KIP-SDM
A.I. Modeling in nursing care - Fall prediction and development of an open-source data repository
The overall goal of the alliance is to simplify data access for caregivers as well as for AI researchers in the use case of fall prevention. Nursing staff will thus benefit from a novel AI solution for risk assessment and intuitive access to relevant patient data. AI researchers benefit from simplified development on otherwise hard-to-access data and faster integration into hospital information systems (HIS).
You are here:
Motivation
Every year, there are almost 5 million registered fall incidents in Germany. For older people in particular, the consequences and complications can be long-lasting and devastating. The annual costs for fall treatments amount to more than 500 million euros. Yet up to 30% of all falls are preventable (Hshieh et al. 2018).
Systems using artificial intelligence (AI) analyze risk factors, predict individual fall risks, and could thus digitally support fall prevention for caregivers. However, existing systems often do not consider all relevant risk factors (Seibert et al. 2020). AI-based fall prevention, for example, is often based on gait analysis, even though the risk of falling is increased by 56 percent for many patients taking half the daily dose of hypnotics and sedatives.
Medication data, however, are often unavailable for use in AI systems because access to the data is technically and legally complex. Thus, the potential of AI-based systems for fall prevention has not yet been fully realized.
Innovations and perspectives
To preserve patient privacy, patient data will never leave the respective institution. Instead, generative deep learning models are learned on the data of the participating institutions, which can generate realistic patient data.
These generative models will be embedded in standardized runtime environments, implemented through Docker containers, and can be shared among the institutions as well as with external AI researchers.
The use of privacy-preserving decentralized Deep Learning approaches thus allows to provide realistic data without actually sharing real patient data. At the same time, providing a standardized runtime environment allows rapid development of AI applications externally as well as data analysis on-site at care facilities under uniform conditions without integration efforts.
The provision of this novel infrastructure now allows for the first time the development and testing of guideline-compliant AI-based nursing fall prevention across multiple facilities.
