Hedging Readmission Risk in Bundled Payment

This article describes an approach for classifying these DRGs as “unrelated”, and therefore for excluding their costs from the episode bundle.  This approach is not based on clinical considerations, but rather on the statistical and financial issues that relate to their random behavior. 

Three caveats should be kept in mind when reading this article.  First, as noted above, the viewpoint shared here is financial, not clinical.  Obviously, clinical issues are paramount in properly managing and reducing unnecessary levels of readmissions, and this article contributes nothing to that effort.  Rather, its goal is to deal with the financial structure of payment for readmissions, and to propose an approach that maximizes the opportunity of financial gain (and minimizes the changes of loss) from true clinical improvement, rather than from random variations in patient mix that occur over time.

Second, this approach is structured for the Medicare Bundled Payment for Care Improvement (BPCI) initiative, and specifically the methodology used to set the bundled payment episode price.  It may not be applicable to other bundled payment systems that do not use historical readmission data for this purpose.

Finally, no attempt is made to present statistically significant results.  The results listed below are anecdotal, reflecting the experience of several hospitals, and are expected to be representative of the situations that other bundled payment participants may find.

Under the BPCI initiative, the episode budget is computed based on the utilization of services, and CMS payments for those services, for selected "index" DRGs during the first nine months of 2009.  All services provided during that period, unless specifically excluded as the “unrelated”, are included in the episode price.  The applicant has the opportunity to propose to exclude DRGs for readmissions, and diagnosis codes for other services, that it believes are unrelated to the index DRG of the episode.  The costs of those unrelated readmissions are not built into the episode bundle amount developed from the 2009 base period costs, and occurrences of those services will not be charged against that budget in subsequent years.

It’s important to note that the costs of readmissions that are not designated as “unrelated” are built into the episode price.  Thus, a recurrence of those same readmission DRGs would be neither an financial advantage nor disadvantage to the provider organization.  Excluding readmissions as being unrelated, however, removes their costs from the episode budget, thereby lowering the budget amount.

For most index DRGs, there are a number of “readmission” DRGs that commonly occur multiple times within a year.  For example, COPD patients are frequently readmitted in COPD DRGs.  Patients with congestive heart failure are frequently readmitted with CHF DRGs.  These DRGs often represent a primary opportunity for reducing unnecessary costs.  They are unlikely to be excludable as unrelated.

Then there are the “singleton” readmission DRGs that only occur once within a year.  Although each readmission DRG only occurs once, there are many such DRGs within an index DRG.

An example of the DRGs for readmissions for index DRG 470 is show in Table 1 at the end of this article.  In this table, 5 DRGs occur more than once.  These readmission DRGs generally have consistency across years, and targeting them for clinical intervention may be successful.  There are also 19 “singleton” DRGs that occur only once. 

The table below shows for four index DRGs, the total number of readmission DRGs (column 1) and the number that are singletons (column 2).  In all cases the singletons represent more than 50% of all readmit DRGs.

DRGs that appear as singletons generally do not reoccur in the subsequent year.  Column 4 in the table below shows the number of singleton DRGs that occurred in both 2008 and 2009 for the respective index DRGs.  As shown below, these percentages are generally low.

As noted above, the lack of repetition of singleton DRGs suggest that these DRGs may behave in a random manner.  Therefore, let’s assume that for any given index DRG there are 100 different potential singleton readmission DRGs, 10 of which will randomly occur in any given year.  Believing it is minimizing risk, a provider may attempt to classify many of the 10 singleton DRGs occurring during the 2009 base period as being unrelated.  (CMS is unlikely to allow this, but for purposes of this example we will assume that it will occur.) In this case, none of the cost of the singleton DRGs is built into the episode price.  If those same 10 DRGs reoccurred in a subsequent period, they would not be counted against the bundled payment budget and would have no financial effect.

However, if 10 different DRGs that had not been classified as unrelated occurred during a subsequent period, there would be no cost built into the episode price to cover them.  This would create a deficit in the bundled payment model caused primarily by random variation in the occurrence of the singleton DRGs.

Therefore, assuming that the occurrence of the singleton DRGs is random, the provider would be better off leaving them in the base period, rather than reclassifying them as unrelated - even if they appeared to be clinically unrelated.  In this way, cost would be built into the episode to accommodate the singletons that would occur in each year.  Of course the mix of these DRGs (and therefore their cost) could differ from those for the base period, but these costs would be partially offset by the presence in the episode cost of the base period singleton DRGs that did not occur in the subsequent period.

Conversely, the best mathematical strategy for provider (although impractical, and unlikely to be allowable by CMS) would be to exclude as unrelated all DRGs that did not occur during 2009.  In this case, the singletons that did occur during 2009 would be built into the base episode cost, but no singleton DRGs occurring in subsequent years would be counted against the episode budget because they would be unrelated.  Therefore, the provider would be financially ahead by the costs associated with the 2009 singletons.

The strategy above may be successful for readmission DRGs having approximately the same cost.  However, singleton DRGs with extremely high cost may occur occasionally during a base or subsequent period.  If these DRGs do not occur during the base period, it is generally wise to exclude them, since their occurrence in a subsequent period would likely exceed the singleton costs in the base period.  The possibility of these DRGs occurring can be approximated by looking at the incidence of readmissions for the entire HRC for the index DRGs under consideration.  Using this much larger sample, high cost DRGs that are clinically unrelated to the index DRG can be more easily identified and excluded as unrelated.

But what if one of those high-cost DRGs occurs during the base period?  This occurred for one of our clients that had a DRG 3 (ECMO or tracheostomy care) readmission occurring during the 2009 base period, with a cost of about $115,000.  In this case it is probably advisable not to exclude this DRG as being unrelated, since its significant costs will be built into the bundled payment rate.  Although this leaves the organization at risk if two of these cases occurred as readmissions, those probabilities are significantly smaller than the probability that none of these cases would occur.

Readmission DRGs that occur with relatively high frequency should be included in a clinical intervention plan, and evaluation of their classification as “unrelated” should be made based on clinical criteria.  However, the same decision for DRGs occurring with low frequency and inconsistently across years should incorporate the statistical and financial effects of this classification to avoid allowing randomness to adversely affect financial results.