Empanelment Guide

The basics of empanelment have remained unchanged for many years. Evolution of the approach will likely be driven by larger forces impacting how primary care is conceptualized, delivered and paid for.

Shift to Whole-Person Care

The movement to address whole-person health in primary care has been on the rise for many years, building from integration of behavioral health to a more recent focus on systematically responding to health-related social needs in addition to physical wellbeing.^[1],[2] Although there is a deep, wide and growing body of literature on the importance and impacts of responding to behavioral and social health challenges in primary care, the implications for empanelment are less well established. While demographic, diagnostic and risk score data have been used to weight panels according to anticipated utilization, these methods account for medical complexity alone.^[3] There is a great deal yet to learn about accounting for behavioral health or social complexity in establishing panel risk adjustment and weighting methods. Wide variation in standards and lack of evidence about optimal panel size persist.^[4],[5] As additional individual and population-level insights accrue relating to behavioral and social health risk factors, methods to adjust panel size based on patient complexity may benefit from integration of these variables.

Virtual Care Opportunities and Challenges

Historically, calculating supply and demand has been based on face-to-face interactions, even as virtual care visits have increased.^[6] Even prior to the enormous expansion of virtual care accelerated by the COVID-19 pandemic, these activities already consumed a significant percentage of both perceived and actual physician time and capacity.^[7],[8] In this environment of rapid change to telehealth implementation, saturation and reimbursement, the best formats and use of virtual primary care are still being defined.^[9]

Nonetheless, as the use of synchronous telehealth and asynchronous virtual care grows, so will the importance of accounting for this workload on the supply of provider and care team time. This is an area of ongoing development, as standard methods of accounting for the total patient care workload are yet to be defined, tested and disseminated.^[10] In the meantime, practices may want to consider using technology to leverage both rising insight into patient behaviors and additional options for visit formats, such as the use of machine learning by Urban Health Plan, a New York community health center, to identify patients with moderate to high risk for appointment no-shows. Following this identification strategy, Urban Health Plan developed targeted interventions to reach people for same-day conversion to telehealth visits and increased the show rate for patients most likely to miss their visits by a stunning 154%.^[11]

Artificial Intelligence Implications for Supply and Demand Management

Machine learning is “the fundamental technology required to meaningfully process data that exceed the capacity of the human brain to comprehend,” where a computer model uses algorithms to learn from a large data set of examples rather than operate on the basis of discrete rules. This capability allows the model to absorb and learn from enormous data sets, thereby acquiring the ability to perform tasks far more complex than simple rules-based coding would enable.^[12]

Machine learning models are well suited to improve predictive accuracy of capacity planning for empanelment by using large retrospective data sets of healthcare utilization behaviors for an accountable population. Although the efforts are nascent, the use of machine learning to enhance supply and demand management for primary care has begun, using historical utilization data to estimate the amount of primary care effort required to care for diverse population segments.^[13]

In addition to improving panel weighting methods using machine learning to estimate demand, artificial intelligence is likely to significantly change the provision of primary care in a variety of ways, with implications for estimates of supply as well.^[14] There are a range of ways that primary care may be augmented by artificial intelligence with some features creating greater impact on the supply of provider and care team time. For example, the use of artificial intelligence for medical advice and triage, diagnostics, digital health coaching, clinical decision-making and documentation could all free up substantial provider and care team time to interface with patients, which would have tremendous impact on the current formulae for calculating provider supply relative to patient demand.^[15]

The Urban Health Plan example in the previous section is an excellent demonstration of early adoption of artificial intelligence to better understand patient demand patterns, as well as creative use of virtual care technology to diversify the formats for healthcare supply. As artificial intelligence technologies are more widely adopted and studied in Empanelment Implementation Guide 28 primary care settings, the implications for empanelment and panel management, both functionally and mathematically, will become more clear.