Medicare Bundled Payment Q&A with CMS

Throughout the BPCI application process, Singletrack Analytics and our partners at DataGen have maintained an ongoing dialogue with CMS about issues related to the bundled payment implementation process.  Below are some of the issues discussed along with our interpretation of the response.

Price Determination

DataGen Question:  This question relates to the Empirical-Bayes formula, which will essentially blend the hospital historical price with the region.  I assume that the region is defined as the HRC or the combination of HRCs received by the hospital.  Some of the HRCs cross CBSAs, which are the basis for wage adjustment of the hospital and institutional post-discharge care (an example is HRC 6 Albany NY – Springfield, MA).  For that reason, it seems that the wage adjustment must be neutralized before combining the region with the hospital – if this is not done, a hospital in Springfield which has a 2013 wage index of 1.3 will look significantly more expensive than a hospital in Albany NY which has a wage index of 0.86.  Blending the regional experience with the hospital without taking this difference into account will under pay a hospital in Springfield and over pay a hospital in Albany, NY.  How is this being adjusted for in the empirical bayes blending?

CMS Response:  At this time, CMMI has not identified a final methodology for setting target prices or bundled payment amounts.  Thus, we have not made a determination of whether the Empirical Bayes method will be used in calculating target prices or the region that would be used as a comparator.

We appreciate your comment and recognize the importance of adjusting a regional benchmark for expected differences compared to an individual awardee when applying the Empirical Bayes method.  We are taking this into consideration in developing the methodology.

We welcome your feedback and intend to provide specifics on the details of these calculations as we approach a final methodology.

We need to monitor CMS’ approach carefully and check to ensure that any/all benchmark prices are neutralized for regional differences.  These adjustments are essential to ensuring that any blended price does not include bias that is not due to care delivery (which is why CMS is removing differences due to teaching and DSH adjustments).

DataGen Question:  I am looking over the converged episode documentation sent to our hospital and am having trouble tying to the numbers, beginning with the actual episode counts.  Can you tell me how to proceed with reconciling the numbers provided?

CMS Response:  We have discussed your question about Bundled Payment Episode Packet (“BPEP”) with our contractor and have a response.  The contractor generated BPEP analysis yielded more episodes because they were working with the national data and were not restricted to analysis of particular HRCs as the applicants were. Though the benchmarks were set using the applicant provided HRCs, the BPEP data analysis was based on identifying claims in the national data using the provider CCN, they did not restrict the claims based on the applicant’s HRCs.

As you know, we have been presenting our preliminary payment polices to you during the webinars.  We appreciate the feedback we receive from candidates on those preliminary payment policies.  As we indicated during our webinars, we will need to take these preliminary policies through a clearance process before we can share with you the final policies, including the methodology for calculating target prices.  We understand your need to preview what your target prices would be for the episodes you are interested in testing. 

We are working towards providing you with estimated target prices after the final payment policies are cleared and before you would be expected to sign the participation agreement with CMS.  

CMS has been very explicit in its warnings not to look at claims data beyond what was necessary to set the hospital’s own price.  CMS and its contractor have access to the entire data set and are looking, not only at the whole HRC, but at the whole US when calculating episode prices. 

We agree that this approach is most appropriate because this is how the program will be implemented; however, the full data set should be made available so that hospitals can assess and analyze current care practices in order to plan for and achieve meaningful improvement.  Providers need to understand fully the differences (or lack thereof) in care patterns between patients who reside within the HRC and patients who travel to receive their care at the hospital.

DataGen Question:  Based on your response and on the numbers I am able to calculate for the region, I believe that the way you developed the regional numbers was to use all episodes (regardless of the residence of the beneficiary) for hospitals in the candidate’s HRCs.  For this reason, the episode counts for the region are considerably higher than what I can find by looking only at the 3 HRCs the candidate received.  Is this true or did you restrict the region numbers to only beneficiaries residing in the HRC’s of data that the candidate received?

