Portal de Periódicos da CAPES

Some aspects of retrospective studies

1960; Pergamon Press; Volume: 11; Issue: 5 Linguagem: Inglês

10.1016/0021-9681(60)90016-3

1878-0679

Jerome Cornfield, William Haenszel,

Meta-analysis and systematic reviews

The or case-control study provides a more economical way of estimating the relative risk of disease than the prospective method because it does not require the devotion of considerable study resources to those who did not develop the disease. In such a study one identifies all or a well-defined sample of the new cases of a disease as they occur during some period of time and only after the occurrence of the disease are they classified by the presence or absence of the characteristic. Those who did not develop the disease during the period are also sampled and similarly classified by presence or absence of the characteristic. A basic assumption for estimating relative risk from the retrospective study is that it is possible to enumerate all new cases of a disease or a representative sample of them without having to observe all individuals in the population from which they arise and watching for cases to develop. This assumption might be accurate if all new patients with the disease sought medical attention all medical sources to which such patients might go were completely canvassed and an effective system for reporting such cases was in operation. In actuality these conditions may be far from satisfied. Another and closely related assumption which requires examination is that the sample of individuals not developing the disease supplies an unbiased estimate of the prevalence of the characteristic under study among the entire nondiseased population of interest. Most retrospective studies are satisfied to select a group consisting of individuals with some disease other than that being investigated and to assume that the prevalence of the characteristic in that control group is an unbiased estimate of the required proportion. A discussion of Pearls study of the association between cancer and tuberculosis is included as an illustration that this can be a dangerous assumption. There are several ways of guarding against the possibility of error arising from the unrepresentative nature of the controls and these are identified. It is assumed in this discussion that the measures of association can be computed without regard to the effects of other known or suspected variables but this assumption will rarely be true. Methods of eliminating the possible effects of other variables must be considered.