Treatment comparisons with screenable endpoints

This paper proposes a method for treatment comparison when the disease endpoint can be discovered by either screening or clinical surfacing. Assuming a zero false negative screening rate, we develop a score test for comparing treatments while adjusting for the mode of discovery based on a three-state progressive disease model with a discrete proportional hazards formulation of the treatment effect. By simulation we evaluate and compare the performance of the score test to the log-rank test. The results suggest that the score test is more efficient when screening rates are balanced between groups, and it greatly reduces the bias in the situation of asymmetric screening intensity.