Table 1.

Characteristics of each prediction study

Author	Cohort	ICU or hospital LOS	Design	Prediction	RMSE	MAE	R2 (%)	Log-scaled ICU LOS	Truncation at 30 days	Excluded	Quadratic/cubic terms	Interaction terms	Normalization	Partition	Cross-validation
Moran et al.(4)	111.663	ICU LOS	131 mixed-type ICUs	Admission	4.50	2.30	22.0	Yes	No	ICU LOS > 60 days	Yes	Yes	No	N/A	N/A
Verburg et al.(10)	32.667	ICU LOS	83 mixed-type ICUs	Admission	7.28	3.43	15.4	Yes	Yes	Hospital LOS > 365 days	No	No	No	Bootstrap	N/A
Houthooft et al.(11)	14.480	ICU LOS	14 medical ICUs	Day 5	N/A	1.79	21.9	Yes	No	ICU LOS > 40 days	No	No	Yes (method not informed)	60/40	N/A
Li et al.(12)	1214	ICU LOS	One ICU	Admission	0.88	0.87	35.0	Yes	No	No	No	No	Z-score	70/30	10-fold
Muhlestein et al.(13)	41.222	Hospital LOS	1,000 hospitals	Admission	0.56*	N/A	N/A	No	No	No	No	No	Z-score	80/20	5-fold
Caetano et al.(14)	26.431	Hospital LOS	One hospital	Admission	0.47*	0.22*	81.3	Yes	No	No	No	No	Z-score	N/A	5-fold

ICU - intensive care unit; LOS - length of stay; RMSE - root mean square error; MAE - mean absolute error; N/A - not available;

^*Root mean square error/mean absolute error calculated using log transformation.