A LINEAR NONPARAMETRIC REGRESSION-MODEL FOR THE EXCESS INTENSITY

Non-proportionality and non-positive excess intensities are two major problems in regression modelling for long-term survival of cancer. In this paper a linear non-parametric regression model for the excess int ensity is presented which overcomes these problems. The model allows t he excess intensity to be varying with time as well as to be non-posit ive. A test for the effect of covariates in specific time intervals, o r for the complete observation period, is given together with a Kolmog orov-Smirnov type test. The method is illustrated by an analysis of lo ng-term survival of all Norwegian male colon cancer patients registere d in the period 1965 to 1974, and who survived the initial 5 years aft er diagnosis.