Reuse of controls in nested case-control studies.

A frequently used study design within epidemiology is large cohort studies. In some situations additional information which is not included in the cohort, for instance biological material stored in biobanks, is required. The availability of such material is usually limited and might be very expensive to analyze for the entire cohort. A reasonable strategy is therefore to restrict the study sample to a subset of the original cohort. Two such designs are nested case-control1,2 and case-cohort3 studies. With a nested case-control design, m controls are sampled for each case. The controls are required to be alive and event free at the time the case experienced the event. Due to those requirements, we say that the controls are matched on time, or at risk status. With a case-cohort design, a subcohort is sampled at the outset of the study, and this subcohort is used as a reference population at all event times. In many studies, more than one endpoint or type of event are of interest. Examples of this can be settings where two or more types of events ”compete”, e.g. death from cancer and death from cardiovascular diseases. Another example could be situations where one endpoint is a subset of another endpoint, e.g. incidence of prostate cancer and subsequent death from prostate cancer. As the controls are matched to their respective cases in a nested case-control design, using them for other endpoints have traditionally not been considered possible. Thus, in the first example only the controls sampled for the endpoint in question can be used. In the second example, all controls sampled for incident cases that did not die from prostate cancer can not be included in the analysis towards death from prostate cancer. The casecohort design, on the other hand, has often been the design of choice if it was to be carried out analyses towards multiple endpoints. Since the subcohort is a random sample from the cohort, it can be used as control population for all types of events. In recent years, methods have been developed that allow for breaking the matching in nested casecontrol designs.4–9 This opens up the possibility of reusing controls for other endpoints. Even though these methods have been around for some time, it does not seem that they have been picked up by epidemiologists yet. We will show that it is fairly easy to reuse controls also within the nested casecontrol design and that this in many situations can give large efficiency improvements. Meyer et al.10 showed in a recent paper that serum 25-hydroxyvitamin D (s-25(OH)D) was positively related to the risk of prostate cancer in a nested case-control study. We will re-analyze this study with inverse probability weighting (IPW).4–8,11 The main endpoint in Meyer et al.10 was incidence of prostate cancer. To investigate efficiency improvements with IPW and illustrate reuse of controls, we are also going to analyze the endpoint death from prostate cancer and subgroups of cases based on metastasis status.