Essay Essays are opinion pieces on a topic of broad interest to a general medical audience.

See all article types »

Observational Research, Randomised Trials, and Two Views of Medical Science

  • Jan P Vandenbroucke
  • Published: March 11, 2008
  • DOI: 10.1371/journal.pmed.0050067

Reader Comments (5)

Post a new comment on this article

Interaction not significant

Posted by rconroy on 31 Jan 2011 at 11:37 GMT

I was curious about the data presented in the paper on the interaction between Factor V Leiden and oral contraceptive use, and thought it would be a good teaching example for my students. However, on analysing the data, I find that my results disagree with Dr Vandenbroucke. The odds ratio for the interaction is 1.3, 95% CI 0.2 to 9.2) so the contention that women who are Factor V positive and use oral contraceptives are at "much higher risk of venous thrombosis than women with either risk factor alone" is not supported by the data provided.
Regrettably, I have no access to the original article.

No competing interests declared.

RE: Interaction not significant - author response

Vandenbroucke replied to rconroy on 06 Feb 2011 at 09:47 GMT

Dr Conoy looks at interaction on a relative risk scale, which is not the way epidemiology looks at interaction: today’s epidemiologists use the risk difference scale to judge whether interaction is causal. This is in contrast to the statistical view which is merely descriptive and in which the scale does not matter. That causal judgement about interaction should be on a risk difference scale was first proposed in the 1980s (1), and is now common teaching, as witnessed in a leading textbook (2), in STROBE (3, item 12b, box 8, and item 17), and in overviews (4). It has been upheld by the latest insight in counterfactual and component causes thinking (5). If data are available on a relative risk scale, they should be transformed to a risk difference scale to judge interaction (1-5). In effect, the reasoning is very simple: oral contraceptives in the absence of factor V Leiden gave a relative risk of about 4 for venous thrombosis. In the presence of factor V Leiden the relative risk was about the same (as calculated by Dr Conoy). However, the presence of factor V Leiden in itself (without contraceptive use) had a relative risk of about 7 for venous thrombosis. When both were present, the relative risk of venous thrombosis, in comparison to the absence of both, became about 35. The last figure indicates a much higher risk with the combined exposure, relative to none, than with either factor V Leiden or oral contraceptives alone – which becomes obvious when subtracting the relative risks (1, 3). In common language, the people who carry factor V Leiden have a large increase in baseline risk; so, if a more or less similar relative risk of oral contraceptives is applied to them, their absolute risk becomes much higher than if they would not have carried factor V Leiden. This phenomenon is sometimes called ‘public health interaction’.

To the credit of Dr Conoy he avows that he had not looked up the original publication when making his comments. At the time of our publication in 1994, we realized that this reasoning would be difficult for many readers – despite the fact that this was already common knowledge in epidemiology for more than a decade. Therefore, we back calculated to incidences. These are presented in table 2 of the original paper (6): the baseline incidence of venous thrombosis per 10,000 women-years was 0.8, for oral contraceptives alone it was 3.0, for factor V Leiden alone it was 5.7 and for both together it was 28.5. This represents a large interaction on a risk difference scale, which now can be seen when subtracting the incidences. In causal terms, this means that there were several women who would not have developed venous thrombosis because of Factor V Leiden alone, nor because of oral contraceptives alone, but only if both factors were present. Actually, this is the same phenomenon that clinicians describe when they say that a bacterium is not sensitive to ampicilline, nor to gentamycine, but completely sensitive to the combination of both antibiotics: in this instance it is called ‘pharmacologic interaction’. As an additional note, to pre-empt any further debate about the statistical significance of our findings, in the discussion section of our paper we explain why we had confidence in our estimates, despite the small numbers of Factor V Leiden controls.

Jan P Vandenbroucke

(1) Rothman KJ, Greenland S, Walker AM. Concepts of interaction. Am J Epidemiol 1980;1112:467-70.

(2) Rothman KJ, Greenland S, Lash TL. Chapter 5: Concepts of interaction. In: Modern Epidemiology, 3rd Ed, Wolters 2008

(3) Vandenbroucke JP, von Elm E, Altman DG, Gøtzsche PC, Mulrow CD, Pocock SJ, Poole C, Schlesselman JJ, Egger M; STROBE Initiative. Strengthening the Reporting
of Observational Studies in Epidemiology (STROBE): explanation and elaboration. PLoS Med. 2007 Oct 16;4(10):e297.

(4) Knol MJ, Egger M, Scott P, Geerlings MI, Vandenbroucke JP. When one depends on the other: reporting of interaction in case-control and cohort studies. Epidemiology. 2009;2:161-6.

(5) VanderWeele TJ, Robins JM. The identification of synergism in the sufficient-component-cause framework. Epidemiology. 2007;18:329-39.

(6) Vandenbroucke JP, Koster T, Briët E, Reitsma PH, Bertina RM, Rosendaal FR. Increased risk of venous thrombosis in oral-contraceptive users who are carriers
of factor V Leiden mutation. Lancet. 1994;344:1453-7

No competing interests declared.