Volume 3, Issue 1, Page 51

Twin Research

March 2000, Volume 3, Issue 1, Pages 51 - 57

	Journal Home
<- Previous	Issue Contents	Next ->

Article

Genetic nature of individual frailty: comparison of two approaches
Alexander Z Begun¹, Ivan A Iachine² & Anatoli I Yashin³
¹Max Planck Institute for Demographic Research, Rostock, Germany     ²Department of Statistics and Demography, Odense University, Denmark     ³Duke University, Center for Demographic Studies, Durham, USA

Correspondence to: Alexander Z Begun, Max Planck Institute for Demographic Research, Doberaner Strasse 114, D-18057 Rostock, Germany. Tel: (+49) 381 20 81 107; Fax: (+49) 381 20 81 407; E-mail: begun@demogr.mpg.de

Keywords

frailty models;   twins;   random hazards

Abstract

The traditional frailty models used in genetic analysis of bivariate survival data assume that individual frailty (and longevity) is influenced by thousands of genes, and that the contribution of each separate gene is small. This assumption, however, does not have a solid biological basis. It may just happen that one or a small number of genes makes a major contribution to determining the human life span. To answer the questions about the nature of the genetic influence on life span using survival data, models are needed that specify the influence of major genes on individual frailty and longevity. The goal of this paper is to test the nature of genetic influences on individual frailty and longevity using survival data on Danish twins. We use a new bivariate survival model with one major gene influencing life span to analyse survival data on MZ (monozygotic) and DZ (dizygotic) twins. The analysis shows that two radically different classes of model provide an equally good fit to the data. However, the asymptotic behaviour of some conditional statistics is different in models from different classes. Because of the limited sample size of bivariate survival data we cannot draw reliable conclusions about the nature of genetic effects on life span. Additional information about tails of bivariate distribution or risk factors may help to solve this problem. Twin Research (2000) 3, 51-57.

Received 30 April 1999; Accepted 20 May 1999