Per Kragh Andersen

Measuring Expected Years of Life Lost

We will study two classes of models for estimating the expected number of life years lost for a given group of subjects. One is phrased in the framework of a (Markov or non-Markov) illness-death model in combination with a population life-table. It is compared to a number of previously suggested measures of life years lost among patients with a given disease. The other is based on the competing risks model and we show that the cause j cumulative incidence function integrated from 0 to t has a natural interpretation as the expected number of life years lost due to cause j before time t. The large sample properties of a non-parametric estimator are outlined, and it is discussed how the expected number of years lost may be related to explanatory variables in a regression model based on pseudo-observations. The methods are illustrated using data on Danish patients with cancer or psychiatric disorder.