We describe a searchable patient record database for decision support. It contains medical histories of real but pseudonymous patients with patterns of diagnosis, chosen treatment, and outcome. To be searchable, the patterns contain a feature vector (for similarity search by calculating distances) and a globally unique "pattern name" which identifies the kind of data which are represented by the feature vector. Patterns with the same pattern name are directly comparable; they represent the same kind of data. For pattern selection the database provides a growing well-structured list of initial diagnoses with associated input masks.
Procedure: The doctor can assume that the database contains patients similar to the current patient if he finds his initial diagnosis in the list. Clicking on it opens an associated input mask which requests specific further data for finer differentiation. After input a searchable pattern group is built from the provided data, and used to search for histories of patients with similar fine diagnostics, and for the most successful treatment decisions at these patients. This information can be very valuable for deciding the treatment of the current patient. Because the database can collect patient histories from all countries, in the long run this could open access to a wealth of experience which by far exceeds the capacity of a today's doctor.