Choosing Episodes in BPCI Advanced

The BPCI Advanced program offers participation in 29 clinical episode types initiated by an inpatient service, as well as three episode types that are initiated by outpatient services. Participants can choose to participate in any or all of these episodes. Since each of these episodes types has widely different characteristics, correctly evaluating the opportunities for success in these episodes will be a major factor in achieving that success. Many hospitals won't delay their episode selection process until the CMS BPCI advanced data is released, and will instead begin their review using models developed from the Medicare standard analytic files, such as those produced by our partners at DataGen. Others may request their historical data from CMS when they submit their BPCI advanced location, and will contract for episode analysis services to perform the types of analyses described below. This document describes the process that Singletrack Analytics uses to select episodes, and our episode evaluation toolset includes the analyses described below.

Overview

Here's an overview of the episode selection process. First, buy-in from the clinical teams associated with each episode under consideration must be achieved. Physicians and other clinical team members must be enthusiastic about the opportunities to achieve success in their clinical areas. Episodes must also have sufficient volume to have sufficient savings opportunity, and also to avoid large range of variations in results. Sufficient "actionable cost" that can be affected by care management must exist within the episode to allow reductions in episode cost to exceed the CMS discount. If readmissions are a significant part of episode cost, the participant must have an effective readmission management strategy. Finally, the CMS target methodology must not disfavor the episodes being selected.

How Financial Success Occurs in BPCI Advanced

Like all similar programs, financial success in BPCI Advanced occurs when episode costs are lower than target amounts. (“Costs” in this article are defined as payments by the Medicare program to providers who provide services to patients within the episode timeframe.) The key difference among these programs, which differentiate the types of providers who can succeed, is the basis on which the targets are computed. In the BPCI program targets are computed based on each provider’s historical cost. Providers who reduce their costs will usually succeed financially – the key in this program is to become better than you were in previous periods. Your cost efficiency relative to other hospitals is generally irrelevant.

By contrast, the targets in the later periods of the CJR program are computed from the average episode costs of other hospitals in the same geographic region. They are not based on the provider’s historical costs. Therefore, the key to financial success in CJR is to be better than the other hospitals in your region.

Note the significant difference here. In BPCI the participants with the greatest potential for success are those whose initial costs are high, and for whom improvements are relatively easy. However, in the later years of the CJR program the most successful participants are those with low costs, since those costs will likely be lower than the regional averages and hence the targets.

The target-setting process in BPCI Advanced is based largely on each participant’s historical cost with some adjustments for patient mix and hospital characteristics. Therefore, the most successful participants will be those who have high episode costs with significant opportunity for improvement. Already-efficient providers will have a more difficult time creating savings and thereby achieving financial success.

Step 1 – Finding Clinical Champions

But where does one start in selecting among 32 different opportunities? The first place is not within the data, but within the clinical teams involved in treating and managing these patients. Success in any of these bundles requires significant revisions to the care provided to these patients, and clinical team members must be enthusiastic supporters of the initiative for success. In particular, without a "physician champion" actively involved in the care redesign process, success is highly unlikely. Therefore, the first step in evaluating episodes is to find physicians and their teams who are excited about this prospect. Without this support in a particular clinical area, episodes initiated in those clinical areas should not be considered for participation. Identifying clinical areas in which clinical support exists is the first step in episode selection.

But let's assume that physicians in several clinical departments have expressed interest in participation. Now, it's time to look at the data.

Step 2 – Evaluating Episode Volume

Having sufficient episode volume is the first triage point for episode selection. Sufficient volume is important for several reasons. First, it's difficult to get a clinical team involved in infrequently-occurring episodes. Busy clinicians aren't going to spend time on a small number of patients.

In addition, the episode volume must offer sufficient opportunity for savings to cover the cost of participating – the cost of care management, project management, analytics, finance, and other areas. Without sufficient volume even a relatively high savings percentage would not generate enough savings dollars to pay for these costs of participation.

Finally, large amounts of variation inhibit the ability to properly measure clinical indicators. For example, when readmission rates vary across quarters by 15% due to randomness, it's difficult to identify a 5% reduction in readmission due to care management. Small volume creates significant variation; larger volume reduces that variation.

Financial Risk

Low volume also creates large amounts of variation in cost and clinical indicators across episodes. This variation creates significant financial risk for the participating organization. Average episode costs vary randomly, with some episode types having higher inherent variation than others, and this variation can cause significant swings in financial results.

The graph below shows the "coefficient of variation", which is the standard deviation divided by the mean, for major joint replacement and congestive heart failure episodes. The horizontal axis indicates the number of episodes for a particular time period (generally a calendar quarter), and the vertical axis indicates the amount of quarterly variation in episode cost expected due entirely to random variation.

