Most falls in older people occur when walking, with a significant proportion occurring shortly after physical exertion. Coordination and reactive balance are temporarily impaired after moderate to heavy exertion, which increases the risk of tripping and falling. At the same time, regular physical activity is essential to maintain mobility, independence and quality of life in old age.

The "PerDic-Life" project is investigating whether the temporarily increased risk of falling can be automatically recognised with the help of modern, wearable sensors. The physiological reference value for the project is the first ventilatory threshold (VT1). This is the training intensity above which the body increasingly resorts to anaerobic energy supply. At present, VT1 can only be determined by analysing gas exchange.

PerDic-Life aims to estimate this threshold from signals that can be routinely provided by wearables, including electrocardiography (ECG), photoplethysmography (PPG) and acceleration data. By combining these signals with machine learning methods, the project aims to make a measurement that is currently limited to the laboratory continuously available in everyday life.

To this end, around 100 people aged 65 and over will be examined at two appointments in the gait laboratory of the Department of Geriatrics at the University of Oldenburg. At the first appointment, the individual VT1 is determined under controlled conditions using a treadmill protocol with gradually increasing speed and incline. A gas exchange analysis is carried out as the gold standard, while ECG and PPG recordings are made at the same time. During the second session, participants perform activities close to everyday life, such as climbing stairs, while wearing mobile measuring devices. In addition, the participants wear small measuring devices for about a week between the two visits, which record physiological signals and movements during their normal daily routine.

The data collected will be used in three increasingly realistic phases to develop and evaluate prediction models. Firstly, the data is analysed on the basis of information from individual sensors using existing laboratory data. In the second stage, multimodal models are used that combine ECG, PPG and acceleration data. In the third phase, personalised models are used that are adapted to demographic characteristics or individual participants. This gradual transition from controlled laboratory data to real-life conditions closes a central gap in previous research, in which physiological thresholds were determined almost exclusively in standardised exercise tests.

In the long term, the results of PerDic-Life could form the basis for a wearable warning system. This is intended to warn older people at risk of falling when their physical activity reaches an intensity that could be associated with a temporarily increased risk of falling.

The project combines expertise from three areas of the Department of Health Services Research at the University of Oldenburg: Geriatrics, AI4Health and Assistance Systems and Medical Technology.