Through programs such as the Comprehensive Primary Care (CPC) initiative and Comprehensive Primary Care Plus (CPC+), the Centers for Medicare & Medicaid Services (CMS) undertook a project by which they encouraged primary care practices to invest in “comprehensive primary care” capabilities. To assist CMS in designing alternative payment models (APMs) that adequately reimburse primary care practices for investing in these capabilities, this project developed and piloted a new method to estimate the expenses that practices incur to provide such capabilities.
A major goal of the project was to develop a method that could address a persistent question in estimating the costs of these capabilities: When various practices report widely divergent costs for a given comprehensive primary care capability (e.g., medication management), to what extent does the cost difference stem from different prices for the same capability, as compared to stemming from the practices having substantially different capabilities? Without detailed descriptions of the capabilities of each practice, this question cannot be answered empirically.
New Practice Expense Estimation Method for Comprehensive Primary Care Capabilities
We developed a mixed-methods strategy to collect cost and capability data in sufficient detail to inform CMS about payment for comprehensive primary care capabilities. Our strategy consisted of an initial interview with practice leaders to identify their comprehensive primary care capabilities and the types of costs associated with each capability, followed by researcher-assisted completion of a workbook tailored to each practice, which gathered data on the labor and nonlabor costs devoted to each comprehensive primary care capability. These were marginal cost estimates, net of any fee-for-service (FFS) revenues received. In a final brief interview, practice leaders reviewed summaries of the cost estimates and made corrections, if needed, before approving them.
To assess the validity of the labor estimates provided by practice leaders, we conducted interviews with frontline personnel and compared their capability-specific estimates of their own labor with those given by practice leaders. These interviews captured explanations for incongruous labor estimates.
Performance of the New Cost-Estimation Method
The cost-estimation method we developed required substantial time commitments from practice leaders and study staff. Even though participating practice leaders understood the importance of the study and were supportive of it, many also reported being overwhelmed by competing priorities, which prolonged data collection. The median interval between initial interview completion and final practice approval of the cost summary was 96 days (with a range of 40 to 232 days).
Data collection from practices affiliated with parent organizations (e.g., hospital systems, independent practice associations [IPAs]) was especially challenging. Gathering data from organization-affiliated practices involved more parties (e.g., both practice and organization-level leaders) and posed greater scheduling challenges than from independent practices. This prolonged data collection for practices affiliated with parent organizations (median 105 days) compared to independent practices (median 75 days).
Practices also had difficulty estimating the startup costs of comprehensive primary care capabilities. For capabilities that were adopted more than a year or two before data collection, practice leaders were frequently uncertain regarding one-time initial labor-hour expenditures (e.g., hours spent on staff training).
Practice leaders used widely divergent methods of estimating panel sizes (e.g., number of patients listed in the practice's EHR as being empaneled to a practice PCP; number of patients seen in a given year, regardless of whether they were truly considered to be on the practice's panel), and many reported high levels of uncertainty regarding their panel-size estimates.
Despite these challenges, participating practice leaders reported that the final capability descriptions were accurate and that their cost summaries had face validity. Several remarked that participation was an eye-opening exercise, generating insight into financial sustainability for their practices.
Costs of Comprehensive Primary Care Capabilities
Fifty practices, sampled for diversity across CPC+ participation status, geographic region, rural status, size, and parent-organization affiliation, completed the data-collection process. Practices varied considerably in the comprehensive primary care capabilities they adopted. The most-commonly adopted capabilities were empanelment (92 percent of practices), same-day or next-day office visits (90 percent), patient education and self-management support (88 percent), and software-based communication infrastructure (86 percent).
The costs of comprehensive primary care capabilities ranged widely. Medication management had the highest annual median cost per full-time-equivalent primary care practitioner (FTE PCP) ($11,496 per year), and extended hours had the lowest ($0), because the majority of practices offering extended hours did so without incurring marginal costs (i.e., they paid no overtime and had no unfilled appointment slots).
In general, cost variation among practices ostensibly providing the same comprehensive primary care capability (e.g., among the multiple practice-reported services categorized as “medication management”) was at least partially attributable to substantial differences in the level of service provided. However, price variation still played a role, such as when high-cost outlier practices appeared to offer the same service as lower-cost practices (e.g., because they used more expensive labor mixes). With a sample of 50 practices, we were unable to estimate quantitatively the relative contributions of level-of-service variation and price variation to the variation in overall costs.
For nearly all capabilities, labor expenses exceeded nonlabor costs, and ongoing annual costs exceeded one-time startup costs. However, practice leaders expressed doubts about the accuracy of startup-cost estimates, especially for costs that had been incurred years earlier.
Frontline personnel generally confirmed practice-leader reports (72 percent average agreement across capabilities) regarding their participation in each capability and the amount of time they devoted to it. However, in some instances, frontline participants disconfirmed practice-leader reports by indicating that they either (1) did participate in a capability when they were not reported to do so (13 percent on average), (2) did not participate in a capability when they were reported to participate (5 percent on average), or (3) did participate but to a different degree (most often with a discrepancy of less than 0.5 days per week) than was reported by practice leaders (11 percent on average).
In addition, capability descriptions provided by practice leaders were generally consistent with those given by frontline staff. Our goal was not to determine whether levels of agreement between practice leaders and frontline participants were adequate (which might depend on the intended use of such data); rather, our focus was on highlighting discrepancies—both to document them and to provide specific examples from frontline participants that explained why discrepancies occurred.
Considerations for Future Data Collection and Payment Policy
Even within a small sample of 50 practices, the heterogeneity we observed in the costs and content of comprehensive primary care capabilities highlights the importance of gathering detailed capability descriptions. Without a basis in such descriptions, linked to capability costs, APMs would risk underpaying for some capabilities (e.g., if payment amounts are insufficient to reimburse practices for a desired high level of service) and overpaying for others (e.g., when practices provide the same level of service, but payment amounts are skewed by high-cost outliers).
A similar mixed-methods approach to estimating the costs of comprehensive primary care capabilities, deployed on a larger scale than in the current study, could serve as a robust basis for future payment models that seek to incentivize and sustain comprehensive primary care. However, gathering these data is labor-intensive for both data collectors and practice leaders, with few opportunities for economies of scale.
To address the challenges of collecting such detailed data, future efforts could plan for longer data-collection periods, experiment with offering greater financial incentives for participation, try to provide other types of incentives, or explore the possibility of compulsory participation (giving financial compensation to the sampled practices). Additional methodological development might be necessary to better estimate the startup costs of comprehensive primary care capabilities, capture the costs borne by parent organizations, estimate patient-panel sizes consistently, and determine how much cost variation is attributable to differences in patient needs. Complementary methods, such as time and motion analysis, could help address the 28 percent disagreement rate between practice leaders and frontline staff regarding labor-cost estimates, thus improving the validity of those estimates.
Improving the accuracy and detail of the estimated expenses of comprehensive primary care capabilities would help CMS achieve its strategic goals for comprehensive primary care and serve as a resource for both practices and payers throughout the United States.