BACKGROUND
The proportion of people aged 60 years and older in the population is increasing, representing a significant challenge. Due to their frailty, there is a higher frequency of unplanned hospitalizations in this population, leading to adverse events. Assisted living facilities can provide improved support for aging adults. Digital tools based on artificial intelligence can help to identify early signs of vulnerability and improve the quality of life.
OBJECTIVE
This study aims to identify the performance of a system that provides alerts when a machine learning algorithm predicts a short-term risk for emergency hospitalization, as well as to explore health treatments offered in response to these alerts and users’ experience.
METHODS
An uncontrolled multicenter trial was conducted between March 2022 and August 2022, on older adults residing in 7 assisted living facilities in France. An eHealth system was set up to alert in case of high risk of emergency hospitalization. Nurse assistants (NA) of the assisted living facilities used a smartphone application to complete a questionnaire on the functional status of the patients, analyzed in real time by a previously designed machine learning algorithm. This eHealth system notified a coordinating nurse or a coordinating NA who subsequently informed the patient's nurses or physician. The primary outcomes were the acceptability and feasibility of the device implementation in the context and to confirm the effectiveness and efficiency of artificial intelligence in risk prevention and detection in practical, real-life scenarios. The secondary outcome was the hospitalization rate after alert-triggered interventions.
RESULTS
In this study, 118 out of 194 eligible patients (61%) were included and had at least one follow-up. A total of 38 emergency hospitalizations were documented, among an average of 78 (66.1%) patients. The system has generated 92 alerts, for 47 (40%) patients. Out of these alerts, 46 (50%) led to 46 healthcare interventions for 14 (12%) patients and have resulted in 4 hospitalizations. While the other 46 (50%) alerts that did not trigger a healthcare intervention resulted in 25 hospitalizations for 64 patients which represent 86% of hospitalizations (P<.001). Almost all hospitalizations were due to a lack of alert-triggered interventions (P<.001). System performance was very good as specificity was 96% and True Negative Rate was 99.4%.
CONCLUSIONS
CE Marked PRESAGE CARE system has been implemented with success in assisted living facilities. It was well accepted by coordinating nurses, predicted 76% of emergency hospitalizations with a very good true negative rate of 99.4%. This system has shown good results in terms of performance and clinical impact in this context, nevertheless, more work is needed to understand the moderate usage rate of NA and improve it.
CLINICALTRIAL
clinicaltrials.gov Identifier: NCT05221697.
The research protocol was approved by ANSM (The French Agency for the Safety of Health Products):
ID RCB: 2021-A02131-40–CPP 1-21-072.