These systems analyze known and identify latent risk factors to predict the risk of falls and thus enable individual preventive measures. However, existing systems often leave out relevant risk factors (Seibert et al. 2020). Frequently, only gait analyses are used and changes in medication, for example, are not taken into account, even though the risk of falling increases by 56 percent in many patients just by taking a half dose of hypnotics and sedatives. One reason for the incomplete consideration of fall risk factors is the technically and legally complex data access. As a result, the potential of AI systems in the care sector has not yet been fully exploited.

The goal of the KIP-SDM project is to research AI-based fall prevention in nursing, as well as to develop a decentralized data repository with nursing treatment data. To achieve this, predictive models for fall prediction will be developed using privacy-preserving decentralized deep learning approaches, which will consider all relevant risk factors. The data used for this purpose comes from two large nursing institutions and a startup. To maintain privacy, this data will not leave the respective institutions. Instead, the generative deep learning models will learn from the data and generate realistic, synthetic patient data. Thus, realistic data can be provided without having to share real patient data. Fall prevention is just one example from a range of similar nursing problems such as pressure ulcers, incontinence, delirium, etc. Using the data integration and data analysis methods developed in the project as well as the AI application, alternative relevant research questions, data, and outcomes can also be evaluated. The novel infrastructure enables for the first time the development and validation of guideline-compliant AI-based nursing fall prevention across multiple institutions.