Exposició GRASS

De 12:00-14:00.

Títol: Analysis of doubly interval-censored data: Review of the current methods and a new proposal

Resum:

Objectives: Doubly interval-censored data (DI data) is defined as a time (T) for which the start (U) and end (V = U+T) are interval censored. This data can be found in many clinical applications. We review the existing methodology and propose a stochastic EM algorithm to overcome some of the limitations of existing methods.

• Existing methods: A number of methods have been proposed to analyze DI data.
• Reduced likelihood methods: We review the methods that simplify the data to apply standard right-censored or interval censored techniques. We give a condition for these methods to work well.

Exiting DI data methods:1) Some methods su_er from the need to pre-specify a set of mass points a priori: we show by simulations the impact of this subjective choice. 2) Some methods only accept right-censored V and do not allow overlapping interval (when the upper bound of U is above the lower bound of V), which is frequently encountered in real life datasets.

Stochastic EM algorithm: We propose to use the EM algorithm to estimate the distribution of T and estimate the impact of covariates on the distribution. The idea is to treat the unknown event times as missing and apply the EM algorithm to estimate the parameters of interest (KM distribution or Cox PH model). Since no closed form for the integrated likelihood is available, we use the stochastic version of the EM to obtain the parameter estimates. We illustrate the use of this algorithm on simulations and on a motivating dataset.