Table 5.

Selection percentages of treatment-covariate interactions in the prediction model for the donation outcome across 100 repetitions of 5-fold cross-validation when l₁-PLS is used to estimate the optimal ITR.

		Variable Selection
Baseline Variables	Variable Type	HGL	GL
Age	Continuous	63	21
Body mass index	Continuous	33	38
SF-36v2 physical component score	Continuous	47	11
SF-36v2 mental component score	Continuous	100	20
Blood donations in the two years before trial enrollment	Continuous	100	100
Hemoglobin level	Continuous	100	100
White blood cell count	Continuous	100	46
Red blood cell count	Continuous	0	11
Mean corpuscular hemoglobin	Continuous	100	96
Mean corpuscular volume	Continuous	62	11
Platelet count	Continuous	100	12
Ethnicity	Categorical	47	97
Blood group	Categorical	99	99
Iron prescription	Categorical	1	74
Smoke ever	Categorical	13	39
Smoke currently	Categorical	98	99
Alcohol ever	Categorical	85	32
Alcohol currently	Categorical	100	100
New or returning donor status	Categorical	100	32

Note: Hierarchical group LASSO (HGL) enforces strong hierarchy between main effects and interactions, and group LASSO (GL) does not impose strong hierarchy between main effects and interactions.