An explorative study on movement detection using wearable sensors in acute care hospital patients

Insufficient physical activity during hospitalization correlates with decreased physical functionality, prolonged stays, and higher readmission rates among the elderly population. Wearable systems provide an approach for monitoring patients' physical activity, data to set achievable goals and motivation for patients to stay active. However, elderly patients often present distinct gait patterns due to walking aids or co-morbidities, and most existing monitoring solutions are trained on data from healthy individuals. Therefore, the main study goal was to develop a wearable based algorithm prototype for three wearing locations (ankle, thigh, wrist) and assess its comparative classification accuracy to determine the optimal location for classifying patient activities during hospitalization. We collected raw accelerometer and gyroscope data from three different body locations (wrist, ankle, and thigh) from 40 patients at the University Hospital Basel. Depending on the patient's mobility status, the protocol comprised up to six activities, including lying, sitting, standing, sit-to-stand, walking, and climbing stairs. We trained two classification models for each location; one based on accelerometer and gyroscope input and the other on accelerometer only. In addition, we assessed the patient experience by questionnaire. The ankle model performs best with an accuracy of 84.6% (accelerometer and gyroscope) and 82.6% (accelerometer). The wrist and thigh models show accuracy results in the 72.4-76.8% range. The patient questionnaire evaluation reveals a high acceptance of 97.7% towards carrying a monitoring device for 8 h throughout the day, regardless of the wearing location. Patients reported the ankle as the least disturbing location in 87.2% cases. Our study showed that the accuracy of the model is clearly dependent on the individual location of the sensor, with the ankle showing the highest weighted accuracy results of the three body sites. In addition, patients reported a high acceptance towards a sensor-based classification system, underlining the feasibility in a clinical setting.Trial registration: The study was approved by the Ethics Committee of Northwest and Central Switzerland (BASEC 202202035) and has been registered at clinicaltrials.gov (NCT06403826).