to make stent costs uniform across VA (Dr. John Rumsfeld, National Program Director for Cardiology, Department of Veterans Health Administration, personal communication, June 2013).
First, as an observational study, we cannot exclude unmeasured confounding. However, we used robust risk adjustment for observable variables related to both mortality and cost outcomes. The goal was to evaluate variation jointly in outcome and cost in actual clinical practice, and hence, the observational study design is the only one possible. Second, there is likely some variation in attributing specific costs to a cost category at a hospital. However, this should not influence the overall cost comparisons (as the VA has a national cost accounting system), and these differences in methods would not explain the large variability in total costs spent at each hospital. Furthermore, there was strong correlation between utilization and costs, suggesting that variability in costs cannot be solely attributed to differences in cost accounting methods. However, it will also be important that future studies apply standard costs to utilization to assess whether the variation in costs persists. Third, although we were able to account for absolute costs, VA has its own cost accounting system, which may not be applicable to other health care systems or payers and may limit generalizability. However, it is the comparison within the system such as the one we did which provides valuable information to identify opportunities for quality improvement. This is particularly relevant given the push toward ACOs, where there will be increasing pressure to improve quality at lower costs. Finally, we assessed 1-year mortality and costs after PCI, which is a common duration of follow-up for clinical trials and registries of PCI. Although future studies should assess different lengths of follow-up, it is likely that variations in costs will persist.
We assessed variations in both 1-year mortality and costs of care among all VA hospitals that perform PCI. We found substantial variations in costs, with much less variation in patient mortality. This suggests that value of PCI care for the VA may be enhanced through reducing unnecessary variation in costs of care without compromising patient outcomes. This approach to the assessment of mortality and costs for the VA system may be useful for other health care systems and ACOs.
COMPETENCY IN MEDICAL KNOWLEDGE: Most variation of longitudinal PCI care occurs as a result of differences in costs with much less variation in patient outcomes. This suggests that value of PCI care for the VA may be enhanced through reducing unnecessary variation in costs of care, without compromising patient outcomes.
TRANSLATIONAL OUTLOOK: Future evaluation will be needed to identify specific reasons for the cost variations. Qualitative research such as structured interviews may help identify reasons for cost variations.
The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the United States government. Drs. Bradley and Maddox are supported by Career Development Awards from VA Health Services Research and Development. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- accountable care organization
- clinical assessment, reporting, and tracking
- decision support system
- percutaneous coronary intervention
- Social Security Administration
- Veterans Affairs
- VA Information Resource Center
- Received March 28, 2014.
- Revision received October 2, 2014.
- Accepted October 30, 2014.
- American College of Cardiology Foundation
- Levine G.N.,
- Bates E.R.,
- Blankenship J.C.,
- et al.,
- ACCF; AHA; SCAI
- Go A.S.,
- Mozaffarian D.,
- Roger V.L.,
- et al.,
- for the American Heart Association Statistics Committee and Stroke Statistics Subcommittee
- Curtis J.P.,
- Schreiner G.,
- Wang Y.,
- et al.
- Weintraub W.S.,
- Grau-Sepulveda M.V.,
- Weiss J.M.,
- et al.
- Kenney K.M.,
- Marzo M.C.,
- Ondrasik N.R.,
- Wisenbaugh T.
- Gao J.,
- Moran E.,
- Almenoff P.L.,
- Render M.L.,
- Campbell J.,
- Jha A.K.
- U.S. Department of Veterans Affairs. VA Information Resource Center. Available at: http://www.virec.research.va.gov/Cost/Overview.htm. Accessed January 30, 2014.
- Byrd J.B.,
- Vigen R.,
- Plomondon M.E.,
- et al.
- Peterson E.D.,
- Dai D.,
- DeLong E.R.,
- et al.,
- for the NCDR registry participants
- Porter M.E.,
- Pabo E.A.,
- Lee T.H.
- Bagai A.,
- Cantor W.J.,
- Tan M.,
- et al.
- Garg P.,
- Normand S.L.,
- Silbaugh T.S.,
- et al.