We were ambushed while eating our lunch before a clinical study session. We had expected to have a relaxing afternoon but suddenly found ourselves agreeing to critically appraise an epidemiological study. The study investigated the risk of venous thromboembolism after a long haul flight. At least the topic was interesting and controversial, but since we had not enjoyed our epidemiology lectures, at first we were reluctant. But epidemiology might not be as mind numbingly boring and difficult as we originally presumed. The study seeks to clarify whether the relation between long haul flights and venous thromboembolism is an urban myth created to deter sun worshippers from travelling to far off countries or is it a terrifying reality?

Why do the trial?

Sometimes young fit patients have presented to hospital with unexpected venous thrombosis or have even died suddenly. We now have advice, guidelines, and preventive strategies but little objective evidence.

The argument, however, seems plausible. Airline passengers are immobile for relatively long periods of time, often restricted in movement, sometimes dehydrated, and perhaps partially sedated by alcohol or benzodiazepines. Clinicians were aware of a potential problem, but epidemiologists had yet to provide confirmation. Finding the evidence is difficult because finding information on passengers and subsequent thromboembolic events is incredibly difficult.

What did the investigators do?

The authors of this study chose Western Australia for a number of reasons. Australia has kept excellent records on international travellers since 1970-the country has an excellent source of data on travellers. Identifying subjects for follow up is important in all epidemiological studies, but this can often be difficult to obtain. The remote location of the airport ensured that all arrivals had undertaken a long haul flight. To identify travellers who developed a venous thromboembolism, data from flight records were probabilistically matched to data from hospitals in Western Australia. The authors used an established protocol that preserves confidentiality and maximises specificity. Maximising specificity ensures that a name in flight records matching one in hospital data is of the same person.

Disadvantages of the study

Sources of data in any study will always have some disadvantages-for example, input error in the flight records or inaccuracies in the hospital discharge summaries, which are often completed by junior doctors with heavy workloads. All epidemiological studies are prone to other forms of bias; this study has three important sources of potential bias. Firstly, data was only collected for arrivals to Perth because it would be nearly impossible to follow up departures travelling to other parts of Australia or further. Secondly, analysis of arrivals focused only on Australian arrivals because they were more likely to stay in the area during the follow up period. Non-Australian arrival data were collected but were not analysed in the same detail because of the potential bias. Finally, as data were analysed over 18 years, between 1981 and 1999, variation in the quality of data may have affected the records. However, this was overcome by analysing data subsets, which showed consistent estimates. Other potential forms of bias include deaths before admission to hospital due to a venous thromboembolic episode immediately after flying, failure to link arrival and hospital admissions with the matching protocol, and increasing outpatient treatment meaning that venous thromboembolism would not show in hospital discharge summaries. The authors did their best to minimise the effects of bias although possible inaccuracies in the data may be considered limitations. This could occur in any epidemiological study.

LYNN SAVILLE/PHOTONICA

Sitting tight: does air travel lead to deep vein thrombosis?

The authors now have information on the proportion of passengers who developed a venous thromboembolism. They now need to know the incidence of venous thromboembolism in the general population. From where the authors got the general population incidence is not totally clear, but state-wide incidences from hospital discharge summaries appear to be the sources. The researchers then had to consider whether the risk of venous thromboembolism after flying is temporarily increased.

What were the findings?

To help while reading the article we put together a simple table to show the 153 Australian arrivals with venous thromboembolism within 100 days of flying between 1981 and 1999 from the 4.8 million arrivals over the same period (table).

Of 16 205 hospital admissions for venous thromboembolism (in 13 184 patients), 153 Australian patients were diagnosed with a venous thromboembolism within 100 days of an international flight-this number was less than expected. We had always been led to believe that travelling by air can increase the chances of developing deep vein thrombosis but the results contradict this. Why? Maybe the groups were not comparable; authors describe this as the "healthy traveller" effect-those travelling are generally fitter and healthier than the normal population.

The authors then looked at the distribution of venous thromboembolism in the period after flying. Assuming that venous thromboembolism was unrelated to flying, cases should be distributed equally in time. However, a much greater number of patients were admitted within the first 14 days (46 out of 153) than would be expected under the equal distribution assumption. During this 14 day hazard period, venous thromboembolism was four times more likely.

The authors also examined the non-Australian group, despite the uncertainties regarding collecting data about this group. In this group there were more episodes of venous thromboembolism within 2 weeks of arrival than would have been expected compared with state-wide incidence. Could this be that the group travelled further, from Europe perhaps, or are less healthy than Australians? As data were not available to compare flight times between these groups we certainly were not convinced that the non-Australian group was unhealthier than the Australian group.

Making conclusions from these findings is difficult. The authors, however, did try to explain their results. There are so many potential factors that identifying particular sources of selection bias to explain findings that were unexpected is unfair.

Statistical analysis

Statistical analysis is simply a tool. The most important part of any study is the method. If we can decide what we really want to know and refine our research question, it helps statisticians to decide which instrument to choose from their toolkit. Different methods need different tests. In this study, the authors wished to calculate the conditional probability of having a thromboembolic event within a period of time after flight. They describe what they did in an appendix to the paper (see bmj.com). The formulas look complex, but they just describe probability, taking different factors into consideration. The concept is straightforward, but, in this case, the mathematics does require statistical expertise.

Was this a good study?

Few other epidemiological studies consider this contemporary issue. This is a good study. The authors produced a paper that even those of us with basic epidemiology can follow, they were aware of potential sources of bias, and they tried to minimise these as much as possible. The results were unexpected but suggested factors could have affected the results. We were not entirely convinced as data were not available for confounding factors, such as flight duration, but at least a breakthrough has been made in our conflict with epidemiology.