Estimating transition coefficients for reconstructing coherent series of mortality by cause of Death

Abstract Regular revisions of the classification of diseases and the consequent disruptions of mortality series are well-known issues in long-term cause-of-death analysis. Given basic assumptions and medical knowledge about possible exchanges across causes of death in the revision years, redistribution of counts of causes of death into a new classification can be viewed as a constrained optimization problem. Penalized likelihood within a quadratic programming framework allows estimation of exchanges that vary smoothly over age groups. The approach is illustrated using both German data on malignant neoplasms and French data on heart diseases