Value is the lynchpin in health care reform. As the basis for the Alternative Payment Model Value is a measure of cost and outcomes, cost are known from payors, outcomes not so clear. The Value proposition is that cost can be contained without sacrificing quality. The engine that drives this is accountability, the basis of measures linked to qualified organizations and providers. Accountability functions by deriving patient costs from claims data, whereby benchmarks are created for defined populations that are then used to judge whether providers are within bounds for costs. The assumption is that providers when so assessed by costs will then impact their practice or practice group to discover efficiencies. The claims set’s originating cost data are used to measure quality basically by tracking adverse events reflected in codes, this is the basis of the current system for accountable organizations, ACO, and PCMH. Accountability is used also for gainsharing programs based on Episodes of Care, a FFS based APM level 2 program. This discussion will show in detail the basis of Value as a numerator made of patient level granular events and a denominator including all similar patients as a population of interest.
Generally the APM Value framework in creating population based benchmarks can be broad or narrow in selecting the population of record to serve as the basis of the benchmarks. For example the Episode of Care APM level 2 compares the sum of episodes assigned to a provider to the population from similar providers of record for patients with the same episodes. Here the issue of risk stratification and small numbers is critical because to allow fair and appropriate comparison of individual providers assigned as quarterbacks for the episode of record, random variation in outcomes is a risk. The ACO model however has larger populations whose global care costs are less vulnerable to the problem of small numbers though not entirely free of this risk. Also accountability is assigned not to an individual provider but to all the members of the ACO. This version of accountability is APM level 3.
The difference in these 2 models for accountability rests on numbers both of providers and of population. The power of the APM is the ability to effect provider’s behavior reflected in quality and cost of the attributed populations that serves as the comparator applied to the provider. Ultimately in the patient centered framework used in this discussion, the decisions of providers at the patient level determines behavior that should be rewarded or penalized. Even for patient level decisions, the context and guidance comes from populations whether small or large. The issue for defining accountability is how to determine the population that serves as the comparator for individual patient level decisions. In MedPac reports on Medicare this core issue determines which provider organizations are eligible to be considered accountable organizations. MedPac recognizes the problem in determining that fee for service is not eligible to be held accountable, and Medicare Advantage and Accountable organizations are. The difference is that accountable organizations have patients attributed to them that serves as a population comparator whereas independent fee for service providers have no ability for attribution of an assigned population. Therefore no benchmarks can be created for the FFS groups. The MACRA law recognizes this and will promote a rapid shift to all accountable organizations.
To complete this brief review of the Alternative Payment Model the 4 APM levels are:
Promotion of desired delivery models linked to APM requires all entities of Networks, from insurance plans, to hospitals, and to individual providers who will be held accountable for cost and quality, have a method for attribution of providers to populations. The attribution can be via the payment model, with simple assignment coming from plans. The delivery model then responds to incentives of payment reform. If the Network is patient centered another approach to assignment is possible whereby the delivery model creates the populations, a reversal of the standard model where attributed populations are derived from plans. This reversal requires only manipulation of information that spans the progress from the single episode instance to the multilevel Network that houses the information of many patient’s atomic events. The events can be bundled in many ways, by global cost profile, disease, and by any parameter that is patient centered. All patients that are included in a care delivery group become the population. It is more interesting than this when patient centering enlists not one but many populations of which a patient is a member . The sum of populations identifies the patient . As discussed in the Patient Centered Network tab, one atomic event proceeds to many in a patient journey on a trajectory to value; as this unfolds the network responds by constantly changing the context of care delivery, which means the population linked to an atomic event becomes dynamic. Here as will be expounded below in the Outcomes bullet, care coordination and communication become natural as patient centered populations involve different providers and evolving venues that accompanies patients on their individual journeys. Also the attribution of becomes easier and can cast a wider net within the provider network.
What is notable about outcomes measures is that they reflect centering of information of network entities whereby averages are created at the scale of network neighborhoods. These neighborhoods can be informal as in referral networks, or formal as with an ACO. And these neighborhoods are too small to generate statistically significant parameters, but are large enough to build benchmarks that will allow comparison of providers for cost and quality. The intent of CMS is to balance the ubiquitous process measures for hospitals and Medicare Advantage plans and others linked to outcomes as the basis of Value equations. For this effort to be transformative, CMS will transition to outcomes measures as the basis for pay for performance programs to 50% of Medicare payments by 2018.
Why is it meaningful to say outcomes are centered on network entities? As the value proposition is referenced mostly by plans, not provider networks, there is an assumption that value means the same in all contexts. Clearly this is not the case. As in APM level 2 where the managed care context is the codes of care transactions, value is construed as cost being the basis of benchmarks. Though outcomes as are used as indicators of adverse or undesirable events within episodes, and the impact on costs, there is no use of outcomes of normative episodes . With risk and business exclusions of episodes, the non-normative outliers which are not included in attributing episode spend to providers, jails outcomes of normal episode journeys. Thus the preferable outcomes when defined only by cost, leaves the analysis of journeys, how they differ, how they are similar, out of the value equation.
If value were centered on provider networks, the parameters differentiating normative from non normative journeys would be clinical descriptors . An example is Hierarchical Condition Categories (HCC) promoted by CMS. This is intended to be the link to more precise clinical measures not captured by Codes, even ICD10, which can lead to more robust outcomes measures focused on journeys. Unfortunately because there is no incentive system to use HCC there is little provider awareness they even exist. If fact CMS uses these measures as risk tools to balance exchange plan costs to allow the same premiums per metal level for varying risk groups.
Similarly there is a unique centering of individual providers. Whereas patient level centering is the most problematic and granular, with many unknowns, provider centering tracks metrics such as specialty, referrals, care coordination and collaboration. This is ultimately the group of variables that centers not just the providers but can be descriptive of the entire network; plans, populations and providers in any type group. By specifying each network entity and the semantics of the hierarchically centered variables, the links and members of the network can be the basis of the Value equation. Here at the level of the provider centered Network, populations can be assembled to serve as the denominator of the Value equation.
Finally what can be said about a Network that is patient centered? What is unique about such a network? The answer lies in understanding how a patient is defined and characterized as the unit of importance in a population. As stated above patients can be members of many populations. As in risk scoring, comorbidities are assembled, each with its own population as is specified in clinical evidence. Risk is intimately linked to past medical history which by definition is unique to the patient. Events of the past are intimately related to future journeys, with the assumption that the trajectory of the journey varies by risk status. The assembly of populations is washed out of population centering in that medical evidence uses comorbidities as covariates, not definitive of a population in and of themselves. For example for a diabetic and hypertensive patient, to predict an outcome after a heart attack is done for the group of heart attack patients with a list of comorbidities of a certain prevalence in the group. Another way to assess outcome is to look at all diabetics, some with a past heart attack, some not. Which population best characterizes and individual patient? The answer is the interaction, easily handled statistically. Consider the wealth of detail presented to the clinician . For each patient, the realm of medicines used by the patient, unique past medical history, family and patient preferences, diversity of problems that exist as factors, but are often known only incompletely if at all. This is the clinician’s dilemma, focusing on the patient and making decisions in a context of uncertainty . In a statistical model patient centered information has unknowns, called latent variables. For patient centering, in contrast to population centering, these unknowns do not go away, but are averaged out in population centering.