Simulation study of the effects of excluding early deaths on risk factor-mortality analyses in the presence of confounding due to occult disease: The example of body mass index

David B. Allison, Moonseong Heo, Dana W. Flanders, Myles S. Faith, Kenneth M. Carpenter, David F. Williamson

Research output: Contribution to journalArticle

40 Scopus citations


PURPOSE: Estimating the effects of continuous chronic disease risk factors on mortality is an area that generates confusion and controversy. The frequently observed U-shaped or J-shaped relationships between the risk factors and mortality are often in contrast with presumed monotone relationships. Therefore, some investigators suggest that subjects dying during the first k years of follow-up (where k is some positive number less than the total length of follow-up) be excluded from statistical analyses. The rationale for this approach is that subjects dying during the first k years of follow-up are likely to have some pre-existing occult disease that confounds the relationship between the risk factors and mortality. Excluding such subjects purportedly reduces bias due to this confounding. The purpose of this study was to test the effects of excluding subjects who die during the first k years of follow-up on the reduction of bias under a variety of situations. METHODS: Using body mass index (BMI; kg/m2) as an example, we conducted Monte Carlo simulations to investigate such effects. RESULTS: Results suggest that under the conditions investigated, the method of excluding early deaths does not reliably or substantially reduce bias due to confounding introduced by occult disease. CONCLUSION: Excluding subjects dying during the first k years of follow-up may not be a judicious strategy for handling confounding due to occult disease. Investigators are encouraged to develop alternative methods.

Original languageEnglish (US)
Pages (from-to)132-142
Number of pages11
JournalAnnals of Epidemiology
Issue number2
Publication statusPublished - Feb 1 1999



  • Body Mass Index
  • Confounding
  • Cox Regression
  • Gompertz Distribution
  • Logistic Regression
  • Survival Analysis

ASJC Scopus subject areas

  • Epidemiology

Cite this