A Decision Process Model for De-Identification Methods on the Example of Psychometric Data

After years of delay, Germany will offer its citizens with statutory health insurance the voluntary use of an Electronic Health Record (EHR) by law from 2021. Patients are able to upload their medical records into the EHR and from 2023 onward, they will be able to donate their data for research purposes. The donated data will be collected and stored at the Research Data Center (RDC). De-identified datasets can be requested from the RDC, which has to choose a method for de-identification and is responsible for the safety of the published data. Since every dataset is unique, de-identification methods have to be found and chosen individually. Within this work, a decision process model is presented and demonstrated on the example of psychometric data from a digital dementia screening application. As privacy model, the k-anonymity is chosen to demonstrate a desired grade of anonymity. Datafly, Incognito and Bottom-Up Generalization are compared in detail as suitable anonymization algorithms to achieve a dataset that fulfills k- anonymity for a given value of k.