Table 4.

Prentice's criteria 3 and 4 for the 3-year landmark analysis (n = 1364)^*

Variable	Coefficient	P†
S(t) = DM
DM (H0: β1 = 0)	2.71	<.001
DM × TRT (H0: β2 = 0)	0.10	.54
Non-DM × TRT (H0: β3 = 0)	−0.31	.39
S(t) = GCTF
GCF (H0: β1 = 0)	3.25	.004
GCF × TRT (H0: β2 = 0)	−0.16	.29
Non-GCTF × TRT (H0: β3 = 0)	−0.02	.98

^*

The following proportional model was used to validate the criteria. Prentice's third criterion, that the surrogate endpoint has a statistically significant impact on the true endpoint, can be tested by H₀:β₁ = 0. Prentice's fourth criterion can be tested by H₀:β₂ = 0 and H₀:β₃ = 0.

where h(t) = hazard rate of prostate cancer-specific death at time t; S(t) = surrogate endpoint (failure = 1 and nonfailure = 0); 1 − S(t) = not having surrogate endpoint failure; TRT = treatment arm analyzed (control = 0 and experimental = 1); DM = distant metastasis; GCTF = general clinical treatment failure; β₁ = coefficient for surrogate endpoint effect; β₂ = coefficient for treatment effect; β₃ = coefficient for the interaction between the not having a failure for surrogate endpoint and treatment.

^†Chi-square test statistic. All statistical tests were two-sided.