A Comparative Analysis of April 2020 Medicare SNF Claims
The human mind is a poor processer of large numbers (“large” in this case is any figure with more than two digits). Accounting for fatalities from 1918’s influenza pandemic is, unfortunately, a logical place to begin a discussion about this evolutionary shortcoming. As curiosity naturally piqued interest about that strain’s impact on the World War I-era population, a bit of research revealed that 17 million to 100 million souls succumbed to the disease. Absent context, this 83 million range seems absurd; but given the extreme variables at play, the relative “margin for error” is understandable. Comparatively, consider the old “jellybeans in a jar” guessing game – an estimate of 100+ candies when there are clearly no more than 17 would be mocked; but find a jar large enough to hold 17 million jellybeans and all bets are off. To make the point on a more practical scale, just fill in the blank (note this is not a riddle or trick question): “The average human adult male has 17,000 _______________.”
Perspective is equally deceiving. Coming of age in the early 1990s, “Generation X” would scoff at the thought of any flu strain representing a greater population threat than AIDS, yet 1918’s H1N1 ended more lives in 24 weeks than HIV/AIDS claimed in 24 years. Recency bias notwithstanding, mental processing of scale is a major challenge that, in real time, warps society’s response. To compensate, we default to “comparison” as a proxy for direct measurement. The impediment to accurate comparative analysis is that all independent variables must be identical when assessing outcomes – aka “apples-to-apples.” For example, ask five people born in different decades the following question:
Who is the greatest basketball player of all time? There is no correct answer. Yes, the game is the game, but enough small, accretive changes over eras invalidate objective performance measures required for scientific comparison (while recency bias alters perspective in more qualitative ways).
COVID distribution reporting is subject to extreme bias for a host of reasons – first and foremost, the virus is selectively more threatening to the health-compromised elderly – a cohort that disproportionately resides in institutional settings. Statistically, introducing one person infected with the coronavirus to a nursing facility would produce outcomes that are non-comparable to introducing that same catalyst into Central Park on a warm spring day.
Technology has improved our ability to process large numbers but not our perspective. Skilled Nursing Facilities care for the most susceptible in society. Despite compliant, robust infection control policies, institutional environments are, by nature, highly conducive to contagion. Furthermore, political folly, media distortion and the law of supply and demand (PPE) skewed statistics against the provider in many ways, including financially.
The “Era” of Comparative Integrity
Comparing RUGs to PDPM remains vexing. CORE’s first data release attempted to differentiate between October’s PDPM performance and a theoretical October RUG-IV per diem rate, had the latter system continued unabated. Our second release compared November’s PDPM performance against CMS’ original financial impact projection (based on 2017 assessments). The bias in that exercise was the evolution of location-specific independent variables – post-acute care had evolved so quickly and unevenly, “era bias” was evident after only three years. There was no distinct inflection point, but enough minor, accretive changes occurred to mitigate the Comparative Integrity of Medicare billing in many markets. CORE found PDPM rates were generally higher than CMS anticipated, but the underlying utilization declines (driven by Medicare Advantage/ISNPs & FFS shared savings initiatives) offset those gains enough to lower “episodic” revenue for many providers.
Medicare’s “Limited Data Set”
Medicare releases encrypted Limited Data Set (LDS) claims that are six to nine months old and stripped of data elements required for detailed analysis. As noted previously, the difference between “most up to date” and “current” is of major relevance today. We will not have LDS billing data specific to the SNF COVID-era for months, as “mature,” generalized Medicare claim compilations will not be available until November or December. However, CORE Analytics offers the first sample of reality, as reported by SNFs on the front lines of the COVID battlefield.
Medicare’s Q4 2019 claims data was “ready for extraction” in mid-May 2020. CORE decodes and includes the same data – it serves a purpose – but its lack of detail and timeliness offers little value to providers seeking insight regarding COVID’s impact on their financial performance. While PDPM is only nine months old, the most recent LDS is already BCE (“Before the COVID Era”); October - December 2019’s generalized claim data is non-comparable to COVID-era outcomes. CORE remains the only tool for analyzing CURRENT claims.
