Fig. 1.

Sample construction for imputation and survival classification. a Sample construction for each patient with missing data to be imputed. b Candidate sample construction procedure. In this example, subject i has n=4 visits in the first three months of screening (one in the first month, two in the second and one in the third) while candidate j has 3 visits in this interval (one visit per month). Since the visit at t_j,2 matches both visits at t_i,2 and t_i,3, its dynamic feature values are repeated twice in the resulting feature vector (sample). c Survival classification sample construction for each patient