Application and efficiency of sequential tests in matched case-control studies

Abstract

For large epidemiological observational studies banks of biological samples are sometimes created. To study specific research hypotheses blood, urine or tissue specimens can be obtained from participants to a study and stored for later analysis in a biological bank. Constantly new biochemical, molecular or genetic laboratory techniques are developed. These allow a large number of (new) etiologic hypotheses to be tested on the stored biological material to investigate interesting associations between an exposure and a disease. The amount of stored biological material however is, in general, limited, in particular for the cases (the persons who develop the disease of interest), when the disease is not so common. Furthermore, with most laboratory techniques biological material is destroyed and cannot be used for another test. To combine the large number of interesting hypotheses with the limited number and amount of biological samples statistical methods are needed that can distinguish between more promising and less promising hypotheses at the expense of as little biological material as possible. Sequential statistical methods offer a researcher the possibility to terminate an investigation as soon as sufficient evidence has accumulated to accept the null hypothesis ("no association between exposure and disease") or to reject it in favour of the alternative hypothesis ("an association exists between exposure and disease"). After each new observation or group of observations the accumulated data are tested. Based on the cumulative test result the study is stopped or more information is obtained. A sequential analysis requires, on average, fewer observations to come to a decision ("accept the null hypothesis or reject it") than the corresponding fixed sample size analysis. Sequential methods are thus an efficient way to handle the available data. Matching is used to make cases and controls more comparable by controlling for possible confounding factors. These factors are related to both exposure and disease and may distort the size of the exposure-disease relation. In a case-control study one or more controls can be matched to a case based on the value of the confounding factor. A matched design requires in general less cases and controls than an unmatched design. Matching is thus another way of efficient data handling. The chapters of this thesis describe various sequential tests that can be applied in matched case-control studies. Sequential tests were developed for continuous and dichotomous exposure variables. (Group) sequential analysis was proposed as an alternative for conditional power to stop a study early for 'futility'. Sequential tests for the analysis of gene-environment interactions in matched case-control studies were developed. In the last chapter three approaches to statistical testing theory and their effects on sequential testing were discussed. In general, sequential analysis in matched case-control studies can lead to considerable efficiency gains by saving valuable biological samples, time or money.