Table 7. Predicted risk of death and readmission at 24 hour intervals in up to the last four days in the ICU.

Percent Risk in Patient Subgroups	Days Prior to ICU Dischargea				Rate of Change
	3	2	1	0	Average per Dayb	95% CI
Risk of Death Model
Survived 48 hours after ICU discharge, Median % (IQR)	0.80(0.26–2.96)	0.55(0.19–1.83)	0.30(0.11–0.89)	0.19(0.07–0.48)	−1.18	−1.21 – −1.16c
Did not survive 48 hours after ICU discharge, Median % (IQR)	7.01(2.02–23.77)	6.44(1.83–24.16)	6.29(1.73–26.63)	8.82(1.73–43.53)	2.47	1.98 – 2.97d
Did not survive the ICU stay, Median % (IQR)	11.11(2.87–36.50)	12.51(3.07–41.41)	16.61(3.88–53.16)	69.51(25.20–94.74)	13.08	12.81 – 13.34e
Risk of Readmission Model
No readmission within 48 hours after ICU discharge, Median % (IQR)	2.71(1.63–4.40)	2.50(1.51–3.98)	2.19(1.32–3.44)	2.04(1.24–3.11)	−0.31	−0.32 – −0.31c
Readmitted within 48 hours after ICU discharge, Median % (IQR)	4.21(2.63–6.93)	3.82(2.43–6.50)	3.42(2.21–5.76)	3.30(2.15–5.52)	−0.35	−0.39 – −0.31d
Did not survive the ICU stay, Median % (IQR)	4.08(2.28–7.07)	4.13(2.19–6.99)	4.14(2.17–7.14)	4.20(2.23–7.37)	0.07	0.03 – −0.10e

^aDay 0 =  predictions generated at the time of ICU discharge; Day 1 =  predictions generated 24 hours prior to discharge; Day 2 =  predictions generated 48 hours prior to discharge; Day 3 =  predictions generated 72 hours prior to discharge.

^bThe average change in predicted risk (%) across the last four ICU days using linear regression with a robust variance estimator clustered by patient across ICU days.

^c600,252 observations.

^d9,830 observations.

^e40,801 observations. All p-values <0.01.