the 6 lenses of population health management

Using the appropriate analytical framework can help health systems plan changes to their business models that will support population health management.

Having "one foot on the dock and one foot in the boat" is an expression commonly used to describe the ongoing transition from a fee-for-service healthcare payment model to one centered on population health and value-based care. With the rapid adoption of value-based payment models, the boat is drifting quickly away from shore.

Nearly 4,0 percent of Medicare patients are in value-based contracts, with almost 3o percent in Medicare Advantage programs and about 10 percent in the Medicare Shared Savings Program and Pioneer accountable care organizations (ACOs).a An estimated 626 ACOs are in operation and covering more than 20 million lives, with the majority represented in commercial contracts.b

Few provider organizations have fully optimized care delivery processes, competencies, and resources for population health management given the significant financial challenges and the requirements for new leadership skills and core competencies. Meeting these challenges and mastering these skills will be imperative to avoid being stuck on the dock as value-based payment models become increasingly predominant.

The Six Lenses

Success in population health management is achieved primarily through a focus on key success drivers and leadership. Improving the cost and quality of care requires taking enormous amounts of clinical and financial data and zeroing in on opportunities at the patient, provider, and organizational levels.

There is no perfect model for population health management, but successful organizations have shown that by viewing their populations and processes through these six analytical "lenses," they can attain an understanding of how to create and implement programs that drive clinical and financial improvements.

Cross-continuum cost and utilization metrics. A health system's ultimate goal for population health is to create a model that provides ready access to care resources and encourages patients to receive high-quality, appropriate care in the lowest-cost setting. In a value-based model, each time a patient receives a service-whether from a facility or a healthcare professional- the cost of that service reduces the potential shared savings.

Paid-claims data from payers are needed to determine the true cost of the care the patient received both within and outside the defined provider network. Such data are required from all sites of care to understand contract performance and to calculate key metrics such as per member per month costs and measurements of utilization in various care settings and within distinct categories (e.g., diagnostic, inpatient, outpatient, home health). These metrics should be tracked and compared with contractual expectations to enable timely correction of trends or outlier behaviors that negatively affect final population expenses.

Predictive modeling. Historical trends in patient care costs and utilization do not necessarily predict future expenses. However, predictive models provide insight that can help proactively identify which patients may be at greatest risk for high levels of utilization and cost in the near term.

Models can predict which patients are most likely to be hospitalized over the next six months, for example, or which are at risk for requiring excessively high-cost care over the next year (see the sidebar on page 64). Many models stratify the population, allowing organizations to assign the optimal intensity of care management resources to patients at each level of risk. Predictive models also can help organizations identify improvement opportunities by analyzing actual versus expected costs for individual patients and patient populations. Most predictive models have been based on claims data, but emerging models incorporate clinical data and social determinants to improve the accuracy of the forecasts.

Quality metrics. Virtually all value-based contracts include quality metrics that serve either as a prerequisite for receiving payment for cost savings or as a multiplier to the cost savings. Successful organizations use various tactics to optimize quality.

For example, these organizations focus particularly on quality metrics that align with the metrics in their payer contracts. When contracts with different payers include similar metrics but different outcome values (as may be the case with HbAiC values, for example), choosing the most stringent metric as the goal helps an organization and its providers avoid having to manage to different standards of care.

These organizations also provide up-to-date quality scores at the patient and population levels to providers and care teams. Access to paper or electronic reports of gaps in patient care is given to providers at the point of care, allowing them to implement appropriate interventions quickly. The providers also receive monthly reports to understand how they are managing the quality of care for the population as a whole.

Because the quality of data plays a large role in the measurements, successful organizations implement processes and workflows to ensure that data are captured completely and accurately at the point of care. Such data capture is optimally achieved either through direct interfaces with provider systems or through manual entry or CPT Category II coding.

Pharmaceutical utilization optimization. Often overlooked in population health management is the opportunity to include pharmaceutical expenditures in value-based contracts. Although pharmacy benefit managers provide some management of formulary compliance and generic utilization, cost savings of $35 or even $50 per member per month can be generated through additional optimization. Through a lens focused on drug spend, organizations can identify opportunities for generic substitution and for switching patients to therapeutically equivalent, less expensive drugs, and then implement programs to act on those opportunities. Taking these steps with even a handful of drugs can make a material difference, and- unlike in other areas of population health management-cost savings in drug spend do not have a negative revenue impact on the health system. For an example of how an organization can generate savings through pharmaceutical utilization optimization, see the exhibit on page 65.

Physician profiling. As organizations become more sophisticated in population health management, they begin to examine variations in physician practice patterns. Studies have disclosed tremendous variation in the cost of care among physicians. For example, a study by the RAND Corporation of 12,724 physicians across 27 specialties in Massachusetts stratified cost of care by physician years of experience.0 For physicians with less than 20 years of experience, the mean per capita cost was determined to be about $15,000, compared with about $10,000 for physicians with more than 40 years of experience. Although the study did not correlate cost with outcomes, the findings clearly demonstrate an opportunity to decrease variability in care.

