Table. Patient Characteristics, Associated Readmission Risk, and Comparison Between Hospitals With High vs Low Readmission Rates^a.

Characteristic	Readmission Risk Associated With Characteristic (95% CI)b	Standard Deviation in Patient Characteristic Across Hospitalsc	Lowest Quintile Readmission Rate (Mean, 8.9%)d	Highest Quintile Readmission Rate (Mean, 14.8%)d	P Valuee
Patient Characteristicsf
Male, %	0.73 (0.61 to 0.84)	3.4	41.9	40.7	<.001
Age, y	0.04 (0.03 to 0.05)	1.4	80.1	79.7	<.001
CCW chronic conditions, No.	0.66 (0.64 to 0.67)	0.76	9.0	9.7	<.001
HCC scoreg	2.08 (2.04 to 2.11)	0.24	2.0	2.3	<.001
Disabled, %	1.94 (1.77 to 2.11)	5.0	13.7	16.8	<.001
End-stage renal disease, %	10.70 (10.37-11.04)	1.8	2.7	3.6	<.001
Long-term nursing home resident, %	1.58 (1.38 to 1.78)	5.0	8.6	11.0	<.001
Insurance characteristicsh
Enrolled in Medicaid, %	3.02 (2.86 to 3.18)	13.0	17.2	24.3	<.001
Enrolled in a Medicare Savings Program, %	1.00 (0.69 to 1.32)	3.9	3.4	4.5
Receives Part D low-income subsidy, %	1.21 (0.81 to 1.62)	1.3	2.1	2.4
No subsidies or prescription drug coverage, %	1.04 (0.87 to 1.22)	4.5	15.0	12.6
Area-Level Characteristics Linked to Beneficiariesi
Poverty rate among residents ≥65 y in ZCTA, %	4.24 (3.10 to 5.37)	4.4	8.9	11.0	<.001
Poverty rate among residents ≥65 y in US Census tract, %	0.70 (0.07 to 1.33)	4.9	10.5	11.5	.005
Household income among residents ≥65 y in ZCTA <$50 000, %	2.51 (1.96 to 3.06)	8.4	61.6	64.5	<.001
Household income among residents ≥65 y in US Census tract <$50 000, %	2.39 (2.00 to 2.79)	8.6	61.6	64.4	<.001
Education among residents ≥65 y in ZCTAi
Less than a high school education, %	1.87 (1.02 to 2.72)	8.2	18.3	23.3	<.001
College or more, %	−1.21 (−1.94 to −0.48)	7.9	23.6	19.9	<.001
Education among residents ≥65 y in US Census tractj
Less than a high school education, %	2.03 (1.40 to 2.66)	8.3	18.3	23.4	<.001
College or more, %	−1.32 (−1.87 to −0.76)	8.0	23.5	19.7	<.001
Residents ≥65 y living alone in ZCTA, %	1.88 (1.08 to 2.68)	3.6	27.3	28.1	.001
Residents ≥65 y living alone in US Census tract, %	0.05 (−0.42 to 0.53)	4.1	27.4	27.6	.41

Abbreviations: CWW, Chronic Conditions Data Warehouse; HCC, Hierarchical Condition Categories; NA, not applicable; ZCTA, zip code tabulation area.

^aAmong patients in a random 20% sample of fee-for-service Medicare beneficiaries with 1 or more index admissions in 2013 or 2014 who met study criteria (see eAppendix 1 in the Supplement).

^bAverage within-hospital association with readmission is reported as percentage-point increase in readmission rate per unit increase in characteristic. Within-hospital associations estimated by individually adding each characteristic (or group of characteristics for race/ethnicity and area-level education) to a linear regression predicting readmission as a function of the characteristic and hospital fixed effects. Coefficient estimates are reported in percentage points of the readmission rate for a 1-unit change in the outcome (or for a 100% change for characteristics expressed as proportions).

^cThe hospital-level standard deviation of patient- and area-level characteristics was estimated from a random effects model. Using the proportion of Medicare patients dually enrolled in Medicaid as an example, the proportion of dual enrollees in hospitals differed by ±13.0 percentage points in hospitals 1 standard deviation above and below the mean among all hospitals (18.8%).

^dMean value of given characteristic among hospitals in the lowest and highest quintiles of unadjusted readmission rates. To address sampling error that would cause some hospitals to have more high-risk patients than others in the 20% sample or in a given year, a random effects model was used to estimate unadjusted hospital-level readmission rates, which were then categorized into quintiles reflecting systematic differences in readmission rates that would be expected to persist, on average, over different samples or years.

^eP value for the difference in means or proportions between hospitals in the highest vs lowest 20% of unadjusted readmission rates.

^fMeans or proportions of beneficiary-level characteristics assessed from Medicare claims and enrollment data. See the Box footnotes for additional information about the characteristics.

^gHCC scores were constructed using Medicare enrollment and claims data from the prior year, with higher scores indicating higher predicted spending in the subsequent year.

^hBeneficiaries were categorized into mutually exclusive categories of insurance coverage. Receiving prescription drug coverage through the Medicare Part D program, an employer, or another source was omitted as the reference category. See the Box footnotes and eAppendix 2 in the Supplement for details of variable definitions.

ⁱCharacteristics of beneficiaries’ ZCTAs or US Census tracts are based on residential address data reported in Medicare enrollment files. Estimates of the average within-hospital associations reflect the expected change in the readmission rate associated with a 100 percentage-point change in the proportion of residents with the area-level characteristics shown.

^jThe omitted category is the proportion of residents 65 years and older with a high school education.