Statistics on mortality related to each disease are usually based on the so-called underlying cause of death, which is selected from the diseases declared on the standardized death certificate using international rules. However, the assumption that each death is caused by exactly one disease is debatable, particularly with an aging population in an era where infectious diseases are replaced by chronic and degenerative diseases. The need to consider multiple causes of death has been acknowledged in epidemiologic research, with a growing body of literature producing statistics based on any mention of a disease on the death certificate. Yet there has not been a formal framework proposed for the statistical modeling of death arising from multiple causes. We propose a model for multiple cause of death data grounded on an empirical approach that assigns weights to each cause on the death certificate. We describe how this model for multiple-cause mortality, which extends the usual competing risks model used to conceptualize single-cause mortality, can serve to study the burden and etiology of mortality related to each disease, particularly using Cox regression methodology. We discuss how the multiple-cause, single-cause, and “any-mention” approaches compare in this regard. A simulation study and an application to a study of socioeconomic inequalities in mortality show the value of the proposed methods for exploiting this precious source of data to gain new insights, especially for certain diseases. See video abstract at, http://links.lww.com/EDE/B84.
Supplemental Digital Content is available in the text.
From the aClinical Epidemiology and Biostatistics Unit, Murdoch Childrens Research Institute, Melbourne, Australia; bDepartment of Epidemiology and Preventive Medicine, Monash University, Melbourne, Australia; cInserm CépiDc, Epidemiology Centre on Medical Causes of Death, Le Kremlin-Bicêtre, France; and dAP-HP, Assistance publique-Hôpitaux de Paris, Paris, France.
Submitted 22 July 2015; accepted 21 June 2016.
Supported by a Centre of Research Excellence grant from the Australian National Health & Medical Research Council, ID#1035261, to the Victorian Centre for Biostatistics (ViCBiostat), and a research grant from Institut National du Cancer (INCa, 2012-1-PL SHS-05-INSERM 11-1) and Cancéropôle Ile-de-France.
The authors report no conflicts of interest.
Supplemental digital content is available through direct URL citations in the HTML and PDF versions of this article (www.epidem.com).
Correspondence: Margarita Moreno-Betancur, Clinical Epidemiology and Biostatistics Unit, Murdoch Childrens Research Institute, 50 Flemington Road, Parkville, Victoria 3052, Australia. E-mail: firstname.lastname@example.org.