Sample sizes for proportional hazards survival studies with arbitrary patient entry and loss to follow-up distributions.

In proportional hazards survival studies, power depends on the observed number of deaths, d*. For a given choice of survival, loss, and patient entry distributions, sample sizes can be determined by equating d* to the expected number of deaths. Approximating the survival and loss distributions with piecewise exponential distributions, and patient entry with a piecewise linear distribution, significantly reduces the computational overhead, and the expected number of deaths can be evaluated routinely. The merits of this approach are illustrated by a clinical trial of chemotherapy for large bowel cancer.