AI-based patient monitoring for fall prevention in stroke patients: a pilot study at a Malaysian acute stroke unit

Falls are an important patient safety concern and stroke patient are at high risk. Artificial intelligence (AI) could be leveraged to reduce patient falls in the hospital but there is scarcity of data. Therefore, the aim of this study is to evaluate the effectiveness of the SMART AI Patient Sitter system-an AI-powered motion-sensing and alert system designed for fall detection and prevention in a real-world hospital setting.

Conducted from January to December 2024 at the Acute Stroke Unit of Hospital Seberang Jaya (ASUHSJ), the study involved 30 stroke patients who consented to AI monitoring. The SMART AI patient sitter system comprised an optical sensor, alert panel, and control panel monitored by AI, which detected patient movement and triggered alerts to the observation counter. Blurred, non-identifiable images maintained patient privacy, and investigators were identified through uniform recognition. Data on mobility and fall events were recorded continuously.

The integration of this system led to an 83.33% reduction in fall incidents and the generation of 1,439 alerts with a 95.34% accuracy rate. Enrolled patients had a mean age of 61 years(SD ± 12.8) years; 63.3% were male; 56.7% were of Malay ethnicity and 83.3% were classified as high fall risk. The median duration of monitoring was 3 days (IQR: 1.0-6.0), with a median of 19 bed exits(IQR: 1.0-85.0) bed exits. The first bed exit attempt occurred at a median of 150 minutes (IQR: 20.0-2103.0) minutes post-admission. Response time to movement alerts was prompt, with a median of 21  seconds (IQR: 4.0-75.0). Only one fall (3.3%) was recorded during the study. The incident involved a moderate-risk patient who attempted to stand abruptly. Staff responded within 29 seconds, and the patient recovered without severe injury.

These findings suggest the system's potential in early detection and timely intervention. Study data demonstrated wide variability in patient mobility patterns, highlighting the need for individualized monitoring. The SMART AI patient sitter system's ability to deliver real-time alerts, ensure patient privacy, and reduce fall incidence demonstrates its value in improving stroke patient safety. Overall, this study supports the integration of AI-based monitoring tools in clinical settings to enhance patient care and reduce preventable incidents like falls.