Figure 1 Episode Cost Variation by Volume

For example, a participant having 50 CHF episodes in each quarter could expect the average episode cost to vary across quarters by about 26% (twice the 13% coefficient of variation, equal to about $4000) from quarter to quarter. At an overall average episode cost of about $15,000 for CHF episodes, this random variation could create quarterly average episode costs varying from about $11,000 up to $19,000 per episode, so at 50 episodes per quarter this variation would be plus or minus $200,000 per quarter. Since targets are constant, the surplus or deficit each quarter would vary by that amount due entirely to random selection of patients.  That amount of random variation may be too large for smaller participants.

Major joint replacement episodes have a lower amount of variation, as noted by the orange line on that graph. Nonetheless, sufficient volume in these episodes is also necessary to avoid wide variations in episode costs.

We generally recommend a minimum episode volume of 100 episodes per year to achieve sufficient statistical and financial stability. Episodes having a lower volume may be too unstable to effectively participate in bundled payment programs. Identifying episodes with sufficient volume is the second step in episode selection.

Reducing risk by combining episodes

A further strategy to reduce variation is to participate in several clinically-similar episode types having different amounts of variation; for example participating in AMI, cardiac arrhythmia and CHF episodes. This allows participation in lower-volume episode types by creating a larger population of episodes the combination of which will have lower variation. Each of these episode types must meet the other criteria described in this article (clinical involvement and actionable cost), but the combination of episodes may allow participation in an episode type that doesn’t meet the volume threshold. This issue is described more completely in this Singletrack Analytics article.

Step 3 – Evaluate Actionable Cost and Readmission Rate

Four primary metrics are used for evaluation of opportunity in bundled payment. First, sufficient episode volume is necessary to create financial stability and critical mass for care management, as described above. Since not all episode costs can be affected by care management, the actionable cost percentage shows the percentage of episode cost that can actually be changed. Finally, since readmission reduction is more difficult to achieve than reductions in institutional post-acute cost, the readmission percentage is a critical element in episode selection. Finally, the discount reduction needed shows the percentage reduction in actionable cost necessary to cover the CMS 3% discount on the entire episode cost, and therefore to avoid a deficit.

Actionable cost

The cost drivers of episodes can be seen in Figure 2 Major Joint Replacement - Hospital A below. Each vertical bar is an individual episode with the cost components of that episode shown on the stacked graph. Noticeable on the graph is that almost all patients receive home health services, about 60% go to SNF and that readmissions, while present, don’t represent a major component of cost. This type of graph gives guidance to the care management strategy that will be successful in these episodes; namely that focusing on post-acute SNF and HHA services will create the greatest benefit. Furthermore, the longer green bars at the right side of the graph indicate long SNF stays, showing that some benefit may be achieved by working with SNF providers to reduce lengths of stay.

Figure 2 Major Joint Replacement - Hospital A

Contrast the above graph with the episode composition graph for CHF below. Note the large predominance of red in that graph indicating a high rate of readmissions with commensurate contribution to episode cost. In contrast to major joint episodes, readmissions must be reduced in CHF episodes to have a significant effect on cost.

Figure 3 Congestive Heart Failure

In both types of episodes there’s significant “actionable” cost, with “actionable” defined as costs other than the index admission (which is paid based on the DRG at a fixed rate and can’t be changed) and “professional” costs that are largely physician costs and rarely are the target of care management activities. Some episodes have high DRG payments and therefore high index admission costs, or have low post-acute costs (or both) and therefore have little opportunity for cost reduction. In Figure 4 Cardiac Valve below note the high percentage of the episode cost that’s composed of the index admission DRG payment and the surgeon fee and other professional costs relative to the post-acute costs; therefore only a small percentage of the episode cost can be affected by any care management activities. Since the CMS 3% discount is applied to the entire episode cost, significant reductions in the relatively small post-acute costs would be necessary to cover the discount amount before any savings could be achieved. So the “percent actionable” cost is one significant metric to be used in the evaluation of episodes.

Figure 4 Cardiac Valve

The percentage of episode costs that are actionable also varies significantly across providers, as depicted in Figure 5 Major Joint Replacement - Hospital B. This hospital has significantly revised its treatment protocols to minimize SNF stays and even sends about 20% of its patients home without home health services. Because cost reduction is the key to success in BPCI Advanced, this episode type may not be a good candidate for participation because the “low-hanging fruit” has already been picked, and achieving further cost reductions may be difficult.

Figure 5 Major Joint Replacement - Hospital B

Comparing cost drivers among episodes

Visualizing these cost drivers across episode types often facilitates understanding the differences in opportunities. The graph below illustrates the differences between AMI (in which in index admission costs are low and institutional and readmission costs are relatively high) with CABG (in which index admission and professional costs account for 80% of the total episode cost). And while AMI and MJR episodes have similar amounts of actionable cost, readmissions account for more than half of the cost for AMI, yet a small percentage in MJR. From this visualization the relative opportunities, and care management activities necessary to achieve those opportunities, can be seen.

