Skip to main content
. Author manuscript; available in PMC: 2021 Mar 31.
Published in final edited form as: Stat Med. 2020 Jul 27;39(26):3685–3699. doi: 10.1002/sim.8687

TABLE 2.

Simulation results comparing PCGLM, GEE.pseudo and RF.pseudo in Simulation Study1 (n=200 in training data and n=200 in validation data; each with 30% censoring). AUC, PE and MSE are the averaged AUC, PE and MSE over 500 simulated data sets, respectively, and ESD is corresponding empirical standard error.

n0 s = 6 s = 12 s = 18 s = 24
MSE (ESD) MSE (ESD) MSE (ESD) MSE (ESD)
PCGLM 0.003 (0.002) 0.004 (0.002) 0.005 (0.003) 0.010 (0.009)
GEE.pseudo 0.003 (0.002) 0.004 (0.002) 0.005 (0.003) 0.010 (0.009)
RF.pseudo 0.010 (0.003) 0.014 (0.004) 0.019 (0.006) 0.028 (0.012)
AUC (ESD) AUC (ESD) AUC (ESD) AUC (ESD)
PCGLM 0.775 (0.04) 0.830 (0.04) 0.827 (0.06) 0.809 (0.09)
GEE.pseudo 0.775 (0.04) 0.830 (0.04) 0.827 (0.06) 0.810 (0.09)
RF.pseudo 0.752 (0.05) 0.805 (0.06) 0.798 (0.06) 0.761 (0.10)
PE (ESD) PE (ESD) PE (ESD) PE (ESD)
PCGLM 0.157 (0.02) 0.172 (0.02) 0.175 (0.03) 0.183 (0.05)
GEE.pseudo 0.157 (0.02) 0.172 (0.02) 0.175 (0.03) 0.184 (0.05)
RF.pseudo 0.165 (0.02) 0.192 (0.02) 0.200 (0.03) 0.215 (0.04)