Chris Birkmeyer, M.S.; Luis Leon; Matthew Pierce, M.S.; Karishma Sekhon, Ph.D.; Douglas Staiger, Ph.D.
Managing risk and succeeding in value-based reimbursement programs such as the Centers for Medicare & Medicaid Services (CMS) Bundled Payments for Care Improvement Advanced (BPCI-A) Model requires clarity, control, and confidence. Three keys to achieving desired program results include appropriate risk/episode selection, performance monitoring, and clinical management.
This report focuses on the first key: Episode selection and specifically episode selection based on challenges and opportunities in target prices. It shows how current and prospective BPCI-A participants can manage risk, maximize reimbursements, and optimize overall results in the program through appropriate bundle selection for future years.
Because success in BPCI-A is measured by the degree to which a participant can beat a target price while delivering quality care, current and prospective participants should take steps to understand the positive or negative effect BPCI-A target price methodology could have on their program results.
CMS designed the program to encourage broad participation with the intention to reward improvement over time. As CMS can rely only on available data, it is understandably bound by a hospital’s past performance in setting target prices. For BPCI-A, it calculated target prices based on historic performance using a hospital’s claims data from 2013 through 2016.1 One of the underlying assumptions made is that the calculated average is truly indicative of a hospital’s performance in a given episode. However, the cases used to set a hospital’s target price do not always represent that hospital’s reality. The past does not predict the future.
CMS has included risk adjustment in the BPCI-A methodology to account for patient case mix and severity that is outside of providers’ control.2 But it has not considered how two key factors affect a hospital’s true performance in a given episode: (1) Variability from limited sample sizes (2) Hospital-specific trends through and beyond the target price performance period.
Current and prospective participants that understand and account for these factors by using statistical analysis in their own program assessments can establish more accurate estimates of how favorable or unfavorable particular bundles are to their hospital. This allows participants to maximize results in the program by opting to participate in bundles where target prices are favorable from day one. For example, hospitals participating with a favorable price in the Cardiac Valve bundle could expect to earn $4,484 or more per episode before even considering any changes to their current delivery of care, per ArborMetrix’s analysis of expected per episode Net Payment Reconciliation Amount (NPRA) (See Figure 2).
To help current and prospective program participants evaluate bundle favorability, ArborMetrix examined the combined effects of the current BPCI-A Model target price methodology. Our analysis accounted for variability due to limited sample sizes and current trends in hospitals’ cost of care across all BPCI-A participants and all bundles in which they were eligible to participate during Model Years 1 and 2. Specifically, we measured favorability of target prices due to the sample size used for BPCI-A target pricing and hospital performance over time.
Each blue dot represents a hospital eligible to participate in the bundle. The vertical Y-axis shows the hospital’s CMS target price for the bundle. (Note: Target prices presented are based on “standardized dollars” that have already eliminated variation attributed to contractual reimbursement rates). These begin at $15,000 and range up to $50,000. The horizontal X-axis shows each hospital’s baseline case volume for the bundle. Moving from top to bottom along the Y-axis and to the right along the X-axis, hospitals begin to cluster and form a funnel shape. This shows how as baseline volume increases, it is easier to reliably determine a true episode cost. The spread of low volume hospitals up and down the Y-axis highlights the difficulty in setting target prices based on a small or unreliable sample size.
Examining Variability from Limited Sample Sizes
Some BPCI-A participants’ target prices were set based on low total case volume, or a small sample size (as few as 41 cases for some). In these instances, the likelihood that the sample represents a hospital’s true efficiency is extremely low. This makes it difficult for CMS to calculate a truly accurate estimate of a hospital’s total cost of care for an episode.
For example, Figure 1 examines target prices and baseline volumes of hospitals eligible to participate in the Major Joint Replacement of the Lower Extremity bundle. This bundle has the largest spread in expected program earnings and losses. It illustrates the challenges in setting target prices for participants with low case volume.4
BPCI-A participants can account for sample-size-driven variability (i.e., chance) by adjusting for statistical uncertainty. ArborMetrix uses a proven statistical methodology called Reliability Adjustment,3 which leverages all available data including performance in related bundles to quantify how a hospital would perform if the case volume were unlimited. The result is a more accurate estimate of a hospital’s true cost of care for episodes in a given bundle.
Comparing these estimates to the target prices set by CMS reveals situations where randomness in the samples used results in a target price being higher (more favorable) than a hospital’s true per episode cost in this bundle. In this situation, a hospital could expect to gain earnings on a given episode from the start of the program, before even considering changes to care.
Accounting for Hospital Performance Over Time
Although CMS’s methodology does control for trends attributable to regions and other relevant characteristics, it does not account for hospital-specific trends in costs over time. ArborMetrix leverages additional data sets including more recent information such as the CMS Limited Data Set. Using the same statistical methodology, we forecast Reliability-Adjusted bundle-specific trends in hospital efficiency and calculate their effects on total episode costs.
By assessing how these metrics, key to BPCI-A success, have been trending historically, we can project their values during the BPCI-A performance period. This trending forward indicates how much savings or loss can be expected without any further clinical intervention.
By accounting for a hospital’s natural upward or downward motion in cost of care, a hospital can gain a sense of whether it is facing a headwind or a tailwind when managing episode costs for a given bundle.
