The semiparametric efficiency bound of the mixed proportional hazard m
odel is derived. The density of the model factors in such a way that t
here exists a complete sufficient statistic for the individual heterog
eneity. The efficient score is shown to be the difference between the
score in the parametric direction and its conditional expectation give
n the sufficient statistic. Applying this result to the single-spell W
eibull mixed proportional hazard model, it is shown that its informati
on matrix is singular and there cannot exist any square-root n-consist
ent estimator sequence. The information of the multi-spell Weibull mix
ed proportional hazard model is shown to be nonsingular in general.