Physician Group Practices Accrued Large Bonuses Under Medicare’s Bundled Payment Model, 2018–20

Abstract

The Bundled Payments for Care Improvement Advanced (BPCI-A) alternative payment model, a voluntary Alternative Payment Model for Medicare, incentivizes hospitals and physician group practices to reduce spending for patient care episodes below preset target prices. The experience of physician groups in BPCI-A is not well understood. We found that physician groups earned $421 million in incentive payments during BPCI-A’s first four performance periods, 2018–2020. Target prices were positively associated with bonuses, with a mean incentive payment of $139 per episode in the lowest decile of target prices and $2,775 in the highest decile. In the first year of the COVID-19 pandemic, mean bonuses increased from $815 per episode to $2,736 per episode. These findings suggest that further policy changes, such as improving target price accuracy and refining participation rules, will be important as CMS continues to expand BPCI-A and develop other bundled payment models.

Introduction

The Centers for Medicare & Medicaid Services (CMS) is shifting towards payment models that aim to reward reductions in medical spending and improvements in care quality.¹ In 2018, CMS launched Bundled Payments for Care Improvement Advanced (BPCI-A), an opt-in alternative payment model for hospitals and physician group practices (PGPs). Based on historical spending patterns, BPCI-A assigns participants individualized target prices for spending on care “episodes.” Episodes begin with a patient encounter and end 90 days after discharge. If participants’ spending falls below their target, CMS rewards them with a financial bonus; if spending exceeds target, they owe CMS a penalty.

Although PGPs make up nearly 40% of BPCI-A participants, most work has focused on the model’s hospital participants.^2–10 How PGPs have fared in the model is incompletely understood. Of particular salience is the accuracy of target prices, which must be generous enough to entice participation in the opt-in model but strict enough to generate financial savings for CMS.^11,12 Miscalibrated targets may systematically advantage some participants and disadvantage others.^13,14 For example, hospitals that receive higher target prices (based primarily on historical spending patterns) tend to receive bonuses in BPCI-A, whereas hospitals with lower target prices tend to pay penalties.⁸ It is unknown whether this association holds for PGPs, for whom target prices are calculated slightly differently.¹⁵ Furthermore, prospectively set target prices may have failed to reflect unexpected disruptions in care utilization (and thus spending) during the COVID-19 pandemic. PGPs’ performance may thus have changed in important ways during this time. Moreover, CMS allowed providers to suspend participation in BPCI-A in 2020, and it is unclear how the PGPs that chose to remain in the model fared thereafter.¹⁶

Understanding PGP participation and performance has financial and policy implications for physician groups and policymakers in light of CMS’s 2022 announcement of a several-year extension of BPCI-A.¹⁷ This study thus had three aims: (1) to describe characteristics of PGPs participating in the first 4 performance periods of BPCI-A (2018 through 2020); (2) to determine associations between PGPs’ target prices and bundle selection with the receipt of bonuses or penalties; and (3) to investigate changes in PGP participation and performance during the COVID-19 pandemic.

Methods

Data

We evaluated PGPs during the first 4 performance periods of BPCI-A, spanning 27 months from 2018 through 2020. Each performance period lasted 6 months, except performance period 1, which lasted 9 months (Appendix Exhibit 1).¹⁸ Across BPCI-A, PGPs could choose to participate in any of several inpatient or outpatient bundles and receive “reconciliation payments” based on their performance in those bundles (Appendix Exhibit 2).¹⁸ A positive reconciliation payment is a bonus paid from CMS to the PGP; a negative reconciliation payment is a penalty paid by the PGP to CMS. CMS does not publicly report model performance for individual PGPs, and PGPs’ fluid billing arrangements (e.g., ability to participate in BPCI-A under new Taxpayer Identification Numbers [TINs] or submit Medicare claims through any number of TINs to which they have billing rights) hinder independent evaluations from claims data.^8,19,20 We therefore obtained PGP characteristics, participation data, and reconciliation payments from CMS through a Freedom of Information Act (FOIA) request. These data did not include the TINs under which PGPs participated in BPCI-A. These data, however, did detail the hospitals at which PGPs initiated inpatient episodes; we characterized these hospitals using information from the 2018 American Hospital Association (AHA) Annual Survey.²¹ AHA survey data were not available for 5 hospitals – out of the more than 1,000 admitting hospitals (Exhibit 1) – in our study.

Exhibit 1.

Characteristics of Physician Group Practices Participating in BPCI-A, patient episodes in bundles selected, and hospitals where episodes initiated, by performance period (2018–2020).

