In theory, the best empirical evidence regarding causation should come from randomized trials in humans. Studies in humans do not face the pitfalls inherent in the reasoning by analogy from animal to human data, whereas randomization allows control for distorting influences by both known and unsuspected confounding factors; in addition, doubleblind designs minimize the potential for several types of bias. Indeed, therapeutic clinical trials are also aetiolo-gical studies exploring the causation of a better clinical outcome by a certain process, the study treatment. However, experimental studies of cancer causation cannot be undertaken in humans for ethical reasons, except when there is evidence that a particular factor may actually protect from cancer (in which instance the absence of the factor can be thought of as the carcinogenic agent). Even under these conditions, experimental studies in humans are still impractical. Among cancer patients, the outcome under investigation (death, metastasis or recurrence) is a relatively frequent effect, and the corresponding study size can be manageable. Among healthy individuals, the frequency of occurrence of any particular cancer is low and the corresponding study size must be very large, making the follow-up and compliance problems exceedingly difficult. Nevertheless, a few such studies have been undertaken, either by investing large resources (e.g. the IARC study targetting the prevention of hepatocellular carcinoma in the Gambia through active immunization against hepatitis B virus) or by reducing the required sample size by focusing on preneoplastic conditions that identify high risk individuals.
Epidemiological studies specifically designed to address a particular aetiological hypothesis are usually called analytical. The objective of analytical studies is to document causation between exposures and a certain disease. In analytical investigations, measurements and categorical assignments apply to individuals, whereas this is not necessary in descriptive epidemiological studies. Thus, in order to examine in an analytical study whether chronic carriers of hepatitis B virus (HBV) are more likely to develop liver cancer than noncarriers, it is necessary to classify the individuals under study according to their HBV carrier state and the development or not of liver cancer during a specified time period (Hennekens and Buring, 1987; MacMahon and Trichopoulos, 1996; Rothman and Greenland, 1998).
Ecological studies in epidemiology, as opposed to individual-based studies, occupy an intermediate position between descriptive and analytical investigations, in that they share many characteristics with descriptive studies, but may also serve aetiological objectives. In ecological studies, the exposure and the disease under investigation are ascertained not for individuals but for groups or even whole populations (Morgenstern, 1982). Thus the prevalence of HBV in several populations could be correlated with the incidence of liver cancer in these populations, even though no information could be obtained as to whether any particular individual in these populations was or was not an HBV carrier and has or has not developed liver cancer.
When an exposure is fairly common (e.g. smoking, sunlight, poverty, even prevalence of HBV carriers), ecological studies should be able to reveal the effects of these exposures. Thus, following the increase in tobacco consumption, the incidence of lung cancer increased sharply over time; skin melanoma is more common in geographic latitudes with more sunshine exposure; stomach cancer is generally more common in low-income social strata; and the incidence of primary liver cancer is higher in populations with higher prevalence of HBV (Tomatis, 1990). As a corollary, the inability of ecological studies to demonstrate an association between a widespread exposure that has rapidly increased over time (e.g. extremely low-frequency magnetic fields) and the incidence of a disease allegedly caused by these fields (e.g. childhood leukaemia) challenges the causal nature of the positive association reported from some analytical investigations.
In analytical epidemiological studies, there are several ways through which an association, or lack thereof, is assessed, but the most common measure is the relative risk. A value equal to 1 implies that the exposure under study does not affect the incidence of the disease under consideration. In contrast, values < 1 and > 1 indicate, respectively, an inverse or a positive association (MacMahon and Trichopoulos, 1996; Rothman and Greenland, 1998). When the relative risk is >2 and the associated unbiased, unconfounded, precise and causal, an exposed-case patient is more likely than not to have developed the disease because of the exposure. When the relative risk is >1 but <2 the exposed patient is more likely than not to have developed the disease for reasons other than the exposure. This is because a relative risk of, say, 1.5 has a baseline component equal to 1 that characterizes the unexposed, and a component equal to 0.5 that applies only to the exposed (MacMahon and Trichopoulos, 1996). For instance, if the risk of a nonsmoking 55-year-old man developing lung cancer in the next 10 years is 1%, and that of a same age and gender smoker is 5% (relative risk 5), only 4% in the smoker's risk (i.e. 4/5 of the total 5%) can be attributed to his smoking.
Was this article helpful?
Save Your Lungs And Never Have To Spend A Single Cent Of Ciggies Ever Again. According to a recent report from the U.S. government. Centers for Disease Control and Prevention, more than twenty percent of male and female adults in the U.S. smoke cigarettes, while more than eighty percent of them light up a cigarette daily.