The CORE Database
CORE’s “active” user base has more than doubled in size since October 2019 and is far more diverse. We define an “active user” as a provider or stakeholder that has uploaded claims within the past 60 days and/or spent more than 60 minutes per month engaged with CORE’s other functional tools such as market data (e.g. Medicare’s LDS), predictive analytics or benchmarking ancillary spending by PDPM group.
CORE’s 1,300+ active users operate in 40 states. Urban SNFs represent 84% of our clients, 91% are forprofit with an average bed complement of 138. Nevertheless, the “sample” remains non-randomly selected and therefore technically lacks statistical significance. Accordingly, our findings cannot be used as a formal reflection of national performance. That said, if our reach extended to every SNF claim filed this year, the dataset would still not reflect national performance – just an amalgamized collection of hundreds of “local” healthcare ecosystems.
The COVID Effect: Analysis of April Claims
March was an incredibly challenging month for all. From an analytics perspective, the confusion and distortion of the “1135 Waiver” in addition to the absence of admissions from post-elective surgery extended into April, but April claims were the first to contain the COVID ICD-10 code. This release distinguishes between SNFs that reported COVID claims and those that did not. We went one step further and analyzed the PDPM rate composition of claims that included a COVID diagnosis against those submitted without the newly designated code.
As explained, March claims cannot be distilled the same way. Therefore, we used “BCE” February claims as our baseline to more accurately evaluate how a COVID diagnosis impacted a SNF’s Medicare reimbursement distribution. Specifically:
1,035 SNFs with at least 8 claims from February and April are included in this analysis – of 32,512 total April claims, 6,470 reported a COVID Dx (~20%). However, to report figures without context is irresponsible, as these residents were clustered in fewer than half of CORE’s SNFs.
410 (~40%) of providers reported at least one COVID-positive Medicare Part A fee-for-service claim. Given the sharply uneven distribution of Medicare Advantage/ISNP among markets, the full impact of the “blanket” 1135 waiver could not be normalized.
Among the providers that reported at least one COVID Dx, the average share of COVID positive claims was 30.4%.
The average per diem rate for COVID patients (using the urban wage index of 1.0) was $669; nearly 9% higher than the comparable CORE rate average from February.
Recent studies of SNF infection patterns leave little room for debate, but at the provider level they did not stratify by pay source; our sample is based on Medicare SNF FFS claims only. We are in no way questioning these more comprehensive studies – we defer to them. CORE reports on reimbursement, and our data clearly shows COVID-positive patients generated higher rates than non-COVID patients. PDPM came along just in time.
RUG-IV would have been a fiscal disaster for SNFs dealing with COVID. Instead of $669 per day, RUG-IV was so distorted by therapy volume, rates could have been hundreds of dollars lower. In fact, Medicaid coverage would have generated higher net revenue than Medicare RUG-IV (with no therapy or isolation) in many states. PDPM is not perfect and requires refinement, but RUG-IV was a statistical disgrace – embraced by many providers for financial survival and perpetuated by an array of complex variables led by geography
COVID claims highlight the lack of rate sensitivity in the PDPM therapy component – patients with a COVID diagnosis received little to no therapy (per charges on the UB-04) yet Medicare reimbursed more for non-positive residents who accrued standard therapy charges. COVID drove PDPM rates higher due to the Nursing and NTA components.
Isolation (ES1) was captured in 35.2% of COVID patient claims. This figure was tempered by the private room requirement for capture – facilities with more private room availability realized higher relative rates for COVID patients than those housed in semi-private rooms. For reference, ES1 pays $310 per day; the February average Nursing component rate was $188.
Non-isolated patients in SNFs with at least one COVID-positive claim were 62% more likely to trigger the Depression end split than patients in facilities with no reported COVID claims (the Extensive Services score for isolation is not sensitive to the Depression end split).
