TABLE 4.

Performance of the recursive prediction models based on selected features (clinical features and superalleles) implemented using various regression technique.

Method	Attribute	HR	P-value
LASSO	Clinical + 14 HLA-superalleles (A*31_A*24_DPB1*10_B*08_DRB1* 03_DRB1*07_B*18_B*55_A*01_C* 05_DRB1*16_DRB1*12_B*49 _DPB1*11)	4.52	8.01E-15
RIDGE	Clinical + 19 HLA-superalleles (DPB1*10_B*50_C*07_B*49_B*55 _B*08_C*01_C*14_DPB1*06_C*05 _DRB1*03_A*30_DRB1*07_A*31 _B*14_DRB1*16_B*13_DPB1* 01_A*01)	3.85	3.35E-12
RF	Clinical + 3 HLA-superalleles (DPB1*11_C*05_B*08)	3.53	2.84E-11
DT	Clinical + 2 HLA-Superalleles (A*01_DPB1*01)	2.59	6.92E-08

#HR: Hazard Ratio; RF: Random Forest; DT: Decision Tree; Attribute: Clinical features and selected HLA-superalleles.