CMStatistics 2018: Start Registration
View Submission - CMStatistics
B0813
Title: Variance estimation for generalised pseudo-values Authors:  Martina Mittlboeck - Medical University of Vienna (Austria) [presenting]
Harald Heinzl - Medical University of Vienna (Austria)
Ulrike Poetschger - Childrens Cancer Research Institute (Austria)
Abstract: Recently, a novel methodology based on generalised pseudo-values was suggested to compare survival of two cohorts, where cohort membership is a latent baseline variable. Patients in one cohort may undergo an intervention over time, dependent on an exogenous time-consuming search process. A typical example is stem cell transplantation, where identification of a suitable donor from existing databases takes time. Cohort membership becomes known if the search process is ended either successfully or unsuccessfully, yet it remains unknown if donor search is ceased due to patients' death or censoring. The calculation of the generalised pseudo-values for the cohort with time-dependent intervention consists of two-parts: Firstly, the survival probability $S0(w)$ before the intervention at $w$, and secondly the survival probability $S1(t*|w)$ from intervention at $w$ until time of interest $t*$. The survival probability $S0(w)$ before the intervention can easily be estimated by Kaplan-Meier. However, variability estimation for the calculation of proper confidence intervals and for testing group differences is not straightforward and often computationally intensive. Different approaches are investigated and compared with respect to coverage of 95 \% confidence intervals.