CMS Response:  Our contractor did in fact use all episodes for hospitals in the candidates’ HRCs regardless of the residence of the beneficiary.  Please let us know if you have any additional questions.

CMS has since reversed itself on this response.  We continue to press for the final answer to this question, because we are unable to match CMS’ episode counts.

DataGen Question:  It seems that the adjustment that uses regional prices should be applied based on either the case count or coefficient of variation for an individual DRG occurring for an individual provider. For example, a provider having only five cases in DRG 469 might have an adjusted price for that DRG, but not for DRG 470 in which that provider had 500 cases. The price for DRG 470 would be entirely based on that provider’s historical cost. Is that similar to your thinking on this approach?

CMS Response:  You are correct that the Empirical Bayes method uses both the number of observations and variation in calculating the provider weight (B).  In the case you describe below where a provider has 500 cases for MS-DRG 470, the provider weight would be very close to 1.

This clarifies our understanding of CMS’ intent for creating blended prices; we continue to press for a more straightforward approach.  CMS’ statement that “500 cases would yield a provider weight very close to 1” seems to imply that every episode price  will have some degree of regional price blending – there are extremely few, if any, hospitals with over 500 episodes in a single MS-DRG.

Outliers

DataGen Question:  I understand that the outlier trim points are the 5^th and 95^th percentiles.  My questions are:

1. Do you set the 5^th an 95 using the region and then apply that to the hospital’s data or do you set the 5^th and 95^th using the hospital data?
2. If using the region, do you use the region that the hospital is geographically located in or the set of HRCs the hospital received for analysis (an example is a New York City hospital who received data for 5 HRC’s – do you use all 5 HRCs or only the HRC the hospital is physically located in).

CMS Response:  The 5^th and 95^th percentiles, as well as the geographic benchmarks, are calculated for the HRCs that were associated with the episode-initiating organization in the submitted application.

This clarifies our understanding and is reasonable.

DataGen Question:  As we understand it, 5% of episodes at each end of the cost distribution for the HRC would be classified as outliers, and episodes having costs above or below the respective cutoffs would be limited to the 5% to 95% cutoff amounts. This means that 10% of cases would be limited (5% by the lower limit, 5% by the upper limit), not 10% of cost because of Pareto's law. In the example below showing all of the episodes in an HRC for DRG 470, 5% of episodes fall outside of the limits, but 3% of cost is below the lower limit and 13% of cost is above the upper limit. We assume that this is the intended behavior, but wanted to verify that our assumption is correct.

CMS Response:  You are correct in your description of the adjustment for outliers that we have described.  As the distribution of episode payments is skewed to higher values, this results in lower overall target prices.  Our preliminary policy is to apply this both in calculating actual payments and target prices during the reconciliation process.

We have concerns about this approach to the outlier trimming and will be examining the data for potentially adverse effects.

DataGen Question:  In a previous question, we asked if we understand the outlier calculation to mean that 10% of cases would be limited by the limits (5% by the lower limit, 5% by the upper limit), not 10% of payments because of Pareto's law. The answer from CMS was that we did understand the calculation correctly. 

However, the Geographic Benchmarking in the Bundled Payment Episode Packet distributed for yesterday’s call states that: “Based on summary statistics, RTI identified the 5^th and 95^th percentile of episode payments. . .”  This conflicts with the answer we previously received.  Can you tell us which statement is correct - are we looking at the top and bottom 5% of episodes/cases or top and bottom 5% of payments? 

CMS Response:  The percentiles are established by the number of cases within a region.  As an example, if 100 cases occurred for an individual MS-DRG in a region, the 5^th least expensive case would establish the 5^th percentile and the 95^th most expensive case would establish the 95^th percentile.  The 5 cases greater than the 95^th percentile would represent greater than 5% of total expenditures, which is why we stated that the percentiles are determined based on cases and not total payments.

These percentiles would be used to trim an individual provider’s cases, regardless of the number of cases that fall outside of the bookend percentiles.

