Читать книгу Understanding Clinical Papers - David Bowers - Страница 21

A neat way of tackling questions about the cause of disease or other health events is to sit in a library (or, more likely, at a computer), locate routinely collected data, and put population data about disease frequency (e.g. regional deaths from lung cancer) together with data about exposure to a risk (e.g. regional data on tobacco consumption). By so doing, you might find that regions with high lung cancer death rates were also the ones with high tobacco consumption. Suppose also that you found the corollary – that low mortality areas were associated with low tobacco consumption – then your findings would support a link between the supposed risk (smoking) and the target disorder (lung cancer).

The real study used as an example in Figure 6.1 concerns whether smoking might be an important risk factor for the loss of life arising from fires in people's homes. Researchers from Atlanta, Georgia looked into this observation by assembling data for the year 2004 from two separate databases – one that holds death certificate data by state and another that provides smoking data by state. The graph set out in Figure 6.1 shows how the researchers use the data to shed more light on whether smoking is associated with death in domestic fires.

Figure 6.1 Findings from an ecological study about smoking and domestic fires.

Source: From Diekman et al. (2008), © 2008, BMJ Publishing Group Ltd.

But you may by now have spotted a flaw in this type of study: we don't know whether the individual people who died in house fires were smokers. Put another way, it is possible for a study of this design to come up with these findings even if every person who died in a house fire was a non‐smoker. This flaw is sometimes called the ecological fallacy and is a consequence of the use of aggregated data rather than the more usual research method of collecting data for each individual study participant. The other three types of analytic study set out below are more satisfactory approaches to cause‐and‐effect questions because they are able to relate the supposed risk factor directly to the outcome in each study participant.