The NTA difference was the result of lower lengths of stays and a shift of claims with zero (NF) NTA points, but this distribution was uneven across state lines. States like New York, where SNFs were mandated to accept COVID-positive admissions irrespective of law and logic, had fewer days covered per admission. Many patients who should not have been admitted to the SNF in the first place expired after a matter of days – the product of an ill-conceived initiative to shift mortality rates from hospital to SNF (politics and population health do not mix well)
Conclusion
This analysis is about Medicare FFS dollars, yet beneath the surface we found the same fractured political and payer mix disparities that belie the concept of skilled nursing as a “national industry.”
CORE data represents only a sample of Medicare billing culled from the first month of claims that enabled such analysis. Our main takeaway is that the sensitivity of the Patient Driven Payment Model was a far superior method of rate setting than would have been realized under the therapy-centric RUG system. That said, we will leave the politics and perspective accounting to a more reliable arbiter for now… time.
And FYI, the average human adult male has 17,000 facial whiskers.
ABOUT CORE ANALYTICS
The UB-04 has been overlooked as a powerful SNF analytics tool since the MDS gained primacy under PPS. PDPM has changed this dynamic. The UB-04 Medicare claim’s diverse item-set is distinct from the MDS or medical record, and as such offers unique insight into provider and market dynamics – ZCORE Analytics was created to unlock this potential.
CORE is a stand-alone system that requires no technology integration. SNFs simply upload 837i files to CORE as they would to Medicare. There are no on-boarding costs, no cancellation fees, no contract lock-in burden... and no more than 30 minutes needed to realize results
THE COVID-EFFECT
Marc Zimmet
A Comparative Analysis of April 2020 Medicare SNF Claims
The human mind is a poor processer of large numbers (“large” in this case is any figure with more than two digits). Accounting for fatalities from 1918’s influenza pandemic is, unfortunately, a logical place to begin a discussion about this evolutionary shortcoming. As curiosity naturally piqued interest about that strain’s impact on the World War I-era population, a bit of research revealed that 17 million to 100 million souls succumbed to the disease. Absent context, this 83 million range seems absurd; but given the extreme variables at play, the relative “margin for error” is understandable. Comparatively, consider the old “jellybeans in a jar” guessing game – an estimate of 100+ candies when there are clearly no more than 17 would be mocked; but find a jar large enough to hold 17 million jellybeans and all bets are off. To make the point on a more practical scale, just fill in the blank (note this is not a riddle or trick question): “The average human adult male has 17,000 _______________.”
Perspective is equally deceiving. Coming of age in the early 1990s, “Generation X” would scoff at the thought of any flu strain representing a greater population threat than AIDS, yet 1918’s H1N1 ended more lives in 24 weeks than HIV/AIDS claimed in 24 years. Recency bias notwithstanding, mental processing of scale is a major challenge that, in real time, warps society’s response. To compensate, we default to “comparison” as a proxy for direct measurement. The impediment to accurate comparative analysis is that all independent variables must be identical when assessing outcomes – aka “apples-to-apples.” For example, ask five people born in different decades the following question:
Who is the greatest basketball player of all time? There is no correct answer. Yes, the game is the game, but enough small, accretive changes over eras invalidate objective performance measures required for scientific comparison (while recency bias alters perspective in more qualitative ways).
COVID distribution reporting is subject to extreme bias for a host of reasons – first and foremost, the virus is selectively more threatening to the health-compromised elderly – a cohort that disproportionately resides in institutional settings. Statistically, introducing one person infected with the coronavirus to a nursing facility would produce outcomes that are non-comparable to introducing that same catalyst into Central Park on a warm spring day.
Technology has improved our ability to process large numbers but not our perspective. Skilled Nursing Facilities care for the most susceptible in society. Despite compliant, robust infection control policies, institutional environments are, by nature, highly conducive to contagion. Furthermore, political folly, media distortion and the law of supply and demand (PPE) skewed statistics against the provider in many ways, including financially.
