Leading health care systems are now in the process of positioning themselves to take on first dollar risk over the next several years as a key strategic business goal. This payment model shift has facilitated an industry-wide movement towards population health that includes the formation of accountable care organizations, shared savings contracts, and capitation. Common to all of these innovative payment and delivery models is an attributed population. Organizations are now responsible for delivering quality outcomes and managing costs for this defined population.
To achieve these goals, providers need a series of discrete resources that can be grouped into human capital, technology and infrastructure needs Human capital needs are driven from the infeasibility of physicians alone to intervene on every patient within a large attributed population. Additional providers working at the top of their license, like pharmacists, community health workers and behavioral health providers will be needed to ensure success. Similarly, technology is a foundational resource in supporting the extensive outreach necessary in managing populations. This includes patient portals, smart phone apps, and telemedicine to extend services and timeliness of traditional providers. This needs to be coupled with a shift from traditional bricks and mortar infrastructure like new hospital towers and towards capital that spans the entire care continuum.
Regardless of the human capital, technology or infrastructure resources chosen, cost constraints dictate that not every patient can receive every resource. And although not everyone can have everything, everyone should have some resources dedicated to their health needs. This underlies the concept in population health of strategic resource allocation. Providers need a method to ensure that the right patients are getting appropriate resources with maximal benefit. Strategic resource allocation becomes an important starting point in determining the mix of assets and people to ensure population health success.
Strategic resource allocation is actualized by understanding the levels of risk within a population. Traditional approaches have largely relied on claims data to risk stratify patients on the basis of their total cost of care. This method of risk stratification is typically represented as a pyramid in which a small proportion of patients are responsible for a disproportionate share of costs.
Tina Esposito, Advocate Health Care
Although common, this approach is fraught with problems. A high cost patient may be managed well, but continue to be high cost due to the nature of a clinical condition. Allocating care management resources to this well-coordinated but expensive patient is inefficient and further costly. More significantly though is that the majority of high cost patients do not stay high cost. Shifting among the risk categories happens frequently. Often by the time a high cost patient has been identified and allocated resources, the patient has begun “self-correcting” or decreasing in total cost of care expended. This regression to the mean is common and often contributes to the lack of efficacy of population health interventions.
These limitations highlight the need to move beyond claims data in performing risk stratification. A more robust approach includes the combination of claims with clinical and alternative sources of data for which providers are best positioned to leverage. Clinical EMR systems can start to provide this further insight. For example, a claims source can identify a patient as diabetic, but the added information of HbA1c levels, blood pressure readings, and lipid control helps characterize the patient as controlled or uncontrolled (beyond just utilization). This begins to paint a more complete clinical picture of patient risk. This analytic clinical data need is different than the point of care data provided by health information exchanges. The simple diabetes example supports the need to capture data beyond what is available in a Continuity of Care Document (CCD) and inclusive of any data about the patient including textual notes by clinicians, non-medical related spending patterns and data containing environmental or social factors that may be available via open sources like census tracts.
To realize the possibilities the data will need to be aggregated in a way that is meaningful and accessible. This includes concept mapping so that data is recognizable regardless of what EMR sourced the data (e.g., all HbA1c results are available and viewable). Further, the data needs to be attributed to the right patient which requires a robust and accurate enterprise master patient index. Only when all these technical hurdles are addressed can the data then be leveraged by clinicians, analysts, and data scientists for further insights.
At Advocate Health Care, a robust big data platform has been utilized for strategic resource allocation in the decision making around a patient’s discharge disposition following hospitalization. On the surface, this may not seem like a resource allocation problem. And in a traditional reimbursement model which promotes siloes of care–it isn’t. But in a population health environment, the post-discharge options of home care, skilled nursing facility, or long-term care are true costs that a health system has accountability for. Now the health system has a responsibility for optimal decision making at discharge to ensure the allocation of post-discharge resources for the greatest value.
At Advocate, the problem of post-hospitalization resource allocation was solved by stratifying patient risk through a transition of care predictive model. The purpose of the model was to objectively match the patient’s need with the right post-acute service. From a patient perspective, the goal of the model was to return an individual to a level of post-hospitalization independence sooner by reducing the likelihood of a readmission and decreasing unnecessary utilization. The model employs propensity score matching based on variables that predict post-acute care placement to create retrospective case and control groups for each
post-acute setting. Simplified, the model matches “like” patients via a clinical profile and identifies what post-acute care location that type of patient is most successful in (readmitted less often).The model output is a patient disposition recommendation to the care manager and provider within their EMR workflow. The model runs on a big-data platform that includes clinical data from all of Advocate’s EMRs (multiple vendors), claims data from payers and internal transactional data. Running the model on three years of historical data has indicated that the system may be underutilizing home health services and over-utilizing more costly venues like skilled nursing facilities. With a large home health division and extensive post-acute network of affiliated skilled facilities the transitions of care model will be used to better target post-acute resources to the right patients.
Strategic resource allocation needs to be data driven, powered by analytics and intelligence that only comes when all available data resources are leveraged. For providers, the insight allowed with a combination of both clinical and claims data is a ‘game-changer’ in an industry that has relied on the non-specificity of claims data. To leverage this potential, healthcare organizations need to invest on technology and resources that will aggregate large amounts of data from across the continuum, pain-stakingly map it for consistency, and then hire resources to use it for knowledge creation. This is the business case of ‘big-data’ in healthcare.