Assess the presence of disruptions in cause-specific mortality series

Carlo G. Camarda, Institut National d'Études Démographiques (INED)
Marketa Pechholdova, University of Economics, Prague (VSE)

Time series of mortality by cause of deaths may present, in the year of ICD revision, disruptions which are undue to variability in the mortality trend. Assessing the presence of these eventual disruptions is of great help for drawing conclusions on redistribution of death counts among other causes of death and, consequently, for the estimation of continuous mortality series. Our approach aims to detect statistically significant discontinuity in an ICD-revision year by the mean of (back-)forecast the estimated trend in the revision year. The estimation of the trends is done via smoothing techniques on the standardized deaths within a Poisson framework for accounting the variability of the phenomenon. A simulation study is given to demonstrate the performance of the method. An actual application on West German cause-specific mortality data illustrate the outcome of the approach.