Choosing controls in non-randomised studies (Lancet 2000 DVT and flying)

I wonder what your views are in relation to the risks of deep vein thrombosis (DVT) and long-haul flights? If you have the opportunity to travel by air do you take an aspirin before you go and perhaps even wear support stockings for the journey (one of my senior colleagues does) as well as getting up and walking about on the flight.

As far as I know none of these approaches has been tested in randomised trials on air passengers, so we have to rely on other types of study such as the research letter published in the Lancet recently (Kraaijenhagen RA, Haverkamp D, Koopman MMW, Prandoni P, Piovella F, B├╝ller HR. Travel and risk of venous thrombosis. Lancet 2000;356:1492).

The authors of this study decided that the ideal control group for patients with deep vein thrombosis, which they could confirm on ultrasound or venogram, was the 75% of patients who presented to hospital with clinical signs of a DVT but tested negatively. On the basis of this control group they found that there was no association between any of the forms of recent travel that they asked the patients about (plane, rail or car travel) and whether their swollen leg was shown to have a clot in the deep veins.

They concluded that this was of some reassurance that travel was not associated with DVT, but there is a major flaw in their reasoning. They have made an implicit assumption (which is not discussed in their research letter) that there is NO association between travelling and swollen legs that do not contain a clot.

I am not sure that a great deal is known about the aetiology of clinically suspicious leg swelling that is found negative on ultrasound or venography, and it is at the least plausible that flying could increase the likelihood of this condition occurring. If for example there was a five-fold increase in both types of swollen leg after flying (that is those that are venogram positive and negative), the odds ratio of having previously flown in the past 4 weeks would still be one when the two groups were compared. For this reason I am not personally reassured by the findings of this study and plan to take whatever precautions I can when I am next on a long-haul flight.

When randomisation is possible the comparability of controls and cases should be less of a problem, but in non-randomised studies the assumptions arising from the choice of controls have to be examined carefully!