The physician-profiling lens typically uses claims data that allow a physician's costs to be compared with those of peers and with external benchmarks. To ensure physician buy-in and avoid skewed results, the analysis should include risk adjustment and allow for outlier patients to be removed. A sophisticated method of attributing patients to physicians also is important, and the number of patients should be large enough for the analysis to be credible and instructive.

Through this lens, physician variability can be compared by total cost, specific categories of cost (e.g., inpatient utilization, outpatient utilization, diagnostic tests, pharmaceutical spend), and episodes of care (e.g., a particular surgical procedure or disease state). The intent is not to single out physicians who perform poorly, but to identify outliers and provide education or change clinical processes to enable performance improvement.

Network management. Network "leakage," which occurs when a patient receives care outside of the provider network, is even more important to manage under a value-based contract than under fee for service. The network loses the revenue for that care episode and has no control over the quality or cost of care provided.

Using a lens for network management, organizations track sites of care and out - of - network referrals to identify potential causes of leakage. It could be a result of self-referral (e.g., patients going to an out-of-network emergency department [ED]), community partner referral habits (e.g., patients being sent to a particular health system by a skilled nursing facility or home health agency), or, more commonly, personal preference or established referral relationships among providers. The data used to identify the causes of network leakage also disclose the most common procedures and illnesses for which patients receive care outside the network. This information gives the organization a greater understanding not only of where its patients are going, but also of service line gaps or access challenges within the organization that drive patients to leave the system.

Such information establishes a level of transparency that can be remarkably helpful, both in facilitating conversations with providers and community partners to ensure that all stakeholders are aware of the consequences of referring outside the network and in identifying a need to correct perceptions or issues regarding the quality and availability of in-network services.

Achieving Clarity with the Six Lenses

Data that drive population health analytics in many cases are incomplete and imperfect due to the challenges of obtaining comprehensive data and the imprecision of claims data. It is imperative to understand which aspects of population health analytics should be measured precisely and which can be more directional in nature to guide further analysis of underlying factors.

Financial and quality metrics should be precise because they are based on distinct events and are used in defining payments for value-based contracts. Conversely, physician profiling is imprecise, but when used in a sophisticated manner, it can shed light on variations in care among high-volume providers and the causes of such variations: Is variation due to a provider's unique population? Do some work processes fail to support efficient care? Do patients lack access to low-cost care settings? Are practice patterns outside acceptable standards of care?

Among the considerations health system administrators should keep in mind when using the six lenses to prepare for population health management are coding, physician attribution, and data acquisition.

Coding. Before value-based care, providers focused on coding the accurate level of care to optimize revenue. The exact diagnosis typically is less important in a fee-for-service bill, and the patient conditions and care documented in the clinical chart often do not match the final billing codes. Under value-based care models, diagnosis codes help health systems determine which patients to include in registries used to manage qualitymetrics, whicharekey components ofvaluebased contracts. In addition, diagnosis and procedure codes drive the risk-adjustment factors and hierarchical condition categories used in contracts with Medicare Advantage, Medicare Shared Savings, and other programs to adjust the shared-savings targets or capitated payments that providers receive for their patient populations. These factors can mean substantial differences in final payment totals.

As part of a population health program, healthcare organizations should invest in educating their clinicians and coding staff to accurately document and code all patient procedures, diagnoses, and comorbid conditions across all settings of care.

Physician attribution. In population health management, primary care physicians can be viewed as CEOs controlling millions of cost-of-care dollars, and specialists can be viewed as the primary drivers of cost for episodes of care. To accurately measure the impact of each physician, patients must be attributed to the correct one. This process can be tricky when care episodes involve encounters with multiple physicians.

A number of available algorithms can attribute patients based on visit history, type of provider, type of visit, and date of last visit. Additional complexities stem from the growth of team-based care and patient-centered medical homes in which a patient receives care from a team or clinic as opposed to a single physician. The keys to using this type of data are to have transparency in the algorithm to share with physicians, to allow providers to exclude patients when appropriate, and to establish realistic expectations about the use and limitations of the data.

Data acquisition. Successful organizations take an opportunistic approach to acquiring population health data, realizing that capturing all data is not yet possible.

Obtaining paid-claims data from payers is crucial. Although many payers include population metrics in reports, the raw claims data are needed to aggregate the information for utilization reports and analyses of data variation across payers, and to produce ad hoc analyses that are specific to an organization's needs. Paid-claims data can be supplemented with clinical and financial data from providers, such as preadjudicated claims, electronic health record (EHR) information, lab results, medications, immunizations, and any other data that can help bring clarity to the six lenses.

Successful organizations seek to implement analyses that are supported by available data instead of designing analyses and then determining whether data are available. Different data sources have distinct levels of complexity and value in helping to understand populations, and each requires an investment of financial and human resources to obtain and support ongoing acquisition. Most organizations take a prioritized, pragmatic approach in which they start with informative data sources that are of relatively low complexity (e.g., claims data, lab data), and expand over time to support increasingly comprehensive and sophisticated tools (e.g., EHR data, patient-reported data.)

