Skip Navigation LinksHome > March 2012 - Volume 23 - Issue 2 > Negative Control Exposures in Epidemiologic Studies
Epidemiology:
doi: 10.1097/EDE.0b013e3182460c23
Letters

Negative Control Exposures in Epidemiologic Studies

Lipsitch, Marc; Tchetgen Tchetgen, Eric; Cohen, Ted

Free Access
Article Outline
Collapse Box

Author Information

Department of Epidemiology Harvard School of Public Health Boston, MA Department of Immunology and Infectious Diseases Harvard School of Public Health Boston, MA Center for Communicable Disease Dynamics Harvard School of Public Health Boston, MA mlipsitc@hsph.harvard.edu (Lipsitch)

Department of Epidemiology Harvard School of Public Health Boston, MA Center for Communicable Disease Dynamics Harvard School of Public Health Boston, MA Department of Biostatistics Harvard School of Public Health Boston, MA (Tchetgen Tchetgen)

Department of Epidemiology Harvard School of Public Health Boston, MA Center for Communicable Disease Dynamics Harvard School of Public Health Boston, MA Division of Global Health Equity Brigham and Women's Hospital Boston, MA (Cohen)

Back to Top | Article Outline

The authors respond:

We thank George Davey Smith1 for his thoughtful comments on our paper2 and for drawing attention to his previous use of a negative control exposure. We agree with his interpretation that the lack of association of paternal smoking during pregnancy (negative control exposure B as shown in Fig. 3 in our original paper,2 and reproduced here in simplified form as the Figure) with birth weight strengthens the causal interpretation of an observed association of maternal smoking (A) with low birth weight (Y). Comparing Figure 1a and b of Davey Smith, paternal smoking has a univariate association with low birth weight, but this association disappears in a model including maternal smoking. This example emphasizes why the negative control exposure (B) should be evaluated in a model including the exposure of interest (A). The negative control is used to see whether there is evidence for a causal arrow from unobserved confounders (U) to the outcome (Y). Even if no U→Y arrow exists, B will be associated with Y under the alternative hypothesis (A→Y) through the path B←U→A→Y. Conditioning on (A) by including it in the model will close this path.

Figure. Causal diagr...
Figure. Causal diagr...
Image Tools

This example also nicely illustrates the subject-matter knowledge needed to interpret negative controls. A potential problem with paternal smoking as a negative control for maternal smoking is that paternal smoking causes maternal passive smoking; therefore, paternal smoking might be associated with the outcome causally. Observing an association of paternal smoking with birth weight would thus not invalidate in utero effects interpretation of the maternal smoking-birth weight association in this study. Nonetheless, the lack of association with paternal smoking in the birth weight case is reassuring.

The later-life BMI outcome in offspring discussed by Davey Smith is possibly more complicated. Postnatal outcomes could be associated with in utero and postnatal exposures. The finding of an association with paternal smoking during pregnancy (B) after adjustment for in utero exposure to maternal smoking during pregnancy (A) could reflect any combination of three nonmutually exclusive possibilities: (i) uncontrolled confounding of the A→Y association; (ii) an additional causal link B→Y via in utero exposure to passive smoke; or (iii) that some or all of the association A-Y and B-Y is due to effects of postnatal exposure to smoke from either parent. That is, the observed association between parental smoking during pregnancy and adiposity may reflect a causal relationship between postnatal parental smoking and adiposity, combined with a tendency of parents who smoke during pregnancy to continue smoking after the baby is born.

Marc Lipsitch

Department of Epidemiology

Harvard School of Public Health

Boston, MA

Department of Immunology and Infectious Diseases

Harvard School of Public Health

Boston, MA

Center for Communicable Disease Dynamics

Harvard School of Public Health

Boston, MA

mlipsitc@hsph.harvard.edu

Eric Tchetgen Tchetgen

Department of Epidemiology

Harvard School of Public Health

Boston, MA

Center for Communicable Disease Dynamics

Harvard School of Public Health

Boston, MA

Department of Biostatistics

Harvard School of Public Health

Boston, MA

Ted Cohen

Department of Epidemiology

Harvard School of Public Health

Boston, MA

Center for Communicable Disease Dynamics

Harvard School of Public Health

Boston, MA

Division of Global Health Equity

Brigham and Women's Hospital

Boston, MA

Back to Top | Article Outline

REFERENCES

1. Davey Smith G. Negative exposure controls in epidemiologic studies [letter]. Epidemiology. 2011;23:351–352.

2. Lipsitch M, Tchetgen Tchetgen E, Cohen T. Negative controls: a tool for detecting confounding and bias in observational studies. Epidemiology. 2010;21:383–388.

© 2012 Lippincott Williams & Wilkins, Inc.

Twitter  Facebook

Login

Article Tools

Images

Share