Electronic medical record-based multicondition models to predict the risk of 30 day readmission or death among adult medicine patients: validation and comparison to existing models

Table 4 Comparison of performance of discharge 30-day readmission risk models^a

	C statistic (95 % CI)	Generalized R²	Predicted event rate by decile of predicted risk, %		NRI index^c	IDI index^d
	C statistic (95 % CI)	Generalized R²	Lowest	Highest	NRI index^c	IDI index^d
E-Risk Model^e	0.68 (0.67-0.69)	0.059	5.9	30.7		0.049
CMS-HWR Model	0.61 (0.59-0.62)	0.011	9.6	20.3		0.012
Difference (95 % CI)	0.075 (0.061-0.089)	0.047			0.198	0.037 (0.033-0.041)
P value	< .001	N/A^b			< .001	< .05

Abbreviations: CI, confidence interval; NRI, net reclassification improvement; IDI, integrated discrimination improvement; E-Risk Model, automated real-time model to identify adult medicine patients at risk for 30-day readmission using electronic medical record data; CMS-HWR, Centers for Medicare and Medicaid Services – Hospital Wide Readmission with Medicine, Cardiovascular, Cardiorespiratory, and Neurology submodels
^aBased on model comparison cohort of 16 937 patients
^bNo test of significance applicable to difference in generalized R² between these non-nested prediction models
^cNet reclassification improvement is the sum of the proportion of patients moving up less the proportion moving down, among patients who are readmitted, and the proportion of patients moving down less the proportion moving up, among patients who are not readmitted
^dDiscrimination slope is difference of estimated mean probabilities for events and nonevents
^eDischarge version: Updated for length of stay, additional diagnosed comorbidities and complications

ISSN: 1472-6947