Missing values and estimands in diagnostic studies

点击翻译按钮获取中文摘要

The concept of estimands has recently been implemented in therapeutic trials. An estimand represents the underlying research question and is defined by the attributes population, treatments, variable for measuring the individual treatment effect, summary statistics and intercurrent events. In particular, the differentiated handling of intercurrent events, which are often equated with protocol violations, is the new element. An example of an intercurrent event would be a rescue medication during the course of the study. Furthermore, strategies for dealing with intercurrent events are prespecified, e.g. the strategy could be not to use the results after a rescue medication (while-on-treatment strategy). Depending on the strategy and the assumptions for possible missing values, a suitable theoretical estimator results. In the example, a possible estimator would be the mean treatment effect, whereby in the case of rescue medication, only the data collected beforehand are considered. In diagnostic studies, intercurrent events as well as missing values and appropriate statistical methods to deal with them are also important topics. However, for diagnostic studies the concept of estimands does not yet exist and there are still large gaps in methods for dealing with missing values. For this reason, the present project aims to transfer the concept of estimands from therapeutic to diagnostic trials. For this purpose, the attributes of the estimands and strategies for dealing with intercurrent events will be adapted for diagnostic studies. Furthermore, statistical methods for dealing with missing values will be further developed on this basis in order to arrive at suitable estimators. This joint concept of estimand and methods for handling missing values will be evaluated in simulation studies and for example studies.Furthermore, adaptive designs for diagnostic studies with regard to estimands and missing values will be developed. Adaptive designs allow prespecified adjustments in the course of the study and can lead to more efficient studies, but they also have their limitations. Therefore, adaptive designs will be developed and evaluated that allow, for example, sample size re-estimations based on frequencies of intercurrent events or design changes that imply modifications of the estimand. Due to the new European regulation of medical devices, the requirements for diagnostic studies will increase considerably in the near future. The research project will provide recommendations, which will be disseminated in the form of corresponding guidance documents and in workshops and at a symposium. In addition, the methods developed are to be implemented in freely accessible software, thus enabling easy application.