Table 3.

Performance of HRS-CMS and ACS-HCUP Models Compared to Current CMS Models, by Cohort

	Pneumonia	Heart failure	Acute myocardial infarction
HRS-CMS model
C-statistic: our full model with clinical risk adjustment + social determinants of health	0.76	0.68	0.79
C-statistic: CMS model with clinical risk adjustment	0.64	0.61	0.64
Likelihood ratio test comparing our full model to a model with clinical risk adjustment only	χ 2(14) = 43.15 p < 0.001	χ 2(14) = 35.41 p = 0.001	χ 2(14) = 17.41 p = 0.24
ACS-HCUP model
C-statistic: our full model with clinical risk adjustment + social determinants of health	0.62	0.55	0.63
C-statistic: CMS model with clinical risk adjustment	0.64	0.61	0.64
Likelihood ratio test comparing our full model to a model with clinical risk adjustment only	χ 2(8) = 28.06	χ 2(8) = 33.25	χ 2(8) = 20.11
	p < 0.001	p < 0.001	p = 0.007

Note: ACS-HCUP refers to merged Healthcare Cost Utilization Project State Inpatient Databases from Florida and Washington (2009–2012) and US Census American Community Survey 5-year ZIP Code Tabulation Area (ZCTA) data, 2008–2012. HRS-CMS refers to Health and Retirement Study data linked with administrative claims data from the Centers for Medicare and Medicaid Services. The likelihood ratio test compares the model with clinical risk adjustment (includes clinical comorbidities, age, and gender) to the “full” model in this analysis that includes the disability and social determinants of health variables in addition to the clinical risk adjustment. Both models are fit using the same set of observations. A small p value indicates a better fit for the model with the added variables. CMS model c-statistics reported in the 2013 Measures Updates and Specifications report.⁶