Table 5.

Evaluation results from the best model with ablation studies for both outcomes

GBM models		Primary outcome		Secondary outcome
		(highest care level within 24 hours)		(care level at the 24th hour)
		AUROC (95% CI)	AUPRC (95% CI)	AUROC (95% CI)	AUPRC (95% CI)
Regime I	Full-featurea	0.88 (0.87–0.89)	0.65 (0.63–0.68)	0.86 (0.85–0.87)	0.50 (0.47–0.53)
	(-) Demographics & ESI	0.88 (0.87–0.89)	0.64 (0.62–0.67)	0.86 (0.84–0.87)	0.49 (0.47–0.53)
	(-) Vital signsb	0.85 (0.84–0.86)	0.60 (0.58–0.63)	0.83 (0.82–0.84)	0.45 (0.43–0.49)
	(-) Meds	0.88 (0.87–0.89)	0.63 (0.61–0.66)	0.85 (0.84–0.86)	0.48 (0.45–0.52)
	(-) Imaging	0.88 (0.87–0.89)	0.64 (0.62–0.67)	0.85 (0.84–0.86)	0.49 (0.46–0.53)
	(-) Diagnosis Codes	0.88 (0.87–0.89)	0.65 (0.63–0.67)	0.86 (0.85–0.87)	0.49 (0.47–0.53)
	(-) Procedures	0.88 (0.87–0.89)	0.65 (0.63–0.68)	0.86 (0.84–0.87)	0.50 (0.47–0.54)
	(-) Lab orders	0.87 (0.86–0.88)	0.63 (0.61–0.66)	0.87 (0.86–0.88)	0.48 (0.46–0.52)
	(-) Lab results	0.88 (0.87–0.89)	0.64 (0.63–0.67)	0.88 (0.87–0.89)	0.48 (0.46–0.52)
Regime II	Simple-featurea	0.82 (0.80–0.83)	0.52 (0.50–0.56)	0.81 (0.79–0.82)	0.41 (0.38–0.45)
	(-) Vitals summaryb	0.75 (0.73–0.76)	0.41 (0.39–0.44)	0.74 (0.73–0.76)	0.32 (0.29–0.35)
ESI-only logistic regression		0.67 (0.65 - 0.70)	0.37 (0.35–0.40)	0.67 (0.65–0.70)	0.28 (0.26–0.31)
		(intercept: 1.022; coefficient: −1.143)		(intercept: 0.774; coefficient: −1.209)

Notes: ^aindicates models without removing any feature types.

^bModels with vital signs as the feature type removed had the most reduction in AUROC and AUPRC. Vital signs are the most important predictors.