A recent article in the New England Journal of Medicine [1] has attracted significant attention because of its apparent implications to accountable care organizations. This article references the Medicare Physician Group Practice (PGP) demonstration project that operated from 2005 through 2010. In this demonstration, various types of provider organizational structures were evaluated based on quality and cost metrics, and were rewarded based on the savings that they were able to achieve relative to a target developed by CMS after meeting specified quality metrics. The article questions whether the currently-proposed ACO model will allow most such organizations to recover their initial costs within a reasonable time period, and whether those organizations would have sufficient capitalization to maintain financial viability during that period. These conclusions are well-founded based on the information that CMS has presented about the demonstration; however other interesting conclusions can also be drawn from that project.
Insights from the PGP demonstration
A useful source of information is the Report to Congress prepared on this demonstration in 2009. [2] This report includes several observations of interest to potential ACOs related to the PGP’s costs as compared to the CMS targets. The metric used to determine whether a cost sharing payment would be made by CMS to the PGP organization, which is also likely to be used for ACOs, is a comparison of the actual cost incurred to a target computed by CMS. These targets are computed based on a severity adjustment process that utilizes hierarchical condition categories (HCCs) to adjust the payment based on the diagnoses of the patients in the population. Organizations whose patients have more complex medical conditions requiring more medical services receive a higher risk score, and hence a higher payment target. Organizations whose costs are lower than the target by a specified percentage get to share in those savings.
Thearticle points out that six of the 10 PGP practices did not receive shared savings payments in the second year of the demonstration. However, several characteristics about the four practices that did receive these payments are interesting. First, none of these practices belonged to integrated delivery systems. All were either freestanding physician groups or were affiliated with an academic medical center. The report suggests that the "presence of a hospital was hypothesized as a potential deterrent to achieving savings" because of the effect of reductions in admissions on the hospital’s revenue. This lack of alignment of financial incentives between managed care organizations and hospitals was a principal reason for the failure of many of these organizations in the 1990s, and apparently had not yet been overcome as of the time of the demonstration. Hospitals considering becoming part of ACOs should seriously question their ability to overcome this barrier.
They couldn’t lose
A second and potentially more interesting conclusion of the report was that the four PGPs that earned cost sharing payments in the second year had "performance {that} was almost matched in the pre-demonstration period." In other words, they were 'winners' before the project even started. The CMS report even suggests that this advantage "may be one of the reasons why they elected to participate in the demonstration." They decided to play because they knew they couldn’t lose.
But how can this happen? How can a team know before the game that they’ll win? Well, perhaps these organizations had achieved a sufficiently low cost structure that they expected to beat the cost targets, whatever those targets might be. Or perhaps, like in any other game, they had read the rule book, analyzed the rules, figured out what it took to win, and already knew the answer [3].
But what are the rules, and do they present a level playing field for all ACO participants? Under the PGP project, the targets were set based on HCCs, which adjust the cost target based on the risk characteristics of the patient population. HCCs are a well-established payment mechanism currently used for payment of Medicare Advantage plans, PACE programs, and several others, and are likely to be utilized for ACOs. Ideally, HCCs accurately adjust for all patient mixes, such that no organization's mix of patients would have an advantage or disadvantage over any other organization’s patient population regardless of the severity of disease in the population.
Level playing field?
A study [4] performed several years ago, however, called this assumption into question. That study concluded that HCCs under-predict the costs of patients with multiple complex diagnoses and chronic conditions. For example, it found that HCCs under-predict the expenses of Medicare beneficiaries with both CHF and osteoporosis by about 30 percent, and by about 20 percent for patients with CHF alone. The amount of under-prediction increased as the functional status of the patients decreased. Therefore, groups having a large proportion of patients with multiple chronic conditions risk being underscored for those patients. Such groups may be "born losers" having little opportunity for financial success in an ACO.
