Table 2.
Variable | Regression coefficient | [95% Confidence interval] |
---|---|---|
ACO characteristics | ||
Hospital-managed (ref: physician-led) | 0.111a | [0.092, 0.130] |
No. of years in MSSP | −0.004 | [−0.023, 0.015] |
Two-sided risk (ref: 1-sided risk) | 0.014 | [−0.006, 0.035] |
Advance payment or ACO investment (ref: no participation) | −0.008 | [−0.128, 0.111] |
No. of assigned beneficiaries (1000’s) | 0.002a | [0.001, 0.003] |
No. of distinct participating organizations (100’s) | −0.035b | [−0.059, −0.011] |
Weighted HCC risk score | −0.422a | [−0.546, −0.299] |
Beneficiary concentration in ACO service counties (HHI) | 0.129b | [0.047, 0.211] |
Home health agency expenses to total health expensesc | −0.057 | [−0.446, 0.332] |
SDOH | ||
Economic well-being | ||
Unemployment ratec | −1.809a | [−2.834, −0.784] |
Median household income ($10 000’s) | 0.023d | [0.000, 0.045] |
Transportation convenience | ||
Commute by public transportationc | 0.442 | [−0.175, 1.058] |
Mean commute time (10 min) | −0.059 | [−0.154, 0.035] |
Healthy food consumption | ||
Low-income population with grocery store accessc | 3.978b | [1.089, 6.867] |
Per capita expenses on fast food ($100’s) | −0.008 | [−0.058, 0.042] |
Health resource availability | ||
Hospital-based registered nurse to bed ratio | 0.108 | [−0.033, 0.249] |
Observed/expected ratio of primary care checkups | 0.570b | [0.180, 0.961] |
Average distance between health care organizations (10 miles) | −0.010 | [−0.023, 0.003] |
Abbreviations: ACO, accountable care organization; HCC, Hierarchical Condition Category; HHI, Herfindahl-Hirschman Index; MSSP, Medicare Shared Savings Program; SDOH, social determinants of health.
Source: Authors’ analysis of data from the MSSP public use files, American Community Survey, Food Environment Atlas, Dartmouth Atlas of Healthcare, and Healthcare Information and Management Systems Society. A total of 220 ACOs in the 2012/2013 cohort are excluded because the Centers for Medicare and Medicaid Services (CMS) did not publish information on ACO service areas in 2013, leading to missing SDOH. Two more observations were removed due to missing list of participants. Thus, the model uses a sample of 3628 (94.2%) ACO-year observations (unit of analysis) that consists of 858 distinct ACOs from 2014–2021.
a P < .001.
b P < .01.
cCovariates are not in percentage format for ease of presenting regression coefficients.
d P < .05.