What we examined: ArborMetrix’s statistical methods and analysis result in a more reliable estimate of the total cost of care for each hospital in the country for every bundle in which they are eligible to participate.
By comparing this more accurate estimate to the CMS target price for each bundle, we calculate the dollar amount each hospital can expect to gain or lose per patient in every bundle in which they are eligible to participate (absent other interventions to change cost experience). In order to demonstrate the range of expected savings, we look at the top and bottom 10th percentile for each bundle. This analysis gives a sense of the financial ramifications of a favorable or unfavorable target price on a per bundle basis.
This histogram shows the number of BPCI-A eligible hospitals in each per Episode NPRA bucket. Buckets represent the dollar amount participants in this bucket can expect to gain or lose on each episode. This histogram includes each participant’s Per Episode NPRA for the Cardiac Valve bundle, where we found the greatest difference between per episode NPRA in the 10th and 90th percentiles. Hospitals participating with a favorable price in this bundle could expect to earn $4,484 per episode before even considering any changes to their current delivery of care.
What we found: The difference in expected savings per case varies drastically across participants. For example, as shown in Figure 2, comparing the 10th and 90th percentiles in the Cardiac Valve bundle, hospitals with unfavorable target prices will need to improve at least $1,352 per patient to break even, while hospitals with a favorable target price can expect to earn $4,484 per patient before even considering any changes to their current delivery of care.
What this means for individual providers: Current and prospective BPCI-A participants who account for the inefficiencies in CMS’ methodology in their own evaluations and analysis of target prices, episode selection, and performance forecasting have significant opportunities to gain greater savings and reimbursement payments through the program.
By leveraging proven statistical methods to gain a more accurate estimate of their hospital spend on episodes in a given bundle, hospitals and provider groups could realize significant financial savings in bundles they would have not otherwise considered participating, or avoid bundles where they may be fighting an uphill battle.
For example, as shown in Table 1, some Model Years 1 and 2 participants selected unfavorable bundles that might be resulting in situations where, to be successful, they must be able to improve significantly (in some cases more than $2,500 per patient), or risk losing money through participation in this program. On the other hand, participants that selected favorable bundles could expect to earn as much as $5,800 or more per patient before considering any changes to the delivery of care.
Appropriately selecting episodes for future program years using reliable, evidence-based analysis of target price data and performance increases likelihood of program success.
Why this matters to the program overall: CMS’s stated goal of the BPCI-A program is to reduce expenditures and improve quality through a voluntary episode based payment model.1 ArborMetrix’s analysis indicates participants could earn financial incentives in the BPCI-A program before ever making changes to delivery of care.
These opportunities illustrate the challenges inherent in the methodologies used for setting target prices. Adjusting for sample size variability and hospital-specific efficiency trends might represent an opportunity for CMS to set target prices more reflective of true performance at program start. This would more accurately measure true performance improvement and reward participants that can actually reduce expenditures.
There are inherent challenges and limitations in setting target prices for bundled payment programs.
For future Model Years, current and prospective BPCI-A participants have an opportunity to realize significant financial savings by participating in bundles they would have not otherwise considered, or avoiding bundles where they may be fighting an uphill battle to succeed. To do this, they should leverage proven statistical methods and powerful analytics to gain a more accurate estimate of their hospital spend on episodes in a given bundle. Appropriate risk/episode selection is key to success.
ArborMetrix advances healthcare through data science by transforming data into insights for decision-making. Our healthcare analytics solutions deliver clinically-rich and relevant evidence at the level of precision needed to measure and improve clinical outcomes, advance care, and optimize performance.
Whether you are already participating in BPCI-A or are preparing to apply, with or without a convener, ArborMetrix can help you gain more savings in the program and right-size your investment in BPCI-A services. Our unique BPCI-A episode analytics and performance monitoring helps you identify bundles where you will succeed regardless of whether you improve, and then assess your performance in real-time on bundles you do select to improve incrementally and achieve even greater results.
By investing strategically in data and analytics that eliminate uncertainty and bridge the gap between your clinical expertise and bundled payment analytics, you will come away with confidence in your BPCI-A performance.
Learn more at ArborMetrix.com/bpci-advanced.
1. BPCI Advanced. Centers for Medicare and Medicaid Services. innovation.cms.gov. Accessed April 2019.
2. ‘Pricing Methodology for Clinicians and Administrators.” Bundled Payments for Care Improvement Advanced. Centers for Medicare and Medicaid Services. innovation.cms.gov. June 2018.
3. Dimick, J.B., Staiger, D.O., & Birkmeyer, J.D. (2010). Ranking hospitals on surgical mortality: The importance of reliability adjustment. Health Services Research, 45 (6), 1614-1629. doi: 10.1111/j.1475-6773.2010.01158.
4. Analysis performed by ArborMetrix based on BPCI-A participation data and Limited Data Sets from the Centers for Medicare and Medicaid Services.
“ArborMetrix provides us with reliable, data-driven insights to clinical and financial performance of our programs, including BPCI-A. Their ability to integrate multiple data sources and visualize results has changed the game for us – really boosting our confidence to both accurately evaluate risk and drive increased effectiveness in our clinical operations.”
President, Sound Physicians