Smoking and risk of myocardial infarction

BMJ 1998; 317 doi: https://doi.org/10.1136/bmj.317.7164.1017a (Published 10 October 1998) Cite this as: BMJ 1998;317:1017

Statistical and biological interactions should not be confused

  1. Thomas V Perneger, Medical epidemiologist
  1. Institute of Social and Preventive Medicine, University of Geneva, CH-1211 Geneva 4, Switzerland
  2. Department of Community and Family Medicine, Chinese University of Hong Kong, Hong Kong
  3. Department of Clinical Oncology, Nottingham City Hospital, Nottingham NG5 1PB
  4. Institute of Preventive Medicine, Kommunehospitalet, DK-1499 Copenhagen K, Denmark

    EDITOR—Prescott et al report that smoking increases the risk of myocardial infarction significantly more in women (relative risk 2.24) than in men (relative risk 1.43).1 Interactions between components of smoke and hormonal factors were suspected.

    Readers may conclude from this study that men and women do not differ at all. On the basis of data on the prevalence of smoking (table 2 in Prescott et al's paper) and from reported relative risks, we can calculate that in women the risk of developing myocardial infarction during follow up is 5.88% (380/6461) in smokers and 2.63% (132/5011) in non-smokers; in men, the risk is 10.62% (902/8490) in smokers and 7.38% (349/4701) in non-smokers. These are best estimates based on published data; the figures would change slightly if former smokers were removed from the group of non-smokers. The difference that is attributable to smoking was therefore 3.25% in women and 3.24% in men. Over 12 years, smoking caused an additional myocardial infarction in one person out of 31—equally distributed between men and women.

    This shows that statistical interaction should not be confused with biological interaction. Statistical interaction concerns the modelling of combined effects of two or more risk factors for a disease in populations, and biological interaction refers to biochemical reactions in an individual. Whether statistical interaction exists or not depends on the specification of the model that is applied to data—“interaction” means that a model that simply adds the effects of two risk factors (in this case sex and smoking) does not accurately describe their joint effect (the risk of myocardial infarction in men who smoke).

    Prescott et al used a multiplicative model and found a significant interaction; …

    View Full Text

    Sign in

    Log in through your institution