	Performance Period
Characteristic	Period 1	Period 2	Period 3	Period 4
PGPs
n	495	485	434	388
Number of bundles, mean	4.4	4.4	3.9	4.1
Target price, mean ($)	$29,535	$29,857	$31,685	$31,934
Reconciliation payments, mean ($)	$478,083	$528,802	$638,430	$466,532
Largest bonus ($)	$21,427,450	$19,172,394	$11,213,113	$11,034,380
Largest penalty ($)	−$9,674,314	−$8,668,145	−$5,583,052	−$4,882,179
Patient episodes
n	116,963	109,852	44,686	58,000
Per-PGP episode volume, mean (n)	236.3	226.5	103.0	149.5
Per-episode reconciliation payment, mean ($)	$815	$989	$2,736	$1,649
Admitting hospitals
n	1,038	1,040	883	849
Teaching (%)	54.7%	54.6%	57.2%	56.6%
Rural (%)	10.0%	10.9%	11.1%	11.2%
Size (%)
Small	30.5%	29.3%	27.4%	27.3%
Medium	30.5%	31.0%	31.9%	31.8%
Large	39.1%	39.7%	40.7%	41.0%
Ownership (%)
Public	7.9%	8.5%	8.3%	8.2%
Non-profit	66.9%	66.8%	64.9%	66.0%
Private	25.2%	24.8%	26.8%	25.9%
Region (%)
Northeast	12.8%	12.6%	10.7%	10.7%
Midwest	22.2%	23.3%	18.4%	19.4%
South	45.7%	45.0%	48.9%	49.6%
West	19.3%	19.1%	22.0%	20.3%

Source: CMS data obtained through a Freedom of Information Act request, linked to 2018 American Hospital Association Annual Survey data. Notes: BPCI-A = Bundled Payments for Care Improvement Advanced, PGP = physician group practice. Performance period 1 = 10/1/18 – 6/30/19, performance period 2 = 7/1/19 – 12/31/19, performance period 3 = 1/1/20 – 6/31/20, and performance period 4 = 7/1/20 – 12/31/20.

Outcomes

Our primary outcome was the average reconciliation payment that each PGP received for each care episode it initiated in its chosen bundles during performance periods 1 through 4.

Analysis

We first assessed baseline characteristics of BPCI-A PGPs and the hospitals at which they initiated inpatient episodes during performance periods 1 through 4. We then used linear regression models to test associations between PGPs’ target price deciles, adjusted for participation in specific bundles, and the magnitude of reconciliation payments received. This analysis did not adjust for additional patient or PGP factors, which are already accounted for in PGPs’ target prices.¹⁵ To address selection bias that may have arisen due to preferential dropout of PGPs during the program, we conducted sensitivity analyses restricted to PGPs that had participated in all 4 performance periods.

We then used linear regression models to test associations between PGPs’ participation in specific bundles with the magnitude of reconciliation payments received. We included bundles in which PGPs initiated more than 1 care episode during each performance period. We accounted for unobservable differences among PGPs and performance periods by including PGP and performance period fixed effects. Our analyses estimated cluster robust standard errors within hospitals and were weighted by patient volume within bundles. All analyses were conducted using Stata/BE version 17.0 (StataCorp LLC).

Limitations

This study has several limitations. First, participation in BPCI-A was voluntary, and results may not generalize to physician groups in other iterations of BPCI-A or in commercial bundled payment programs. Second, although this is the first peer-reviewed study to evaluate PGP participation and financial performance in CMS’s largest bundled payment model, including during the first year of the pandemic, we could not directly evaluate changes in care utilization or episode spending for PGP-initiated episodes, which represent important areas for future work, as these data were not available under the FOIA request. Third, in 2021, CMS modified the BPCI-A target price calculation methodology and providers were required to participate in groups of clinically related bundles rather than being allowed to selectively participate in individual, unrelated bundles. Changes to PGP participation and performance under these new incentives will require further study.⁹

Results

PGP participation

495 PGPs participated in performance period 1 of BPCI-A; participation decreased steadily to 388 in performance period 4 (Exhibit 1 and Appendix Exhibit 3).¹⁸ In performance period 1, PGPs participated in a mean of 4.4 bundles. The most frequently selected bundles were major joint replacement of the lower extremity (selected by 60.6% of participating PGPs), hip and femur procedures (27.9%), and major joint replacement of the upper extremity (24.4%) (Appendix Exhibit 4).¹⁸ PGPs initiated a mean of 236.3 patient care episodes across their bundles. The hospitals at which PGPs initiated inpatient episodes were mostly large (39.1%), nonprofit (66.9%), and teaching (54.7%) hospitals (Exhibit 1).

