Sammanvägning av parameterskattningar i fallkontrollstudier med matchade och slumpmässigt valda kontroller

Abstract

In statistical studies, it is a common practice to have a number of study objects that are sampled from a population. Then they are followed for the purpose of studying how a factor of interest develops over time. For epidemiological studies in particular, this is a typical approach. However, if the event of interest in the study occurs rarely, this approach is often impractical. In these cases, it might be more appropriate to carry out a retrospective study, a study of events that have already occurred. One type of retrospective study that is frequently used, is the case-control study. The basic concept of a case-control study is that a number of cases of the event of interest is selected, from a population. Then controls, “non-cases”, are sampled from the same population, and a statistical analysis is performed on this group of cases and controls. The controls can be sampled randomly, or they could be selected by matching certain factors against those of the cases, for example gender or age. These two different methods for selecting controls come with various advantages and disadvantages. Therefore, it would be beneficial to be able to combine results produced using each of the methods, in a way that preserves their advantages, but minimizes their drawbacks. This is the main purpose of this study. We want to investigate to what degree it is possible to improve the accuracy of the results from a case-control study, by using a combination of the results from two substudies, using matched and randomly sampled controls, but with the same cases. The data used in our study are produced from computer simulations. For simulations and calculations we use different packages within the R programming language. The statistical methods we mainly use are logistic regression, conditional logistic regression, bootstrapping and the least squares method. We conclude from our analysis that it is possible to achieve an essential improvement in accuracy of the results, by comparing results from studies with matched and randomly sampled controls.