Table 1.
Adjusted differential changes in total Medicare spending per beneficiary in ACOs vs. control group for post-contract years 2013 and 2014a
| Difference between ACOs and control group in pre-contract period, $ (95% CI) | Difference in trend between ACOs and control group in pre-contract period, $/year (95% CI) | Differential change for ACOs vs. control group | ||||
|---|---|---|---|---|---|---|
| 2013 performance year | 2014 performance year | |||||
| Estimate, $ (95% CI) | P value | Estimate, $ (95% CI) | P value | |||
| 2012 MSSP cohort (N=114 ACOs) | 135 (−78,348) | −21 (−76,35) | −146 (−274,−18) | 0.03 | −264 (−398,−130) | <0.001 |
| Subgroups b | ||||||
| Organizational structure: | ||||||
| Independent physician groups (N=69) | 33 (−322,387) | −22 (−111,66) | −217 (−449,15) | 0.07 | −334 (−559,−110) | 0.004 |
| Hospital-integrated (N=45) | 166 (40,292) | −14 (−68,41) | −48 (−144,47) | 0.32 | −179 (−285,−72) | 0.001 |
| Baseline spending: | ||||||
| Above regional average (N=57) | 501 (149,852) | 9 (−62,80) | −228 (−434,−21) | 0.03 | −371 (−586,−156) | 0.001 |
| Below regional average (N=57) | −266 (−398,−134) | −43 (−105,19) | 3 (−102,109) | 0.95 | −105 (−205,−4) | 0.04 |
| 2013 MSSP cohort (N=106 ACOs) | 3 (−113,119) | −8 (−44,28) | 3 (−100,106) | 0.96 | −94 (−194,6) | 0.07 |
| Subgroups b | ||||||
| Organizational structure: | ||||||
| Independent physician groups (N=59) | 52 (−133,238) | −32 (−84,21) | −192 (−339,−46) | 0.010 | −175 (−341,−9) | 0.04 |
| Hospital-integrated (N=47) | 38 (−74,150) | 14 (−29,58) | 99 (−17,215) | 0.09 | −66 (−197,65) | 0.32 |
| Baseline spending: | ||||||
| Above regional average (N=53) | 292 (159,426) | −17 (−62,27) | −25 (−154,103) | 0.70 | −156 (−285,−27) | 0.02 |
| Below regional average (N=53) | −336 (−508,−164) | 21 (−28,70) | 27 (−120,174) | 0.72 | −27 (−189,134) | 0.74 |
| 2014 MSSP cohort (N=115 ACOs)c | 20 (−118,158) | −1 (−27,26) | - | - | −49 (−137,38) | 0.27 |
All estimates are adjusted for hospital referral region (HRR), HRR by year fixed effects to control for local changes over the study period, and the following sociodemographic and clinical characteristics of patients: age, sex, race and ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, or other), Medicaid coverage, disability as the original reason for Medicare eligibility, long-term nursing home residence, end-stage renal disease, indicators of conditions from the Chronic Condition Data Warehouse (CCW) being present at the start of the study year, indicators of multiple conditions, hierarchical condition category (HCC) risk score determined from the prior year of claims, and average educational attainment and poverty rates assessed at the level of patients' zip code tabulation area of residence. Estimates for 2013 were previously published,1 but 2013 estimates presented in Table 1 differ slightly from previously reported estimates because beneficiaries assigned to ACOs entering in 2014 were removed from the control group. Unadjusted estimates of differential changes were similar to adjusted estimates, and sensitivity analyses indicated that changes over the study period in composition of physicians billing under ACO taxpayer identification numbers did not significantly affect estimates.
Subgroup analyses were conducted for ACO cohorts with overall spending reductions in 2014 (the 2012 and 2013 cohorts). We categorized ACOs as financially integrated with hospitals vs. independent physician groups using CMS descriptions and information on participating organizations' websites.1 We categorized ACOs as having above vs. below regional average baseline spending by comparing whether risk-adjusted spending for its attributed beneficiaries was above or below risk-adjusted spending for the control group in its service area.1 To eliminate bias from regression to the mean, we used claims from 2008 (prior to the study period) to assess baseline spending. Estimates for all subgroups in a cohort were produced by the same model; thus, differences in spending reductions between ACOs with different organizational structures were adjusted for differences in spending reductions related to differences in baseline spending. Subgroup estimates were also adjusted for differences in spending reductions between ACOs in areas with high vs. low areas, which were not statistically significant.
Four ACOs were excluded from the 2014 ACO cohort because they had previously participated in the Pioneer ACO model.