Performance period 3 (January 1, 2020 through June 30, 2020) marked the beginning of the COVID-19 pandemic and participants’ optional suspension of BPCI-A. In this performance period, the 434 PGPs that opted to continue in the model participated in a mean of 3.9 bundles (Exhibit 1). The bundles most frequently selected were major joint replacement of the lower extremity (selected by 36.2% of PGPs), sepsis (24.4%), and congestive heart failure (22.6%) (Appendix Exhibit 4).¹⁸ PGPs initiated a mean of 103.0 episodes across bundles during this time. There were minimal changes in the characteristics of hospitals at which PGPs initiated inpatient episodes, although lower proportions of episodes were initiated at hospitals in the Northeast and Midwest and higher proportions were initiated in the South and West (Exhibit 1).

Target prices and reconciliation payments

There was a positive relationship between PGPs’ predetermined target prices and reconciliation payments received (Exhibit 2). PGPs in the lowest decile of target prices received a mean reconciliation payment of $139 (95% confidence interval [CI]: -$207, $485) per episode, while PGPs in the highest decile received a mean per-episode reconciliation payment of $2,775 (95% CI: $2,312, $3,236). A similar pattern was observed across individual performance periods (Appendix Exhibits 5 and 6).¹⁸

Exhibit 2.

Mean BPCI-A reconciliation payments to Physician Group Practices by target price decile, performance periods 1 through 4 (10/1/18–12/31/20).

Source: CMS data obtained through a Freedom of Information Act request.

Notes: BPCI-A = Bundled Payments for Care Improvement Advanced, PGP = Physician Group Practice. Target price decile 1 = lowest, decile 10 = highest. Lines indicate 95% confidence intervals. Positive reconciliation payments signify bonuses paid to PGPs, while negative reconciliation payments signify penalties paid by PGPs.

Bundle selection and reconciliation payments

The magnitude of reconciliation payments differed widely across the bundles in which PGPs chose to participate (Exhibit 3). For example, episodes initiated in congestive heart failure and acute myocardial infarction bundles were associated with mean per-episode reconciliation payments that were more than twice as large as those associated with the percutaneous coronary intervention or cellulitis bundles. Across all performance periods, incentive payments were lowest for back and neck procedures (mean per-episode incentive payment = -$1,031, 95% CI: -$4,126, $2,064) and highest for spinal fusions ($9,180, 95% CI: $5,203, $13,157).

Exhibit 3.

Mean BPCI-A reconciliation payments to Physician Group Practices, by bundle selected, performance periods 1 through 4 (10/1/18–12/31/20).

Source: CMS data obtained through a Freedom of Information Act request.

Notes: BPCI-A = Bundled Payments for Care Improvement Advanced, PGP = Physician Group Practice. Lines indicate confidence 95% intervals. Positive reconciliation payments signify bonuses paid to PGPs, while negative reconciliation payments signify penalties paid by PGPs.

PGP performance before and after the COVID-19 pandemic

Across all 4 performance periods, CMS paid a net of $421,923,168 in financial bonuses to PGPs participating in BPCI-A (Appendix Exhibit 7).¹⁸ In performance period 1, the mean reconciliation payment was $815 per care episode and $478,083 per PGP (Exhibit 1). The magnitude of reconciliation payments increased following the start of the COVID-19 pandemic in performance period 3, when the mean reconciliation payment was $2,736 per episode and $638,430 per PGP (Exhibit 4).

Exhibit 4.

Mean BPCI-A reconciliation payments to Physician Group Practices, total and per episode, by performance period (2018–2020)

Source: CMS data obtained through a Freedom of Information Act request.

Notes: BPCI-A = Bundled Payments for Care Improvement Advanced,

Similarly, while participation in some bundles was associated with the receipt of a negative mean reconciliation payment, i.e., a penalty, in performance periods 1 and 2 (Appendix Exhibit 8),¹⁸ participation in most bundles was associated with positive reconciliation payments, i.e., bonuses, in performance periods 3 and 4 (Appendix Exhibit 9).¹⁸ The increase in mean per-episode reconciliation payments in performance periods 3 and 4 was observed across the range of target prices (Appendix Exhibit 6).¹⁸ The values of per-episode reconciliation payments were not substantially different in sensitivity analyses restricted only to those PGPs that participated in all 4 performance periods (Appendix Exhibits 10 and 11), suggesting that these PGPs had not received systematically higher per-episode reconciliation payments compared to the overall population of PGPs that participated for at least 1 year in the program.¹⁸

Discussion

To our knowledge, this is the first peer-reviewed study to examine the experience of physician groups in BPCI-A, CMS’s flagship bundled payment program. We found that, during the first 4 performance periods of BPCI-A, participating PGPs accrued large financial bonuses under the model. This is consistent with the dispensation of considerable financial bonuses to PGPs under BPCI-A’s predecessor and other episode-based payment programs.^22,23 This is also consistent with the considerable financial bonuses received by hospitals participating in BPCI-A.^8,10 Furthermore, we found that PGPs that were preassigned higher target prices tended to receive larger bonuses compared to PGPs with lower target prices. The magnitude of reconciliation payments was similarly asymmetric across individual bundles chosen for participation. Several of the most commonly selected bundles were associated with a bonus of several thousand dollars per patient care episode. Lastly, the PGPs that chose to remain in the program during the COVID-19 pandemic received larger financial bonuses for each care episode and larger bonuses overall, compared to the pre-pandemic cohort.

