Table 4.

Predictability of four residual kidney function index models for all-cause mortality using net reclassification index, integrated discrimination improvement, and c statistic

Models	NRI (SEM; 95% CI)	P Value	IDI (SEM; 95% CI)	P Value	c Statistics (95% CI)	Mean of Differences in c Statistics (95% CI)a
Base modelb	Reference		Reference		0.75 (0.73 to 0.78)
Base model and GFR-24U	0.14 (−0.05 to 0.25)	0.09	0.002 (−0.001 to 0.01)	0.12	0.76 (0.73 to 0.78)	0.002 (−0.004 to 0.01)
Base model and eGFR-urea,creat	0.19 (−0.24 to 0.27)	0.47	0.001 (−0.001 to 0.01)	0.44	0.76 (0.73 to 0.78)	0.001 (−0.001 to 0.004)
Base modelc and eGFR B2M	0.23 (−0.16 to 0.34)	0.13	0.01 (−0.001 to 0.01)	0.11	0.76 (0.74 to 0.79)	0.01 (0.003 to 0.03)
Base model and UV	0.11 (0.01 to 0.21)	0.01	0.01 (0.001 to 0.02)	0.01	0.76 (0.73 to 0.78)	0.01 (0.001 to 0.01)

NRI, net reclassification index; 95% CI, 95% confidence interval; IDI, integrated discrimination improvement; GFR-24U, GFR by 24-hour urine collection; eGFR-urea,creat, estimated GFR calculated from the serum urea and creatinine concentrations; B2M, β2-microglobulin; UV, urine volume.

^aDifferences in c statistics were calculated using bootstrapping with 1000 replicates.

^bBase model included age, sex, dialysis modality, dialysis vintage, modified Charlson comorbidity index, smoking status, and serum albumin concentrations.

^cB2M data were available in 1640 patients. Comparison between base model and base model and eGFR B2M was performed in 1640 patients.