Competing risks in absence of independence:

Abstract 

The increase in lung cancer (LC) mortality can produce a decrease in mortality from other causes, including ischemic heart disease (IHD). This problem (called the competing risks problem) has been addressed usually assuming independence between the competing causes of death. Our purpose is to show that assuming dependence of causes of death allows obtaining a better estimation of cumulative mortality. We use a clinical epidemiological example on the impact of AIDS in liver function failure in a cohort of drug users. The competing effect under dependence is 47% higher than under independence. This result is compared with a population-based example on LC and IHD mortalities in Spanish people in 1992. LC and IHD share tobacco smoking as a common risk factor, so independence cannot be assumed. Under the independence assumption, both life expectancy and number of deaths from IHD are underestimated. The difference is small compared to the model computed under dependence and it occurs mainly in the elderly (0.3% more deaths in people aged 70 and over).

Keywords:  AIDS, Competing risks, Ischemic heart disease, Liver function failure, Lung cancer, Mortality