Our findings have several financial and policy implications for physician group participation in BPCI-A. We found that being preassigned higher target prices was associated with the accrual of larger bonus payments. Higher target prices were associated with greater bonuses likely because they offered a more achievable goal for reaching targets through a combination of spending reductions and mean reversion.^{8,10,11,15,19} This finding is consistent with prior work demonstrating an association between target prices and bonuses among BPCI-A hospitals, for whom target prices were calculated slightly differently than for PGPs.^8,15 Our findings build upon evaluations commissioned by CMS by investigating PGPs’ heterogeneous performance across clinical bundles and demonstrating this linear relationship between target prices and reconciliation payments. While CMS has acknowledged that target prices in general were too high, it is important for policymakers to note that episodes in the lowest decile of target prices experienced a mean per-episode reconciliation payment of $193, while those in the highest decile of target prices experienced a mean per-episode reconciliation payment of $2,775.

Together, our findings suggest that target spending prices were miscalibrated for both hospital and PGP participants and highlight the nearly deterministic effect of target prices on financial rewards and penalties in BPCI-A and more generally in episode-based payment models.^11,24 High-spending participants in BPCI-A were assigned higher target prices and thus disproportionately received bonuses.⁸ In other payment models where spending benchmarks are based on regional spending trends and not on each participant’s own historical spending, high-spending participants disproportionately fail to meet the regional benchmark and are penalized.¹⁴

Importantly, BPCI-A modified its target price calculation methodology in 2021 — after the conclusion of our study period — by incorporating a retrospective adjustment to set more accurate target prices.²⁵ Under this rule, preliminary target prices preassigned to participants before each performance period are modified post-hoc by up to 10% based on spending trends of participants’ peers during the performance period. Beginning in 2021, BPCI-A participants were also required to choose whole “service lines” of clinically related bundles, rather than being allowed to selectively participate in unrelated bundles with the most enticing target prices. While it is unknown whether these policy changes were associated with a more balanced distribution of incentive payments for PGPs across the range of target prices, it is already apparent that participation in the model has decreased after these changes. 317 PGPs participated (22,131 patient episodes) in performance period 5 (Model Year 4) declining to 245 (18,827 patient episodes) in performance period 7 (Model Year 5).²⁴ This dropout highlights the fundamental issue that plagues voluntary, opt-in payment models. CMS must set target spending prices that are high enough to be achievable and thus attract participants, but low enough to generate savings for CMS.^11,24 As long as providers are able to cease participation when target prices or participation conditions become too unfavorable, CMS will likely continue to struggle to set appropriate spending benchmarks in future episode-based payment models, BPCI-A or otherwise.^11,24

We also found that PGPs – across the range of target prices – accrued much greater financial bonuses during the first year of the COVID-19 pandemic compared to the pre-pandemic cohort. The PGPs that chose to remain in BPCI-A at the start of the pandemic, for example, may have differed in significant and advantageous ways from those that opted to suspend participation. However, analyses restricted to these PGPs that participated in all 4 performance periods did not demonstrate advantageous performance compared to the overall pre-pandemic cohort.

On the other hand, the increase in bonuses to PGPs during the first year of the COVID-19 pandemic likely resulted from temporary changes in care utilization. Participants in bundled payment models generally reduce episode spending, and thus accrue bonuses, by reducing utilization of post-acute care (PAC), such as skilled nursing and inpatient rehabilitation facilities.^6,11,26–29 Notably, PAC faced widespread disruptions in 2020 and beyond, with health systems shifting towards less frequent utilization of PAC during the early phase of the pandemic.^30–32 This systems-wide decrease in care utilization was likely associated with 90-day episode spending that was well below PGPs’ historical baselines and thus below their target prices, allowing for the accrual of larger financial bonuses during this time. Evaluating whether rebounds in PAC usage after 2020 were associated with decreases in incentive payments to pre-pandemic baselines remains an important direction for future research.

Conclusion

We found that physician group practices that participated in the first 4 performance periods of BPCI-A accrued large financial bonuses under the model, amounting to $421 million in incentive payments. PGPs that were preassigned higher target prices tended to receive larger bonuses compared to those with lower target prices. Bonuses to participants increased drastically during the first year of the COVID-19 pandemic. These findings suggest that further policy changes, such as improving target price accuracy and refining participation rules, will be important as CMS continues to expand BPCI-A and develop other bundled payment models.