If that’s the case, then why shouldn't an ACO simply avoid enrolling patients with multiple complex conditions? There are two reasons: first, because chronic conditions are prevalent throughout the Medicare population, and second because chronic care patients are the primary drivers of healthcare costs and represent the best opportunity for an ACO to create change in health status and thereby reduce cost. Attempting to structure an ACO without chronic care patients would be extremely difficult and counterproductive. However, ACOs with an unusually high proportion of patients with multiple chronic conditions risk being under-scored for those patients, and may therefore lose opportunities for shared saving bonuses from CMS.
Other potential inaccuracies and biases may also exist in HCC scoring. Given that overall risk scoring is calibrated to be neutral, if some patients are underpaid then others may be overpaid. Like any such statistical aberration these differences probably become less problematic with a larger population, such as those in Medicare Advantage programs. However, ACOs will be significantly smaller than these large health plans, and may be more subject to the randomness of HCC scoring.
Check your stats before the game starts
In these situations, forewarned is forearmed. Because the opportunity to receive the shared savings payments from CMS is envisioned to be the revenue source that will make up for revenue lost through utilization reductions to the ACO's providers, it will be important for ACOs to have in-depth knowledge of the risk scores that will be applied to their patients. Hopefully, they can do this before deciding to participate in an ACO, so that they have some advance idea of their potential for receiving these payments. Perhaps some organizations like those in the PGP demo will find, to their delight, that they’re "winners going in" and will enjoy early success. Others may reach the opposite conclusion, and make different decisions.
Singletrack Analytics is a healthcare financial and data consulting firm specializing in assisting healthcare providers and purchasers achieve success through better use of data and analytic techniques. For information about Singletrack Analytics, please visit www.singletrackanalytics.com. A detailed article about risk analytics for ACOs entitled "Analytics for ACOs" that discusses these methodologies can be found on the "Articles and White Papers" page of the Singletrack Analytics website at www.singletrackanalytics.com.
[1] “The ACO Model – A Three-Year Financial Loss?”, NEJM 10.1056/NEJMP1100950 NEJM.ORG
[2] “Report to Congress, Physician Group Practice Demonstration Evaluation Report”, Secretary of Health and Human Services 2009, available from http://www.cms.gov/demoprojectsevalrpts/md/itemdetail.asp?itemid=CMS1198992
[3] For other irrelevant but interesting sports-based uses of analytics, refer to “Moneyball” by Michael Lewis, or The Extra 2%” by Jonah Keri.
[4] “Medicare Capitation Model, Functional Status, and Multiple Comorbidities: Model Accuracy”, The American Journal of Managed Care, Vol 14, No, 10, October 2008
Insights from the PGP demonstration
A useful source of information is the Report to Congress prepared on this demonstration in 2009. [2] This report includes several observations of interest to potential ACOs related to the PGP’s costs as compared to the CMS targets. The metric used to determine whether a cost sharing payment would be made by CMS to the PGP organization, which is also likely to be used for ACOs, is a comparison of the actual cost incurred to a target computed by CMS. These targets are computed based on a severity adjustment process that utilizes hierarchical condition categories (HCCs) to adjust the payment based on the diagnoses of the patients in the population. Organizations whose patients have more complex medical conditions requiring more medical services receive a higher risk score, and hence a higher payment target. Organizations whose costs are lower than the target by a specified percentage get to share in those savings.
Thearticle points out that six of the 10 PGP practices did not receive shared savings payments in the second year of the demonstration. However, several characteristics about the four practices that did receive these payments are interesting. First, none of these practices belonged to integrated delivery systems. All were either freestanding physician groups or were affiliated with an academic medical center. The report suggests that the "presence of a hospital was hypothesized as a potential deterrent to achieving savings" because of the effect of reductions in admissions on the hospital’s revenue. This lack of alignment of financial incentives between managed care organizations and hospitals was a principal reason for the failure of many of these organizations in the 1990s, and apparently had not yet been overcome as of the time of the demonstration. Hospitals considering becoming part of ACOs should seriously question their ability to overcome this barrier.
They couldn’t lose
A second and potentially more interesting conclusion of the report was that the four PGPs that earned cost sharing payments in the second year had "performance {that} was almost matched in the pre-demonstration period." In other words, they were 'winners' before the project even started. The CMS report even suggests that this advantage "may be one of the reasons why they elected to participate in the demonstration." They decided to play because they knew they couldn’t lose.
