Tree-based, two-stage risk factor analysis for spontaneous abortion

Am J Epidemiol. 1996 Nov 15;144(10):989-96. doi: 10.1093/oxfordjournals.aje.a008869.


The authors utilize an analytic strategy that combines tree-based analyses with the method of Mantel-Haenszel to evaluate the association of 11 putative risk factors to spontaneous abortion by controlling for 19 potential confounders. Logistic regression is also used for comparison. The data for this study were collected in southern Connecticut, during 1988-1991. The putative risk factors are employment; standing, walking, or sitting more than 2 hours at work; exposure to vibration at work; commuting to work; reaching over the shoulders at work; carrying loads over 9 kg on the job; drinking alcohol or coffee in the first month of pregnancy; and gynecologic problems before pregnancy. The potential confounding factors are maternal age, marriage status, race, years of education mother's height, use of birth control, number of pregnancies, smoking before pregnancy, years of smoking, whether the mother stopped smoking, smoking marijuana, passive exposure to marijuana, chronic problems, infertility cocaine use and history of negative pregnancy outcomes. This analysis indicates that carrying loads over 9 kg on the job at least once a day increases the risk of spontaneous abortion by 70% (relative risk (RR) = 1.71, 95% CI 1.25-2.32). Drinking three or more cups of coffee daily in the first month of pregnancy also elevates the risk of spontaneous abortion (RR = 2.34, 95% CI 1.45-3.77). Reaching over the shoulders at least once daily has a marginally significant impact on spontaneous abortion (RR = 1.35, 95% CI 1.02-1.78). The authors conclude that this analytic strategy offers an efficient approach to the exploration of new risk factors for a disease where many potential confounders already exist.

Publication types

  • Comparative Study
  • Research Support, U.S. Gov't, P.H.S.

MeSH terms

  • Abortion, Spontaneous / epidemiology*
  • Confounding Factors, Epidemiologic
  • Connecticut / epidemiology
  • Decision Trees*
  • Female
  • Humans
  • Logistic Models
  • Pregnancy
  • Risk Factors