Figure 6 Episode Cost Driver Comparison

Discount reduction needed

The “discount reduction needed” metric computes the percentage of actionable cost that must be reduced to exceed the CMS discount that’s applied to the entire episode cost. This metric is a function of the percent actionable cost, but is shown in a way that highlights the relative difficulty of achieving a surplus in some episodes that have a low percentage of actionable costs. In PCI episodes, for example, actionable costs need to be decreased by 8% to overcome the 3% CMS discount. And given that most of the actionable cost in PCI is in readmissions that are difficult to manage, achieving the necessary amount of cost reduction may be unrealistic.

Figure 7 Discount Reduction Needed

Readmission rate

The other major episode evaluation metric is the readmission rate of the episodes. This is because readmission reduction is much more difficult to achieve than reductions in other types of post-acute costs such as SNF. While SNF costs can be reduced through applications of broad guidelines for post-acute care, readmissions almost always require individual physician attention at the patient level. This is significantly more resource-consumptive and often less effective than management of PAC institutional costs, and therefore a high readmission rate indicates more limited opportunity for success. Additional analysis into readmissions is useful when considering participation in episode types having high readmission rates; for example it’s useful to look at the DRGs of the readmissions and the times within the episode that they occur. This may provide guidance as to the ease or difficulty of reducing the readmission rate.

Comparing episodes

The graph below shows both of these metrics for several higher-volume episode families for a specific hospital. This graph shows that MJR episodes have a high percentage of actionable cost and a low readmission percentage, indicating significant opportunity for cost savings. CHF episodes have a similar percentage of actionable cost but a much higher readmission rate indicating that cost reductions will be more difficult to achieve in these episodes. Finally, Percutaneous Coronary Intervention episodes have a relatively low percentage of actionable cost and a high readmission rate, indicating that achieving success in these episodes will be difficult.

Figure 8 Episode Selection Dashboard

This process is the third step in episode selection.

Step 4 – Target Review

The final step after episode opportunities have been identified is to determine the effect of the CMS target setting methodology on these episode types. In the BPCI program in which targets were based on historical episode cost, there was a direct financial relationship between episode cost reduction and financial success. However, due to uncertainties in the effects of the risk adjustment process used by BPCI advanced, there may be episode types that are favored or disfavored by the risk adjustment methodology. This occurs in the Oncology Care Model, in which certain cancer types have a significantly higher proportion of financial successful or unsuccessful episodes. Since CMS has not yet released the details of the risk adjustments for targets, we don't yet know whether these biases in target computation will exist. Once the targets have been released, our analytics team will perform thorough analyses to identify areas in which the risk adjustment process itself has a significant effect on the financial results. BPCI Advanced applicants should likewise compare their historical costs to the targets to identify any areas in which targets significantly create significant advantages or disadvantages in episodes. However, targets themselves rarely create opportunities for success – those opportunities only arise from evaluation of the clinical characteristics of the episodes and the potential for care management activities to affect those costs.

Step 5 – Putting It All Together

The final step in episode evaluation is an overall look at all episodes to be included in the program. An example of such a selection is shown below for group of cardiac episodes. Among these episodes it's notable that the two highest volume episodes have extremely high readmission rates, so readmission reduction will be key to success. The overall coefficient of variation is relatively high, although the inclusion of AMI reduces the average slightly. All episodes have a high percentage of actionable cost, allowing PCI episodes to be included even though their actionable percent is low. The overall discount percentage needed is about 5%, which is about average for most episodes. The combined episode volume exceeds 1000 per year, which should create sufficient critical mass for clinical teams. Therefore, this combination of episodes represents a reasonable choice for participants wishing to focus on cardiology episodes. Similar groups of selections of episodes could be made for group having a different clinical focus.

What about Benchmarks?

Some applicants focus on “benchmarks” when evaluating episodes, but this can be unhelpful for several reasons. First, the natural step when evaluating current performance against a benchmark is to hope for performance that’s better than the benchmark, and perhaps select episodes in which you’re already performing well. As noted above, though, being historically efficient is a deterrent rather than an opportunity in a program using historical targets. Choosing episodes in which you perform poorly against benchmarks is the right strategy, but it’s counterintuitive.

Also – benchmarks don’t provide as much useful information about opportunities as episode composition graphs do. While benchmarks provide gross numerical comparisons of metrics, episode composition graphs lay out a complete picture of the cost drivers of each episode, providing a far better picture of the exact areas in which opportunities exist.