This post deals with the CMS concept of Behavioral Adjustment (BA), its origin, purpose, validity and impact on home health reimbursement.
In the 2024 final rule, CMS describes the background of the BA. When PDGM was developed, they had requirements to reduce the service period used for payments from 60 days to 30 days and to eliminate the connection between the payment model and therapy visits. CMS refers to the regulatory requirements under CFR Title 42, Section 484.220 as the legal basis for their authority to define the PDGM payment model. Using ChatGPT, here is a summary of these regulations:
“The Code of Federal Regulations (CFR) Title 42 Section 484.220 outlines the computation of payment amounts under the prospective payment system for home health services. Here is a summary:
The section states that the Centers for Medicare & Medicaid Services (CMS) uses the prospective payment system to calculate payments for home health services for beneficiaries who are under a home health plan. This calculation involves five components:
Computation of a base payment amount that involves amounts attributed to six different home health resources groups.
Adjustments of the base payment amount for case-mix and wage index.
Any necessary adjustment to the national, standardized 60-day period payment rate.
Adjustments to account for episodes of care that are outlier cases, which involve either extraordinarily high costs or low costs relative to the average.
Adjustments to ensure that there are no further projected or actual changes in aggregate payments.
The Application of the adjustments stated in points 2, 3, and 5 are adjusted in such a way to result in the same estimated total payments under the sections as would have been made without those adjustments. Hence, the adjustments ensure budget neutrality.”
It is the last item, #5, that is the basis for the BA.
The first year of PDGM was 2020. The final rule for 2020 defined how this new payment model would work and provided the basis for the original BA. CMS came up with three assumptions they felt would be leveraged by providers to increase their revenue above what was calculated in CMS PDGM projections based on actual claim data from prior years simulated as PDGM payments. Based on CMS interpretation of these regulations, they not only identified these behaviors, but they calculated the expected financial outcome based on the CMS assumptions regarding these behaviors and the rate they would occur. Using this process, they came up with projections that CMS home health spending would increase by 8.389% in 2020 due to these provider behaviors.
Most of us remember when this concept was introduced and the backlash it created. CMS appeared to relent to industry pressure and reduced this in half (-4.36%) in the final rule. This is a pattern they have maintained each subsequent year when they calculated the BA, introduced it in the proposed rule and then reduced it by 50% in the final rule.
In the 2023 final rule, CMS discusses these original assumptions and declared that their predictions were correct and that these behaviors did occur.
Like their assumptions regarding home health profit margins, these claims are greatly exaggerated. With the benefit of hindsight, let’s look at each of these original behavioral assumptions, what they cost HHAs, +how they would have been implemented by providers and the data associated with each of them.
Clinical Group Coding (-6.40%)
“Among available diagnoses, one leading to highest payment clinical grouping classification designated as principal”
The Workflow of Potential “Upcoding”
I remember when I first read the 2019 proposed rule and these original BA, this was the one that hit home for me as an assumption that could not possibly be valid. Not because there were problems with the math or logic, but in the assumption itself and how it would have to be implemented by providers.
Back in the 1980’s when the DRG prospective payment model was introduced for hospitals, I was beginning my career. I was working as a programmer at a company that provided software for the first generation of IBM hospital information systems. Through Medicare, I was able to obtain the tables and code used by 3M for the first DRG grouper. Using this, I was able to reproduce the grouper in our patient accounting system and calculate the expected payment for our clients as soon as the data elements were available. This allowed them to get alerts, for example, if current posted charge costs approached expected revenue.
Like PDGM, in DRGs, the first categorization in the payment formula is based on the primary diagnosis. Using different diagnosis codes as the primary can increase or decrease your expected reimbursement. At some point during the time that my grouper was in production, I received a request to add a feature that would allow the user to “maximize” the reimbursement by having the software calculate expected payment with each available diagnosis code as the primary diagnosis. The results displayed the diagnosis code sequence required for maximum revenue. With the press of a function key, this could be saved to the patient account and provided to Medicare through the claim.
As most of you probably have guessed, this feature was not available for long until someone got in trouble for using it. I don’t know the details as they were above my pay grade at the time, but I was told to remove this feature a few months after it was implemented.
The reason for this story is that it demonstrates how HHAs would need to implement this expected behavior in their workflow. The average home health claim has about eight diagnosis codes. Those claims coded by trained professional coders will have many more than the average. In order to execute this process, the HHA, or their coding service vendor, would code each chart as they do now and then process the results through new software provided by their EMR vendors that could “maximize” reimbursement by resequencing these codes, calculating the expected revenue for each combination, and then potentially changing the primary diagnosis solely on this basis. This is not a process that could be executed without such tools.
