Table 4.

Performance of multivariable predictive models for composite outcome^a

Model components	c	Nagelkerke pseudo-R2	Brier score
TSLb	0.716	0.066	0.112
TSL + age + sex	0.708	0.066	0.112
YCLSc	0.590	0.022	0.116
YCLS + age + sex	0.597	0.027	0.116
PROMIS scalesd	0.555	0.010	0.117
PROMIS + age + sex	0.563	0.015	0.116
TSL + YCLS	0.695	0.074	0.111
TSL + PROMIS	0.703	0.068	0.111
TSL + YCLS + PROMIS	0.665	0.074	0.110
YCLS + PROMIS	0.536	0.025	0.186
Random foreste	0.715	0.060	0.112

^aModels are calibrated against a composite outcome (non-elective rehospitalization – defined as a hospitalization that began in the emergency department – and/or death within 30 days after hospital discharge). c = c statistic, or area under receiver operator characteristic curve. All results are after fivefold cross validation

^bTSL = Transition Support Level score. This score is assigned at 6 AM on the day of discharge to all adult hospitalized patients in Kaiser Permanente Northern California. The score, which is expressed as a percent, is calibrated against a composite outcome (non-elective hospitalization and/or death within 30 days of discharge). It is based on a patient’s LAPS2, COPS2, length of stay, recent hospital and emergency department utilization preceding the current hospitalization, and discharge care directive (full code or not); see Escobar et al. [17] for details. Patients with a TSL score of ≥ 25% receive additional assessments and follow-up calls and appointments

^cYCLS = Your Current Life Situation questionnaire. See Appendix 1 for details

^dPatient Reported Outcomes Measurement Information System Cognitive Function bank v. 2.0 and Physical Function bank v. 2.0. See Appendix 1 for details

^eThe random forest model included the following variables: age, sex, individual components of the TSL score, the two PROMIS scales, and the 6 YCLS components listed in Table 2