Post a new comment on this article
Post Your Discussion Comment
Please follow our guidelines for comments and review our competing interests policy. Comments that do not conform to our guidelines will be promptly removed and the user account disabled. The following must be avoided:
- Remarks that could be interpreted as allegations of misconduct
- Unsupported assertions or statements
- Inflammatory or insulting language
Why should this posting be reviewed?
See also Guidelines for Comments and Corrections.
Thank you for taking the time to flag this posting; we review flagged postings on a regular basis.close
Thrombosis after Travel: reply to a Perspective
Posted by plosmedicine on 31 Mar 2009 at 00:25 GMT
Author: Suzanne Cannegieter
Position: No occupation was given
Institution: Leiden University Medical Center, Clinical Epidemiology
Additional Authors: Astrid van Hylckama Vlieg, Saskia le Cessie, Frits R. Rosendaal, Jan P. Vandenbroucke
Submitted Date: April 28, 2008
Published Date: April 29, 2008
This comment was originally posted as a “Reader Response” on the publication date indicated above. All Reader Responses are now available as comments.
Kenneth Rothman raised some important methodological issues in his ‘Perspective’ on our paper about travel-related venous thrombosis . In our case-control study, we used individually matched controls who were the partners of the cases, to study the effect of travel on the risk of venous thrombosis . Because partners tend to travel together, a simple contrast between cases and controls would give rise to bias towards the null, but this effect was removed by using a matched analysis.
In addition to the overall effect of travel, we also looked at the interaction between travel and other risk factors for thrombosis, such as oral contraceptive use. As partners were of opposed sex in almost all cases, it seemed illogical to analyse by matched pairs when assessing the effects of oral contraceptives, since men never would use oral contraceptives. Thus, it seemed impossible to us to analyse the interaction between travel and oral contraceptive use in a direct stratified analysis over the pairs. We therefore performed a case-only analysis, in which a synergy index of multiplicativity (SIM) is derived from the cases, whichwas then used to calculate the joint effect of travel and oral contraceptives by multiplying with the separate effects that we had analysed before. However, as Rothman pointed out, a case-only analysis assumes that travel is unrelated to oral contraceptive use. Furthermore, a SIM only gives departure from multiplicativity and is difficult to interpret as a departure from an additive relation.
In his ‘Perspective’, Rothman proposed an interesting solution: instead of restricting the analysis to women only, he suggested to include the men simply as persons who were not exposed. Although somewhat counterintuitive at first, it does make sense when one realizes that the reason for not being exposed, even when it is sex, is irrelevant (with the assumption of absence of confounding). To better understand what happens when men are included as unexposed subjects we simulated several situations that are akin to the situation in the original paper to determine the optimal analysis for matching on opposite sex. From these simulations we conclude that including the men results in a more powerful analysis, when no confounding by sex is present. When there is confounding, sex should be included in the model, and be adjusted for. Thus, in this study, where we matched on opposite sex, it would be appropriate to include the men in the analysis (allowing for sex by an indicator variable), as it would permit us to study the combined effect of oral contraceptives and travel directly.
In brief, the original analysis was performed in 1025 women aged under 50. Non-pill users who did not travel were used as the reference group. We found a case-only estimate of the synergy index of multiplicativity (SIM) of 2.4 for women who traveled by car, bus or train. This would imply a 20-fold increased risk of venous thrombosis for women who travel and use oral contraceptives as compared to women who do neither (2.4 times an OR of 4 for oral contraceptive use times an OR of 2 for travel). Likewise, we calculated a SIM for oral contraceptive use of 4.9 for travel by air, which suggested a 40-fold increased risk for women who travel by air and use oral contraceptives .
The new analysis suggested by Rothman was performed in the complete study population, i.e. 1906 cases and 1906 matched controls. We used conditional logistic regression, including travel, oral contraceptive use and an indicator variable for sex in the model. The results are shown in table 1. The odds ratios for the interaction calculated this way are lower than the ones we estimated using the synergy index from the case-only analysis, i.e. 9.7 (95% CI 4.3-22.0) for travel by car, bus or train and 12.9 (95% CI 4.0-42.0) for air travel.
The obvious explanation for this discrepancy is that oral contraceptive use and travel are linked in this population of relatively young women, which we also found in our data: an OR for oral contraceptive use and travel by car bus or train of 1.97; an OR of oral contraceptive use and air travel of 1.7. From this new analysis we conclude that our original estimates of the additional effect of oral contraceptive use in a travelling population were too high due to the case-only analysis. Nevertheless, the corrected results still show a clearly synergistic effect for the two exposures. Results of similar order of magnitude were recently described by Kuipers et al .
The second point raised by Rothman concerned the way we displayed the number of cases by week since travel (our fig 1 in original paper), rather than the relation between relative risk and time since travel. His suggestion was to solve this by using spline regression, a method that estimates separate regression line segments for different time periods since travel. Although this is indeed a useful method for such a situation, in our particular study it turned out not to be feasible. The problem was that the cases and the controls were strongly linked in their travel behaviour, so, of the cases displayed in the figure, only a small number had travelled without their partner at the time. This resulted in erratic variation in odds ratios when calculated per week (table 2), which made the linear regression model as well as a spline model very difficult to fit. The same applied for the analysis of duration of travel. Generally however, one may assume that the travel frequency in the population is roughly equal over a 3 month period, which implies that the graph we made with a decreasing number of cases over time, does indeed reflect a higher risk up to two months after travel.
To view Tables 1 and 2, please paste the following link in your browser - http://www.plos.org/press...
1) Rothman KJ. Thrombosis after travel. PLoS Med. 2006;3:e300
2) Cannegieter SC, Doggen CJ, van Houwelingen HC, Rosendaal FR.Travel- related venous thrombosis: results from a large population-based case control study (MEGA study). PLoS Med. 2006;3:e307.
3) Kuipers S, Cannegieter SC, Middeldorp S, Robyn L, Büller HR, Rosendaal FR. The absolute risk of venous thrombosis after air travel: a cohort study of 8,755 employees of international organisations. PLoS Med. 2007;4:e290.