The establishment of outlier trim points can minimize risk, but may also minimize opportunity, depending upon the number of cases within a particular DRG for the HRC.  10% of episode in a high-volume DRG is a lot of volume to exclude and may not always be appropriate.  We will be examining the impacts of this.

DataGen Question:  In the RTI reports, it appears that the hospital stay payment and the post-discharge component payments are each subjected to the episode outlier limits, as is the total payment.  But unless there are separate outlier limits computed for hospital stay and post-discharge period, this comparison would be meaningless because the total payment 95% outlier limits would almost always exceed each of the component payment amounts (each of which will be smaller than the total).  Similarly, the component payments would frequently exceed the 5% total payment outlier limit, so applying this limit would understate these costs. 

This presentation suggests that different outlier limits are calculated for each of the components (inpatient stay and post-discharge period), and also for the total payment, yet there’s no reference anywhere else to outlier limits on the components. Were such limits computed strictly for this report but won’t be used elsewhere?  Or am I missing something?

CMS Response:  You are correct that in the analyses, the hospital stay and post-discharge component payments are subjected to trimming.  Separate trimming thresholds were calculated for these components individually to avoid the issue you note below.  This was provided in this report for informational purposes only and would not be applied during the program.

During the program, trimming would occur for the entire episode and so the total episode payment would be most relevant for understanding this policy.

This is confusing.  We will be replicating the CMS stated methodology, as opposed to the methodology used in the RTI reports, to determine the impacts and implications of the outlier trimming.

Episode Attribution

DataGen Question:  This question relates to the “Multiple episodes per beneficiary” issue discussed in today’s seminar.  Please confirm that the information below is correct, or let me know of any errors in my interpretation.

This relates to a beneficiary who is initially admitted to a participating hospital in a participating MS-DRG, and is subsequently admitted within the episode timeframe to another hospital.  The outcomes depend on whether the readmission hospital is participating in BPCI for the readmission DRG, and whether the readmission DRG is related to the anchor DRG.

* Being a BPCI participant isn’t sufficient; the provider must also be participating for the readmission DRG.  Otherwise the readmission hospital is no different than a non-participating hospital.

CMS Response:  We can confirm your interpretation.

This confirms our understanding.

DataGen Question:  One issue that we've been addressing in several ways is the assumption of which providers are participating in BPCI. This is important because a potential initiating admission for one provider may in fact be a readmission in episode initiated by another provider, if that provider is a BCPI participant. Until we know which hospitals are participating, however, we must make one of the following assumptions:

• All other hospitals are participating in BPCI. Any admission that cannot be a readmission for another episode becomes an anchor admission for the hospital. But any admission occurring within 90 days of previous admission for the same beneficiary gets attributed to the hospital in which the previous admission occurred, and does not initiate an episode at the second hospital.
• No other hospitals are participating. Any admission at our hospital will initiate an episode regardless of any other previous admission for the same beneficiary at another facility.

(Both assumptions assume that all of the admissions are for valid MS-DRGs and all other criteria are met.)

Ultimately neither of these assumptions will be correct, since it's likely that some (but not all) other hospitals within the same region will participate, and that admissions to one hospital may actually be readmissions for another participating hospital’s episode.

In the document e-mailed for today’s call, it suggests that an episode initiates with inpatient stay "in an awardee hospital". Since we don't know the awardees are, what should we assume?                            

CMS Response:  We understand the concern you are raising. At this point, we would suggest assuming your second scenario, that any admission at your hospital will initiate an episode regardless of any other previous admission for the same beneficiary at another facility.

This makes it extremely difficult to analyze the data set for more than one hospital at a time.  There is no one correct way to analyze the full data set that will yield results to simulate the interrelationships between providers and beneficiaries. 

Inclusion/Exclusion Criteria

DataGen Question:  The file BPCI_Episodes.xls distributed last week included lists of excluded DRGs and excluded part B services.  Do the excluded Part B services apply to all settings other than acute inpatient (SNF, Home health, OPD, carrier, inpatient rehab, LTCH) or ONLY to services paid under part B? 

