• Cause-specific hazards;
  • Competing risks;
  • Frailty;
  • Informative censoring;
  • Pancreatic cancer;
  • Time-dependent covariates

Summary.  The paper describes a complex multistate modelling approach for the analysis of association between the onset of diabetes and the diagnosis of pancreatic cancer among families with hereditary pancreatitis. The model allows for competing risks, correlated survival times and time-dependent covariates, so taking several interrelated factors into account: the consequences of pancreatic resection, genetic similarities between family members and the possibility that the onset of diabetes is the first symptom of an undiagnosed pancreatic cancer. The model is applied to data on diabetes and cancer outcomes for 593 pancreatitis patients, taken from the world's largest registry of patients with heredity pancreatitis.