The “Era” of Comparative Integrity
Comparing RUGs to PDPM remains vexing. CORE’s first data release attempted to differentiate between October’s PDPM performance and a theoretical October RUG-IV per diem rate, had the latter system continued unabated. Our second release compared November’s PDPM performance against CMS’ original financial impact projection (based on 2017 assessments). The bias in that exercise was the evolution of location-specific independent variables – post-acute care had evolved so quickly and unevenly, “era bias” was evident after only three years. There was no distinct inflection point, but enough minor, accretive changes occurred to mitigate the Comparative Integrity of Medicare billing in many markets. CORE found PDPM rates were generally higher than CMS anticipated, but the underlying utilization declines (driven by Medicare Advantage/ISNPs & FFS shared savings initiatives) offset those gains enough to lower “episodic” revenue for many providers.
Medicare’s “Limited Data Set”
Medicare releases encrypted Limited Data Set (LDS) claims that are six to nine months old and stripped of data elements required for detailed analysis. As noted previously, the difference between “most up to date” and “current” is of major relevance today. We will not have LDS billing data specific to the SNF COVID-era for months, as “mature,” generalized Medicare claim compilations will not be available until November or December. However, CORE Analytics offers the first sample of reality, as reported by SNFs on the front lines of the COVID battlefield.
Medicare’s Q4 2019 claims data was “ready for extraction” in mid-May 2020. CORE decodes and includes the same data – it serves a purpose – but its lack of detail and timeliness offers little value to providers seeking insight regarding COVID’s impact on their financial performance. While PDPM is only nine months old, the most recent LDS is already BCE (“Before the COVID Era”); October - December 2019’s generalized claim data is non-comparable to COVID-era outcomes. CORE remains the only tool for analyzing CURRENT claims.
The CORE Database
CORE’s “active” user base has more than doubled in size since October 2019 and is far more diverse. We define an “active user” as a provider or stakeholder that has uploaded claims within the past 60 days and/or spent more than 60 minutes per month engaged with CORE’s other functional tools such as market data (e.g. Medicare’s LDS), predictive analytics or benchmarking ancillary spending by PDPM group.
CORE’s 1,300+ active users operate in 40 states. Urban SNFs represent 84% of our clients, 91% are forprofit with an average bed complement of 138. Nevertheless, the “sample” remains non-randomly selected and therefore technically lacks statistical significance. Accordingly, our findings cannot be used as a formal reflection of national performance. That said, if our reach extended to every SNF claim filed this year, the dataset would still not reflect national performance – just an amalgamized collection of hundreds of “local” healthcare ecosystems.
The COVID Effect: Analysis of April Claims
March was an incredibly challenging month for all. From an analytics perspective, the confusion and distortion of the “1135 Waiver” in addition to the absence of admissions from post-elective surgery extended into April, but April claims were the first to contain the COVID ICD-10 code. This release distinguishes between SNFs that reported COVID claims and those that did not. We went one step further and analyzed the PDPM rate composition of claims that included a COVID diagnosis against those submitted without the newly designated code.
As explained, March claims cannot be distilled the same way. Therefore, we used “BCE” February claims as our baseline to more accurately evaluate how a COVID diagnosis impacted a SNF’s Medicare reimbursement distribution. Specifically:
Conclusion
This analysis is about Medicare FFS dollars, yet beneath the surface we found the same fractured political and payer mix disparities that belie the concept of skilled nursing as a “national industry.”
CORE data represents only a sample of Medicare billing culled from the first month of claims that enabled such analysis. Our main takeaway is that the sensitivity of the Patient Driven Payment Model was a far superior method of rate setting than would have been realized under the therapy-centric RUG system. That said, we will leave the politics and perspective accounting to a more reliable arbiter for now… time.
And FYI, the average human adult male has 17,000 facial whiskers.
ABOUT CORE ANALYTICS
The UB-04 has been overlooked as a powerful SNF analytics tool since the MDS gained primacy under PPS. PDPM has changed this dynamic. The UB-04 Medicare claim’s diverse item-set is distinct from the MDS or medical record, and as such offers unique insight into provider and market dynamics – ZCORE Analytics was created to unlock this potential.
CORE is a stand-alone system that requires no technology integration. SNFs simply upload 837i files to CORE as they would to Medicare. There are no on-boarding costs, no cancellation fees, no contract lock-in burden... and no more than 30 minutes needed to realize results