But how can this happen? How can a team know before the game that they’ll win? Well, perhaps these organizations had achieved a sufficiently low cost structure that they expected to beat the cost targets, whatever those targets might be. Or perhaps, like in any other game, they had read the rule book, analyzed the rules, figured out what it took to win, and already knew the answer [3].
But what are the rules, and do they present a level playing field for all ACO participants? Under the PGP project, the targets were set based on HCCs, which adjust the cost target based on the risk characteristics of the patient population. HCCs are a well-established payment mechanism currently used for payment of Medicare Advantage plans, PACE programs, and several others, and are likely to be utilized for ACOs. Ideally, HCCs accurately adjust for all patient mixes, such that no organization's mix of patients would have an advantage or disadvantage over any other organization’s patient population regardless of the severity of disease in the population.
Level playing field?
A study [4] performed several years ago, however, called this assumption into question. That study concluded that HCCs under-predict the costs of patients with multiple complex diagnoses and chronic conditions. For example, it found that HCCs under-predict the expenses of Medicare beneficiaries with both CHF and osteoporosis by about 30 percent, and by about 20 percent for patients with CHF alone. The amount of under-prediction increased as the functional status of the patients decreased. Therefore, groups having a large proportion of patients with multiple chronic conditions risk being underscored for those patients. Such groups may be "born losers" having little opportunity for financial success in an ACO.
If that’s the case, then why shouldn't an ACO simply avoid enrolling patients with multiple complex conditions? There are two reasons: first, because chronic conditions are prevalent throughout the Medicare population, and second because chronic care patients are the primary drivers of healthcare costs and represent the best opportunity for an ACO to create change in health status and thereby reduce cost. Attempting to structure an ACO without chronic care patients would be extremely difficult and counterproductive. However, ACOs with an unusually high proportion of patients with multiple chronic conditions risk being under-scored for those patients, and may therefore lose opportunities for shared saving bonuses from CMS.
Other potential inaccuracies and biases may also exist in HCC scoring. Given that overall risk scoring is calibrated to be neutral, if some patients are underpaid then others may be overpaid. Like any such statistical aberration these differences probably become less problematic with a larger population, such as those in Medicare Advantage programs. However, ACOs will be significantly smaller than these large health plans, and may be more subject to the randomness of HCC scoring.
Check your stats before the game starts
In these situations, forewarned is forearmed. Because the opportunity to receive the shared savings payments from CMS is envisioned to be the revenue source that will make up for revenue lost through utilization reductions to the ACO's providers, it will be important for ACOs to have in-depth knowledge of the risk scores that will be applied to their patients. Hopefully, they can do this before deciding to participate in an ACO, so that they have some advance idea of their potential for receiving these payments. Perhaps some organizations like those in the PGP demo will find, to their delight, that they’re "winners going in" and will enjoy early success. Others may reach the opposite conclusion, and make different decisions.
Singletrack Analytics is a healthcare financial and data consulting firm specializing in assisting healthcare providers and purchasers achieve success through better use of data and analytic techniques. For information about Singletrack Analytics, please visit www.singletrackanalytics.com. A detailed article about risk analytics for ACOs entitled "Analytics for ACOs" that discusses these methodologies can be found on the "Articles and White Papers" page of the Singletrack Analytics website at www.singletrackanalytics.com.
[1] “The ACO Model – A Three-Year Financial Loss?”, NEJM 10.1056/NEJMP1100950 NEJM.ORG
[2] “Report to Congress, Physician Group Practice Demonstration Evaluation Report”, Secretary of Health and Human Services 2009, available from http://www.cms.gov/demoprojectsevalrpts/md/itemdetail.asp?itemid=CMS1198992
[3] For other irrelevant but interesting sports-based uses of analytics, refer to “Moneyball” by Michael Lewis, or The Extra 2%” by Jonah Keri.
[4] “Medicare Capitation Model, Functional Status, and Multiple Comorbidities: Model Accuracy”, The American Journal of Managed Care, Vol 14, No, 10, October 2008