In home health, the primary diagnosis is the code that justifies the need for home health services. This is supported by clinical documentation finalized before any ICD10 diagnosis codes are created. For this “maximize” process to work, either the results submitted in the claim would not match this documentation or the documentation itself would have to be “maximized” to match the codes after they were resequenced.
In order to generate this additional 6.4% increase in revenue that CMS predicted, all agencies would have to participate in this process.
I believe that these types of assumptions are what happens when you involve too many data scientists and not enough people familiar with the revenue cycle process in the development of these models. All of the people reading this understand that this new workflow would not happen and why.
This “behavior” could and probably would be considered fraud by auditors. All claim and reimbursement data is thoroughly examined by CMS contracted auditing firms that look for behavior like this. They make money by finding it. Anyone implementing tools that manipulated claim data in a manner that maximizes reimbursement would be easily identified by these auditors and the supporting evidence would be irrefutable. Even if a provider was not concerned with the ethical implications of this practice, they understand the power that CMS holds through takebacks on current reimbursement when any previous payments are in question.
The other problem is that this change in workflow would have to be incorporated into the revenue cycle and coding process. This means retraining your staff to include the objective of selecting the best paying diagnosis code when submitting the final claim instead of the one that justifies their care in the home. How would that play out?
From my own perspective, the most ridiculous part of this assumption was that CMS predicted that everyone would do it. It really does not matter what new revenue cycle opportunity might exist or its potential benefit, everyone would not implement it. This is especially true when it applies to changes in the revenue cycle workflow or the associated technology.
The CMS Data “Supporting” Upcoding
In March of 2023, CMS held a webinar describing the process used to measure the actual behavior changes by home health agencies used to calculate the permanent and temporary adjustments. They published a powerpoint deck from this webinar. In this deck, they describe the data associated with these three original assumptions. Here is the slide documenting the actual relative changes in PDGM clinical groups by year, through 2021.
The chart is included in the final rule, but the interesting part is the CMS conclusion documented on the right side of the slide. Their own documentation and data demonstrates that this upcoding assumption never occurred since the clinical groups they list as having decreased during the first two years of PDGM are the higher reimbursement groups.
This is a list of the PDGM groups sorted by their average Case Mix Weight (CMW) at the time. This weight determines the relative reimbursement of each clinical group. The four groups they identify as decreasing under PDGM (in red), are four of the top five.
LUPA Threshold (-1,88%)
“One third of LUPAs 1-2 visits away from threshold get extra visits and become case-mix adjusted”
This one is a distant second when it comes to the financial impact. Like the upcoding theory, it is not only false, but the exact opposite happened. Instead of fewer LUPAs, the rate of LUPAs increased under PDGM. Table B3 from the 2024 proposed rule documents this:
This assumption error occurred because of data turbulence created from another then unexpected impact of PDGM, the reduction of visits per period.
HHAs have always tried to minimize LUPAs using varying techniques. This was not new to PDGM since LUPAs existed before this payment model. The fact that LUPAs exist at all is due to the impact of data normalization associated with the variable delivery of visits to patients. Visits can be scheduled and delivered, but it does not always happen as planned. However diligent you may be toward delivering care according to a plan, things happen, many beyond your control. Visits can be canceled, care discontinued, scheduling conflicts, clinician and patient availability changes. There will always be a relative disconnect between what is planned and what happens. The degree of this disconnect is often related to the efficiency of the organization, but no one does it perfectly.
When you collect enough data, it will normalize around a mean measurement for that data. From the 2024 proposed rule, here are the average visits per 30-day period by year:
This average period visits for each year sits at the top of a bell-shaped curve of data distributed on each side of this median with more visits than average on the right and less visits than average on the left (as I imagine it). This is what it looks like:
Each year, as the average visits decreased, the entire distribution of data in the curve shifted by the change in this data. In other words, as average visits went down, the curve shifted to the left. If you were an HHA that did not change your visit patterns, your point on this curve would appear to move to the right.
During PDGM, there have not been significant changes to the LUPA thresholds. They have remained constant. This means that as average visits decreased, the number of LUPAs increased because as visits went down, the fixed LUPA thresholds moved right on this curve with more and more 30-day period data points falling to the left of the moving point where LUPA thresholds are defined on the curve.
