Last update:
24-Apr-2007

Most studies in health sciences investigate the etiology of diseases. Data which are collected from applied research provide the possibility to investigate the relation between disease frequency and the characteristics of the patients (determinants). However, the probabilistic relation between a determinant and a disease can be considered as causal only if the relation still exists after counteracting all known and unknown confounders. This could be done only experimentally, but experiments on human beings are prohibited. How then can it be answered whether a frequency relation is causal? A widely adopted approach is the use of causal criteria. These criteria are certain prerequisites which it is believed that a probabilistic relation should fulfill in order to be considered as causal. Serious criticism has been made of this approach because it has many inherent difficulties. A more correct approach would be to develop and test competing hypotheses. With this approach, the probabilistic relation between a disease and a characteristic of a person could be considered causal if any attempt to show that this relation is due to another determinant or a study bias has failed. In effect the characteristic is not a complete causal mechanism (sufficient cause), but only one of its necessary elements, called a component cause. In practice, there are usually many different sufficient causes which produce the same effect, each one of which is composed of many components, and therefore, many valid and precise studies should be made in order to investigate all component and sufficient causes.

Key words: Causal criteria, "Causal pie" model, Component cause, Necessary cause, Sufficient cause.