American Statistical Association New York City Metropolitan Area Chapter Memorial Sloan-Kettering Cancer Center Biostatistics Seminar

In numerous medical studies, participants may experiences a succession of landmark events. In cystic fibrosis, repeated pulmonary exacerbations episodes are an example of same-type events. In a cancer clinical trial, a sequence consisting of therapy initiation time, end of the toxicity period (TOX), end of the disease-free period (DF) and death represents a series of different-type events. Regardless of the event nature, the inter-event times are usually referred to as gap times. Conditional modeling of the gap times is often times challenging and practical implementation in mainstream statistical packages is not easily available. We propose a flexible, computationally efficient, novel modeling strategy based on jackknife pseudo-observations (POs). This construct requires an (approximately) unbiased nonparametric estimator for the joint distribution of the gap times. In essence, the problem is translated into the more accommodating realm of generalized linear models. Simulation studies show that the method proposed produces virtually unbiased covariate effect estimates, even for moderate sample sizes. An example from a medical study further illustrates the practical advantages of this new approach. Other time-to-event settings where this modeling strategy is extremely advantageous will also be discussed.