Definition of Variables in Hospital Discharge Data: Pitfalls and Proposed Solutions.
|Autor||Nimptsch U, Spoden M,
DOI: 10.1055/a-0977-3332 
In Germany, the Diagnosis-Related Group Statistics (DRG Statistics) supply full coverage of inpatient episodes in acute care hospitals. The Research Data Centres of the Federal Statistical Office and the Statistical Offices of the Federal States provide the microdata of the DRG Statistics, namely hospital discharge files of each inpatient case, for scientific research. Hospital discharge data are generated for administrative purposes. As well as other data sources, they have specific features and characteristics, which should be considered in planning and designing research studies. A key challenge is the appropriate and sophisticated operationalization of units of analysis, targets variables, and other study variables. The methodological approach should consider, among other factors, differing coding behaviour between hospitals in order to minimize the risk of bias. This contribution shows by practical examples what should be incorporated in variable definition to ensure that the risk of bias by coding behaviour or other factors is minimized to the greatest possible degree. First of all, the features and characteristics of the German hospital discharge data are outlined. Based on the authorsʼ experiences, basic steps and challenges in observational health services research studies are described. Examples are illustrated by our own calculations, derived from previous studies based on the microdata of the DRG Statistics. The reliability and validity of analyses based on hospital discharge data are crucially dependent on the appropriateness of variable definition. To minimize the risk of bias and misinterpretation, extensive preliminary considerations are required which involve clinical aspects, as well as the context of data collection and technical classification opportunities. Hopefully, there will be greater acceptance of research based on hospital discharge data, so that these valuable data will be used more frequently for research purposes in the future.