Table 5 Performance of the prediction model by survival hazard ratio decision tree using different model setting.

Setting	Imputation method	Use One	Weight method	Validation method	Validation ratio	N folds	Train set size	Test set size	Parameters	Training Performance	Test Performance
1	MICE/CART	F	nothing	Cross-validation		5	2796	932	cp = 0.01/maxdepth = 6	0.73	0.71
1	MICE/CART	T	nothing	Cross-validation		5	2796	932	cp = 0.01/maxdepth = 6	0.71	0.68
1	MICE/CART	T	nothing	One validation	0.285		2796	932	cp = −1/maxdepth = 2	0.72	0.67
1	MICE/CART	F	nothing	One validation	0.285		2796	932	cp = 0.004/maxdepth = 14	0.92	0.59
2	nothing	F	nothing	One validation	0.285		930	310	cp = 0.028/maxdepth = 4	0.78	0.80
2	MICE/CART	F	nothing	Cross-validation		5	2796	932	cp = 0.008/maxdepth = 4	0.74	0.69
2	MICE/CART	F	nothing	One validation	0.285		2796	932	cp = 0.01/maxdepth = 6	0.77	0.68
2	MICE/CART	T	nothing	Cross-validation		5	2796	932	cp = 0.008/maxdepth = 6	0.71	0.68
2	MICE/CART	T	nothing	One validation	0.285		2796	932	cp = −1/maxdepth = 2	0.72	0.67
2	nothing	F	nothing	Cross-validation		5	930	310	cp = −1/maxdepth = 12	0.94	0.62
2	nothing	T	nothing	Cross-validation		5	930	310	cp = 0.002/maxdepth = 12	0.94	0.61
2	nothing	T	nothing	One validation	0.285		930	310	cp = 0.02/maxdepth = 8	0.91	0.60
3	MICE/CART	F	nothing	Cross-validation		5	2796	932	cp = 0.012/maxdepth = 4	0.72	0.68
3	MICE/CART	T	nothing	Cross-validation		5	2796	932	cp = 0.014/maxdepth = 4	0.70	0.65
3	MICE/CART	F	nothing	One validation	0.285		2796	932	cp = 0.01/maxdepth = 6	0.74	0.61
3	MICE/CART	T	nothing	One validation	0.285		2796	932	cp = 0.014/maxdepth = 4	0.72	0.58

*Test performance were presented as concordance index for time to graft failure data. Model setting 1: Use overall attributes with multiple imputation. Model setting 2: Use attributes except DSA, with/without imputation. Model setting 3: Use attributes except 3 month follow up data, with imputation. Test ratio fix 0.3. Use One False (F) is used if the follow-up period is shorter than the period to be predicted in the classification; these cases were excluded from the training process. Use One True (T) is used as a positive example if the patient has experienced a graft failure even though the follow-up period is short. Weight method by Zupan et al. is used when the follow-up period is short, with both positive and negative examples used, but with different weights⁶.

Search