Versorgungsnahe Daten zur Bewertung der vergleichenden Effektivität von medizinischen Behandlungen: eine Bestandsaufnahme der verfügbaren Datenquellen in Deutschland unter besonderer Berücksichtigung von Registern

Using real-world data (RWD) for comparative effectiveness research has gained increasing attention. Real-world data is usually not collected with the primary aim of answering questions about the comparative effectiveness of medical interventions. Therefore, data collection is often not optimally designed for this purpose. For this reason, using it can be associated with several data-related or analysis-related problems.

This article has two aims: First of all, we will outline the basic requirements for comparative non-randomized studies based on RWD. In addition, the advantages and disadvantages of potentially eligible RWD in Germany will be considered. In particular, the use of registry data will be discussed, as these currently appear to be the most suitable for comparing the effectiveness of interventions.

There are now various databases in Germany that contain RWD and can potentially be used for comparative effectiveness research. If they contain all the necessary information and if the data are of sufficient quality, they could offer a timely and efficient way of determining the effectiveness of medical interventions. However, our analysis shows that the use of this data is sometimes associated with considerable limitations. RWD is often highly aggregated and thus may not be sufficiently detailed to select the subjects precisely or to emulate the intervention or control interventions satisfactorily. In addition, many data sources only encompass a limited set of variables and limited time horizons (e.g., only hospitalization) according to their intended purpose (e.g., billing). Therefore, it is often questionable whether this includes all endpoints on benefit and harm that are important for the assessment and whether sufficiently long observation horizons/follow-up periods are given. Similarly, it is often questionable whether all necessary data to avoid bias are included. Furthermore, it is often difficult to assess suitability in advance due to the lack of available information.