Abstract
Results concerning the dependence of life durations on alcohol use of individuals who die of coronary heart disease are brought into question when no account is taken of the mortality risks associated with other competing diseases. Procedures which restrict the analysis of coronary heart disease to respondents who die of this disease will produce biased estimates of the parameters of interest. Other dimensions of coronary heart disease are neglected by focusing on durations of life and well-constructed sample surveys can provide important new information on how alcohol consumption affects the risks of getting coronary heart disease.
Mathematics Subject Classification:
Notes
This is a generalization of the Cox proportional hazard model. See Lancaster (1990), van den den Berg (2001) and Abbring and van den Berg (2003).
There is a large literature on competing risk models. Some useful references are Kalbfleisch and Prentice (2005), Van den Berg (2001), and Lancaster (1990).
This assumption simplifies the analysis but in no way affects the results which are the same when there is noninformative censoring like that which arises when, for example, the data collection exercise is terminated for administrative reasons.
Additional conditions for nonparametric identification are required. These are described in Abbring and van den Berg (2003, p. 704).
Kenneth Mukamal, in a personal communication, reported that applying a competing risk model did not change any of their 2010 results.
The Canadian Community Health Surveys which were used for some of the author’s research, McMillan et al. (2007) and McIntosh (2008). This is a good well run survey but it could be improved by a more thoughtful questionnaire design.