Long-Term Spending of Accountable Care Organizations in the Medicare Shared Savings Program

Abstract

Importance: Evidence from initial cohorts of accountable care organizations (ACOs) participating in the Medicare Shared Savings Program (MSSP) found modest reductions in health care spending. Little is known about whether these effects have changed over time.

Objective: To determine long-term changes in spending for MSSP ACO participants.

Design, setting, and participants: Using 2010 to 2019 traditional Medicare data, difference-in-differences analyses were performed to compare spending changes for patients attributed to ACOs relative to changes for patients at non-ACO organizations. Outcomes included total Medicare spending and spending by category. Three- and 6-year effects and estimated differential changes overall and by ACO characteristics were calculated, including size (small defined as <10 000 patients), rurality, and whether an ACO included a hospital (hospital-associated ACO) or not (physician-group ACO).

Exposure: Attribution to a medical group or clinic in an ACO during the first 2 years of ACO tenure.

Main outcomes and measures: Total annual per-patient Medicare spending.

Results: The sample included 41 973 272 Medicare patient-years. Baseline characteristics for 2 719 406 ACO patients and 5 523 652 control patients were similar (average age, 72 years; 58% female; and 82% to 84% White) prior to ACO formation in 2010 and 2011, and unadjusted annual per-patient spending was slightly lower in the ACO group vs control group ($12 147 vs $12 318; difference, -$171 [95% CI, -$223 to -$118]) in the 2 years prior to ACO formation. ACO formation was associated with a mean differential reduction of $142 (95% CI, -$193 to -$92) in annual per-patient spending over 3 years and $294 (95% CI, -$347 to -$241) over 6 years. Spending reductions associated with ACO formation increased over time: compared with control patients, ACO patients experienced a mean reduction of $234 (95% CI, -$298 to -$171) in year 3 and $584 (95% CI, -$680 to -$489) in year 6. Physician-group and small ACOs generated larger spending reductions. Spending changes resulted in $4.1 billion to $8.1 billion in savings to Medicare between 2012 and 2019.

Conclusions and relevance: During the MSSP's first decade, ACOs generated meaningful reductions in spending, with larger effects over time.

Figure 1.. Flow of Study Population for Accountable Care Organization (ACO) and Control Groups

^aSee eMethods in Supplement 1. ^bPractice defined by Tax Identification Number and clinic by Centers for Medicare & Medicaid Services Certification Number. ID indicates identification.

Figure 2.. Differential Change in Medicare Spending Between Accountable Care Organization (ACO) and Control Groups, 3 and 6 Years After ACO Formation

^aAll spending was inflation-adjusted to 2019 dollars, using the Consumer Price Index for Medical Care. ^bAverage annual per-beneficiary spending for ACO-attributed patients in the second year prior to ACO formation. ^cAverage annual per-beneficiary spending for ACO-attributed patients during the first 3 or 6 years after ACO formation, including year of formation. ^dAverage annual per-beneficiary spending for non-ACO–attributed patients (control group) for each ACO cohort in the second year prior to ACO formation. For example, non-ACO observations in 2012 would represent control observations in the year of ACO formation for the 2012 cohort, 1 year prior to ACO formation for the 2013 cohort, and 2 years prior to ACO formation for the 2014 cohort. Non-ACO observations were weighted to align with the size of the ACO cohort. ^eAverage annual per-beneficiary spending for non-ACO–attributed patients (control group) for each ACO cohort during the first 3 or 6 years after ACO formation, including year of formation. Non-ACO observations were weighted to align with the size of the ACO cohort. ^fDefined as annual Parts A and B Medicare spending (allowed amounts), excluding spending on durable medical equipment. ^gIncludes other nonfacility spending billed in locations not included in the listed categories, such as ambulance transport, chiropractic care, and vision, hearing, and speech services. ^hDifference-in-differences change in annual spending during first 3 or 6 years after ACO formation (including year of formation) for ACO-attributed patients vs non-ACO–attributed patients within the same hospital referral region. Changes were estimated using the Sun and Abraham package with indicators for year, ACO, and year since formation, as well as patient characteristics, including age, sex, race and ethnicity, dual eligibility status, disabled status, end-stage kidney disease status, and Hierarchical Condition Category risk score (Supplement 1).

Figure 3.. Difference in Annual Medicare Spending per Beneficiary Between Accountable Care Organization (ACO) and Control Groups by Year

A, Unadjusted mean spending for ACO-attributed patients and non-ACO–attributed patients by year. ACO group includes ACOs that were formed between 2012 and 2017. B, Adjusted estimates of the differential change in total Medicare spending between ACO-attributed patients and non-ACO–attributed patients (control) in a given year relative to ACO formation. Changes were estimated using the Sun and Abraham package with indicators for year, ACO, and year since formation, as well as patient characteristics, including age, sex, race and ethnicity, dual eligibility status, disabled status, end-stage kidney disease status, and Hierarchical Condition Category risk score (Supplement 1). The reference year is 2 years prior to ACO formation. The year prior to formation (gray triangle) was considered a washout period. Estimates include ACOs that formed between 2012 and 2017. For interpretation, the estimate for year 0 relative to ACO formation represents the estimate during the first year after ACO formation, the estimate for year 1 relative to ACO formation represents the estimate during the second year after ACO formation, etc. Bars represent the 95% CIs. Spending was inflation-adjusted to 2019 dollars, using the Consumer Price Index for Medical Care.

Figure 4.. Differential Change in Medicare Spending 3 and 6 Years After Accountable Care Organization (ACO) Formation by ACO Subgroup

Because all non-ACO observations were included as the comparison group for each ACO subgroup, non-ACO spending is the same for each row in columns with blank cells. ^aAll spending is inflation-adjusted to 2019 dollars, using the Consumer Price Index for Medical Care. ^bAverage annual per-beneficiary spending for ACO-attributed patients in second year prior to formation. ^cAverage annual per-beneficiary spending for ACO-attributed patients during first 3 or 6 years after formation, including year of formation. ^dAverage annual per-beneficiary spending for non-ACO–attributed patients for each ACO cohort in second year prior to formation. For example, non-ACO observations in 2012 represent control observations in the year of ACO formation for the 2012 cohort, 1 year prior to formation for the 2013 cohort, and 2 years prior to formation for the 2014 cohort. Non-ACO observations weighted to align with size of ACO cohort. Unlike ACO group, all non-ACO observations included. ^eAverage annual per-beneficiary spending for non-ACO–attributed patients for each ACO cohort during first 3 or 6 years after formation, including year of formation. Non-ACO observations weighted to align with size of ACO cohort. Unlike ACO group, all non-ACO observations included. ^fSmall ACOs defined as those with fewer than 10 000 attributed patients. ^gRural ACOs defined as those with at least 25% of practices or clinics located in nonmetropolitan areas (using rural-urban commuting area codes). ^hDifference-in-differences change in annual spending during first 3 or 6 years after ACO formation (including year of formation) among ACO-attributed patients vs non-ACO–attributed patients within same hospital referral region. Changes estimated using the Sun and Abraham package with indicators for year, ACO, and year since formation, as well as patient characteristics, including age, sex, race and ethnicity, dual eligibility status, disabled status, end-stage kidney disease status, and Hierarchical Condition Category risk score (Supplement 1). Only ACO observations in the relevant subgroup were included in model; all non-ACO observations included in model.