Accurate training of the Cox proportional hazards model on vertically-partitioned data while preserving privacy

Table 6 Lung dataset

Covariates	coef_lib	coef_newton	coef_mpc	se_lib	se_newton	se_mpc
Inst	\(-\)0.011852	\(-\)0.011861	\(-\)0.011856	0.010921	0.010921	0.010930
Age	0.000026	0.000027	\(-\)0.000046	0.009779	0.009779	0.009848
Sex	\(-\)0.251135	\(-\)0.251185	\(-\)0.251434	0.163212	0.163212	0.163282
ph.ecog	0.615030	0.614995	0.615158	0.204500	0.204500	0.204551
ph.karno	0.023395	0.023392	0.023376	0.010189	0.010189	0.010203
pat.karno	\(-\)0.009492	\(-\)0.009487	\(-\)0.009491	0.007027	0.007027	0.007047
meal.cal	\(-\)0.000080	\(-\)0.000080	\(-\)0.000076	0.000227	0.000227	0.000227
wt.loss	\(-\)0.011039	\(-\)0.011043	\(-\)0.011063	0.006606	0.006606	0.006607

Coefficients (coef) and standard error (se) are listed for each implementation. Convergence was reached in two iterations for ‘lib’, in three iterations for ‘newton’ and ‘mpc’. The secure implementation ‘mpc’ took 3073 seconds to complete

ISSN: 1472-6947