CMS Response:  Exclusion lists for primary ICD-9 codes apply only to Part B services during the post-discharge period for Models 2 and 3.  With the exception of readmissions, which are excluded based on MS-DRG, other Part A services are not excluded from the episode.  As discussed during Session One of the Candidate Awardee Webinar Series, these services are generally comprehensive and are considered related to the episode regardless of the primary diagnosis code.

DataGen Question:  If a physician claim has an included ICD9 code but occurs during an excluded readmission, should the physician claim be included?  Conversely, f a physician claim has an excluded ICD9 code but occurs during an included readmission, should the physician claim be included? 

CMS Response:  In Models 2 and 3, Part B services are excluded based on primary diagnosis code regardless of whether they occur during an inpatient admission.  In Model 4, Part B services are included only if they occur during an included readmission.

This clarifies our understanding of how CMS handles excluded diagnoses, but we are still a bit unclear as to which event trumps when a physician provides a service during an excluded readmission.

DataGen Question:  Does the episode initiating stay (in the column called “CANDIDATE AWARDEE SPECIFIC: Mean Episode- Initiating Hospital Stay Payment.’) include the physician Part B payments or only the MS-DRG payment?

CMS Response:  In Models 2 and 4, physician Part B payments during the index admission are included in the episode.  No Part B payments are excluded during the index admission.

We understand that these Part B payments are included in the episode, but we remain unclear where CMS “puts” them when they break the episode into two components.

DataGen Question:  In the Construction of the Bundled Payment Episode Packet document discussed on today’s call is the following bullet: “Note that non-assigned physician claims are included in this analysis.  However, these claims will be excluded from both target pricing and payment reconciliation during the implementation of BPCI.”  My question is, what is an assigned vs. non-assigned physician?  Does this affect Model 2?

CMS Response:  ‘Non-assigned claims’ is a phrase used to refer to claims from physicians who do not accept assignment in the Medicare program.  These physicians have elected not to be paid the rates that are set based on the Medicare Physician Fee Schedule, electing instead to bill their beneficiaries and receive payment from those beneficiaries directly.  Non-assigned physician claims are a very small percentage of claims, and we do not believe you would be able to identify them in the data files that you have available to you.  However, this guidance would only apply to physician claims, which would be relevant to Model 2.  Claims that are submitted for services rendered by physicians who have not accepted assignment under current Medicare rules will be excluded from the BPCI program entirely.  They will not be included in the setting of target prices or bundled payment amounts, nor will they be included when reconciling against target prices.

Interesting, but we cannot match CMS’ logic with the data set they provided.

DataGen Question:  During today’s call, in the explanation of excluded readmissions, you detail how to identify psych hospitals to apply the drg exclusions.  I thought all psych hospital services were excluded?

CMS Response:  Services at Inpatient Psychiatric Facilities (IPFs) will be considered readmissions for purposes of the Bundled Payments for Care Improvement initiative.  They will be excluded by MS-DRG (based on the MS-DRG listing provided in the BPCI_Episodes.xlsx table provided).  They are not wholesale excluded – rather, they are excluded if they are unrelated, based on the lists provided.

Interesting nuance that may have an impact on episode pricing.

DataGen Question:  In the Construction of Bundled Payment Episode Packet document, you identify psychiatric hospitals as those with provider numbers 4000-4499.  How should we treat subproviders (those with third digit of M or S?  Also, do you consider providers with third digit of T or R as rehab providers or something different?

CMS Response:  Providers with an M or S as the third digit of their provider number are psychiatric units within inpatient facilities, and should be considered in the same way as freestanding psychiatric hospitals (provider numbers 4000-4499).  Providers with a T or an R as the third digit of their provider number are rehabilitation units within inpatient facilities, and should be considered in the same way as freestanding inpatient rehabilitation facilities (provider numbers 3025 – 3099).

This confirms our understanding.