Electronic Health Record-Based Risk Stratification: A Potential Key Ingredient to Achieving Value-Based Care

In preparation for value-based care (VBC) reimbursement and risk contracting, a significant proportion of health systems and physicians in the United States are engaging in population health management (PHM) activities, such as creation of patient-centered medical homes and accountable care organizations. In addition to this move from fee-for-service to value-based reimbursement, increase in health care demand from such factors as the aging population and ongoing COVID-19 pandemic have providers expressing urgency in developing PHM strategies to meet the health needs of all patients, COVID and non-COVID, and deliver quality care at low cost to all patient populations. ¹

A successful PHM strategy consists of the ability to stratify patients who are at high risk for adverse outcomes, identifying gaps in care coordination, and effectively implementing interventions to manage those patients, such as team-based care. Risk stratification (RS) models assign an individual to different health risk strata, using various data sources representing clinical, lifestyle, and social factors. RS is often used to align providers with VBC approaches through retrospective analysis of risk-adjusted payments as well as to prospectively develop effective PHM interventions.

Traditionally, health insurance plans–with access to patient claims data, analytics resources, and case managers–have played a major role in leveraging RS models for management of their beneficiaries. Most health plans have integrated specialized tools that use an individual's diagnosis and medication information over a given period to assign an individual a risk level. Estimated risks usually include health care medical/pharmacy costs or utilization (eg, hospitalization, emergency room visits). However, this approach may have limited effectiveness for managing patients of a health system because there is less availability of insurance claims data to providers in care settings.

Data captured in electronic health records (EHRs) is increasingly seen as a potential substitute to claims for undertaking RS modeling. Previous studies have shown that, even though the diagnosis and medication data contained in claims and EHRs differ by conditions, EHR-based models may perform comparably to claims-based predictive models to stratify risk in a given population. ² Moreover, EHR data obviate the need to reconcile claims from different payers covering various patient populations seen by health care organizations to develop risk models, which is complex, costly, and not readily available to many providers. Another distinguishing characteristic of EHR data is the real-time availability of patient care information allowing development of more time-sensitive or near-term risk models, such as 30-day readmissions. ³ Hence, EHR-based risk models can be clinically relevant for health care providers compared to claims-based models that may be outdated by the time they are available at the point of care.

Most EHRs are designed to capture data on demographics, encounters, diagnoses, medications, and procedures that also are available from claims data. However, EHRs capture additional data–such as vitals, disease acuity and severity, laboratory results, provider- and patient-reported data on health behaviors, sign and symptoms, and clinical outcomes–that can be leveraged to develop more accurate risk prediction models. RS models developed using EHR data also can be rolled out more efficiently and effectively in health systems as minimal effort is needed to integrate results into decision support tools and PHM platforms built in the existing EHR systems. Indeed, an increasing number of EHR vendors are providing RS modeling and population registry architecture as a part of their PHM applications offerings, which is accessible at the point of care.

EHR data come with challenges that need to be addressed to undertake more effective and scalable RS activities. Data captured in the EHR are limited to patients who come in contact with the health care system, thus making it challenging to develop comprehensive RS models that capture all covered lives. Moreover, a combination of EHR application, health care system, and patient factors contribute to various well-documented limitations to leveraging EHR data for prediction modeling, ranging from data quality issues (eg, errors, inconsistencies, missingness, unstructured data) to interoperability challenges of sharing data consistently across different organizations. ⁴

With the possibility of leveraging EHRs for RS, health care providers need to facilitate development of an accurate, timely, and clinically relevant population-based data infrastructure. First, there is a need to evaluate the quality of EHR data and develop a standard approach for data evaluation before use for RS. Building on and enhancing the 2009 Meaningful Use program of Health Information Technology for Economic and Clinical Health Act, the 21^st Century Cures Act final rule of 2020 outlines a minimum set of policies and technical standards to advance interoperability and health information exchange between health care organizations with EHRs. It is anticipated that standardized data and formats of health concepts such as medications, problems/diagnosis, procedures, and utilization will be adopted in EHRs across health care settings (eg, outpatient, inpatient) soon, leading to more complete data being available for all patient populations. Increased quality EHR data can be leveraged to build generalizable RS models and track health outcomes of all patients across health care organizations as a part of the VBC delivery goal.

Second, as EHRs store information in different proprietary data structures, developing common data models that are agreed on by all stakeholders and led by health care providers will be key in the development and adoption of generalizable EHR-based RS models. Well-established technical solutions, such as Common Data Models (CDMs) and Minimum Data Sets (MDSs), should be implemented for system-wide ability to use EHR data to risk stratify patients. CDMs lead to normalization of clinical data to standard terminologies and formats that enable the ability to develop a risk model in one health setting and apply it in another setting with minimal modifications. Some examples of widely used CDMs are Observational Medical Outcomes Partnership, Health Care Systems Research Network, and PCORnet CDM. ⁵ MDSs help in setting the standard definition of essential clinical data elements for a range of RS analyses that are important to a given health care provider. ⁶

Third, although using diagnosis and pharmacy information for RS has been widely studied, the integration of other EHR data types (eg, laboratory testing, vital signs) into actionable RS models will require further evaluation. VBC requires providers to shift from reactive to more proactive care models. An initial step toward VBC goals can be the development and implementation of RS models using all EHR data types that provide additional insights to health care providers to develop timely and actionable interventions at the point of care. ⁷ Moreover, with the integration of disparate data sources such as social determinants of health and patient-reported data in EHRs, RS models can be built to identify areas in which health care systems may be able to offer assistance by working with community organizations and social service agencies to undertake PHM interventions that are more comprehensive and equitable.

A comparison of the advantages and challenges of using EHR data versus claims data is shown in Table 1.

As health care organizations are dealing with the COVID crisis, an effective PHM strategy is increasingly needed to assure improved outcomes guided by VBC principles. RS is a key analytic element of the PHM effort; however, traditionally, population-level risk modeling has been limited to health plans using claims. Recent evidence has shown the value of EHRs in providing adequate data quality to be used for RS. Health care providers should embark on using their EHR data prudently for RS to improve their PHM efforts for all patients, now and in the foreseeable future.