Acting on the Information

Each of the six lenses informs different types of actions and interventions. Successful organizations not only have adopted these lenses to view data, but also have made substantial investments in an infrastructure to act on the data, knowing the information is only as valuable as its potential to affect outcomes.

For example, gaps in care identified through the quality-metrics lens can be addressed through outreach programs, provider education, and standardized work processes to support physicians at the point of care. The physicianvariability lens can provide a basis for engaging physician leaders to identify root causes and educate their peers on appropriate best-practice standards, and can help identify how organizational processes could be changed to better support physicians in a value-based care environment. Patients identified through the predictive-modeling lens as being high-risk should be enrolled in intensive caremanagement programs. An ED utilization report could highlight the need to open an urgent care center in a specific geographic region or to extend the office hours of a clinic. The network leakage report could reveal a primary care provider who continues to refer patients to an out-of-network orthopedist for hip replacements.

Surviving the Transition

Health systems today need robust clinical and financial analytics to thrive. Expense-reduction analyses, capacity planning, and cost-accounting tools are still necessary, but the movement to population health management requires investing in a new set of competencies. Preparation requires a new level of focus on individual patients, providers, and the network to identify areas where higher-quality care can be delivered more cost-efficiently.

The population health boat is moving quickly, yet these competencies take years for organizations to master. Given the pace of change, health systems and networks should develop the required skills now so they are positioned to succeed moving forward.

AT A GLANCE

Organizations should use six analytical lenses to prepare for population health management:

* Cross-continuum cost and utilization metrics

* Predictive modeling

* Quality metrics

* Pharmaceutical utilization optimization

* Physician profiling

* Network management

Learn how St. Vincent'sHealth Partners of Bridgeport, Conn., used an analytical lens to optimize the quality of its patient care in preparation for population health management at hfma.org/SVHPInitiative.

Population Health Risk Models

CFOs should become familiar with the uses, types, and limitations of risk models, which are critical tools for population health management. Most population health tools incorporate risk models for a variety of purposes, such as to select individuals for care management programs through stratification and segmentation of populations or to understand actual versus expected costs.

The majority of risk models are based on medical claims data, but they can also incorporate clinical, socioeconomic, demographic, and medication claims data.

There are two primary types of models.

Concurrent models. A concurrent risk score, also known as a retrospective risk score, represents a patient's risk level (i.e., expected expenditure relative to the whole population) during a previous time period. Concurrent risk scores typically use the previous 12 months of medical claims with or without pharmacy claims to gauge what costs 'should have been' during the period based upon the observed levels of risk. Fundamentally, concurrent models attempt to quantify differences in illness burden across patient populations. This type of analysis is usually used for risk adjustment in provider profiling or for assessing overall provider performance.

Prospective models. A prospective risk score predicts a patient's risk level, or expected cost, in the coming year. These models are typically based on illness burden as revealed by demographic data and diagnosis codes, and some also take into account historical utilization patterns. Unlike concurrent models, prospective models give more weight to chronic conditions that are likely to persist, and thus are used more widely in the care management process. These models are used for identifying patients who are candidates for active population health management and for setting payment levels or capitation rates.

Risk models have significant limitations. The precision of these models is affected by the completeness and accuracy of the data, the amount and types of data used as historical inputs, and the type of population being analyzed compared with the type of population used to create the models. A report issued by the Society of Actuaries in 2007, A Comparative Analysis of Claims-Based Tools for Health Risk Assessment (available at soa.org), compares the predictive power of various risk models and provides a useful overview.

Optimizing pharmaceutical spend often represents a substantial opportunity (or savings in risk-based contracts. Analytical tools can model drug interchange programs to identify patients with one or more medications that can be changed to on-formulary or other equally effective but less expensive therapeutic regimens. This hypothetical report details the number of possible interchanges per provider, which would be used to seek provider approvals for the recommended changes for each patient.

To accurately measure the impact of each physician, patients must be attributed to the correct one, which can be tricky when care episodes involve multiple physicians.

a. "Medicare Advantage Fact Sheet," Kaiser Family Foundation, May 1,2014.

b. Petersen, M., Gardner, P., T u, T., et al., Growth and Dispersion of Accountable Care Organizations: June 2014 Update, Leavitt Partners, June 2014.

c. Mehrotra, A., Reid, R.O., Adams, J.L., et al., "Physicians with the Least Experience Have Higher Cost Profiles Than Do Physicians with the Most Experience," Health Affairs, November 2012.

About the authors

Andrew Mellln, MD, MBA, is vice president and medical director, population and risk management, McKesson Technology Solutions, Minneapolis, and a member of HFMA's Minnesota Chapter (andrew.mellin@ mckesson.com).

Catrina Funk, MD, MBA, is vice president of medical affairs, population and risk management, McKesson Technology Solutions, Seattle ([email protected]).