TWISTed survival: identifying surrogate endpoints for mortality using QTWIST and conditional disease free survival

Traditionally, the standard endpoint in most cancer clinical trials has been overall survival. For many forms of cancer, including colon cancer, the time from diagnosis to the time when this endpoint is reached can take many years. Hence, researchers and patients must wait a considerable amount of time to see if a treatment is effective. We propose an alternative surrogate endpoint which would occur in less time but still be as effective at determining treatment differences. The discovery of such an endpoint would be of Public Health importance to both patients and researchers as it would allow treatments to be tested in a shorter time and subsequently allow patients to have quicker access to a beneficial treatment. \ud Our approach is to use the methodology of QTWIST and conditional survival estimates, conditioning on disease free survival, to produce a potential surrogate endpoint for overall survival. To do this, we examine whether a surrogate endpoint could be theoretically produced for colon cancer. We analyzed NSABP trials C-03 through C-07 to determine the effect of conditional survival and the choice of conditioning sets in colon cancer. In doing this analysis, we examine the impact of determining the probability of surviving an additional y years given that a patient has already been alive and disease free for x years. QTWIST, quality-adjusted time without symptoms, methodology is then reviewed focusing on the underlying methodology of using weights, called utility coefficients, and how they could be applied to a partitioning of disease free survival states. Methodology combining conditional survival and the statistical methodology of QTWIST were then performed on six different sets of potential weighting coefficients. Finally, the success of the methodology was evaluated by comparing the Kaplan-Meier treatment difference p-values to the treatment difference p-values for each of the six utility coefficient approaches tested in our methodology. It is our hope that this methodology will produce a viable predictor for overall survival and one that is more predictive than using standard disease free survival estimates.\u