Daily caregivers in long-term care centers need to measure the vital signs of residents and find out the residents with abnormal values based on personal physiological information and history records of each resident every day. Jubo.health aimed to develop an artificial intelligence model to find out these values automatically.
We used residents' background information, such as sex, BMI, age, and medical history and so on, as features and nursing notes as labels to train tree-based machine learning models. Ultimately, we used LightGbm for residents with different BMI, age, vital sign status (standard deviation, median) to give different vital signs abnormal criteria. The model Recall reaches 0.81 which is a 14% improvement over existing academic values and significantly reduced false positives.
Aug 2019 - Dec 2019
A high standard deviation of systolic blood pressure indicates that the blood pressure of residents is prone to instability (a characteristic of a certain disease), and statistics change over time to take into account external environmental, emotional, and medication
Observation：Anomalies in single vital signs are less likely to be treated by the model as abnormal
Presumption：Abnormal residents' physiology will be reflected in more than one vital sign value
Implement AI model on
Jubo.health smart nursing platform
We used LightGbm model to determine whether this vital sign is abnormal. Then, the important features are observed with the SHAP analysis model and are highlighted. The page would show the changes in the value of each vital sign in the past week. The users can track the important vital sign items by clicking "Track" button.