When we looked previously at visits and costs, costs per 30-day period decrease when visits decrease. The costs also increase when the cost of your clinical staff increases. What we saw in costs per period was a net of these two data curves as well as any other unknown or undocumented factors that might contribute to this measure.
When we look at these LUPA rates provided by CMS, they may include some behavior adjustments as described in the CMS projections, but this behavior has been overwhelmed by the powerful data current associated with visit reductions and its impact on LUPAs through the shift in this curve of data points..
Like the previous prediction regarding clinical upcoding, the assumptions behind this CMS prediction are suspect. First, I would love to hear the story of how they came up with “One third of LUPAs 1-2 visits away from threshold get extra visits”. This again shows an ignorance of the actual difficulty of HHAs in identifying these situations in time to act on them and making these visit adjustments with this level of precision. This one third prediction applies to all agencies and all claims in their analysis. For every agency that takes no action toward this goal, another of equal size would have to double their effort.
Like the clinical grouping prediction, CMS assumes that new software with miraculous new techniques for managing these visits would somehow appear for HHAs and solve this problem in conjunction with the introduction of PDGM. LUPAs are not new and although the thresholds have changed, the methods for avoiding LUPAs have not. The tools for this problem and HHA adoption are improving, but it is still as much an art as a science.
As far as provider implementation of this behavior, once again, the CMS vision of the home health industry as a well oiled machine capable of turning, in unison, toward any opportunity for margin improvement is faulty. In my previous posts, we have seen that many providers, especially smaller ones, are incapable of this level of precision when scheduling their staff.
Comorbidity Coding (-0.25%)
“Assigns comorbidity level based on comorbidities appearing on HHA claims and not just OASIS”
This one bugs me in particular because it has nothing to do with provider behavior, but changing data sources used for diagnosis codes. In OASIS, they retain six diagnosis codes. The primary and five additional codes. On the claim, you can provide up to 25 total diagnosis codes. If you compare the claim and OASIS codes, you will normally find all the OASIS codes on the claim, but some codes on the claims that are not in OASIS due to this limitation.
The way that comorbidity works in the PDGM formula is that if certain specific codes exist in the set of additional claim DX codes, alone or in conjunction with others, they trigger a value of Low or High rather than None, increasing reimbursement for that claim.
It stands to reason that having more codes included in the PDGM grouper will increase the level of comorbidity detected on claims.
That being said, this is one behavior that I witnessed being taken very seriously by HHAs several months prior to the implementation of PDGM and was widespread among agencies as they prepared for PDGM. Before PDGM, the purpose of documenting these codes was mainly related to providing an accurate and complete medical record. PDGM made them more relevant for reimbursement.
With a heads up on this, most agencies worked toward a more comprehensive effort to document and collect all potentially relevant medical conditions, even those that might have been considered trivial in the past. More focus was applied to the coding process and improving clinical documentation and the average number of diagnosis codes in the claims increased.
The reason why this behavior actually happened and not the clinical upcoding is that this behavior was aligned with current staff training and provider goals. Increasing coded diagnoses had a value. Along with the revenue cycle benefit, it has made clinical and claim documentation more accurate instead of invalidating it. Finally, increasing the quantity and quality of these additional diagnosis codes created no new workflows and required no new software to implement it. Claims contained 25 DX codes before and after PDGM. Creating more and better diagnosis data just required additional focus on the task, not new tasks or software.
The CMS data supports this behavior change, this is the slide from their presentation:
We can see that the number of claims with no comorbidity decreased under PDGM and these claims, coded to a higher degree of detail, were classified as either Low or HIgh.
Review of CMS BA Conclusions
If we go back and review the CMS predictions of behavior used through the BA to adjust payments in 2020, we can see that these predictions failed to come to pass for two of the three methods they expected providers to implement to increase revenue.
If we went back and used only the CMS BA assumptions supported by data, this is what the actual impact of this BA should have been for 2020 and the underpayment by CMS to HHA claims as a result of these assumption errors:
This prediction error created a liability for CMS that would have had to have been addressed in future payments. In other words, CMS created their own “temporary adjustment” that would have had to have been repaid back to HHAs when this actual data was eventually compared to their original behavior assumptions, as I have done in this post.
We are talking about hundreds of millions of dollars owed to agencies due to these faulty predictions. As we know now, this money was never repaid and will not be. In my next post, we will explore the CMS pivot to a new definition of the BA and how it saved them from this embarrassment, the associated liability and any explanation of these errors.
Comments