A model for predicting fall risks of hospitalized elderly in Taiwan-A machine learning approach based on both electronic health records and comprehensive geriatric assessment

Abstract

BackgroundsFalls are currently one of the important safety issues of elderly inpatients. Falls can lead to their injury, reduced mobility and comorbidity. In hospitals, it may cause medical disputes and staff guilty feelings and anxiety. We aimed to predict fall risks among hospitalized elderly patients using an approach of artificial intelligence.Materials and methodsOur working hypothesis was that if hospitalized elderly patients have multiple risk factors, their incidence of falls is higher. Artificial intelligence was then used to predict the incidence of falls of these patients. We enrolled those elderly patients aged >65 years old and were admitted to the geriatric ward during 2018 and 2019, at a single medical center in central Taiwan. We collected 21 physiological and clinical data of these patients from their electronic health records (EHR) with their comprehensive geriatric assessment (CGA). Data included demographic information, vital signs, visual ability, hearing ability, previous medication, and activity of daily living. We separated data from a total of 1,101 patients into 3 datasets: (a) training dataset, (b) testing dataset and (c) validation dataset. To predict incidence of falls, we applied 6 models: (a) Deep neural network (DNN), (b) machine learning algorithm extreme Gradient Boosting (XGBoost), (c) Light Gradient Boosting Machine (LightGBM), (d) Random Forest, (e) Stochastic Gradient Descent (SGD) and (f) logistic regression.ResultsFrom modeling data of 1,101 elderly patients, we found that machine learning algorithm XGBoost, LightGBM, Random forest, SGD and logistic regression were successfully trained. Finally, machine learning algorithm XGBoost achieved 73.2% accuracy.ConclusionThis is the first machine-learning based study using both EHR and CGA to predict fall risks of elderly. Multiple risk factors of falls in hospitalized elderly patients can be put into a machine learning model to predict future falls for early planned actions. Future studies should be focused on the model fitting and accuracy of data analysis.

Digital health technology combining wearable gait sensors and machine learning improve the accuracy in prediction of frailty

“…However, the FP test only evaluates the physical aspects of frailty, while frailty is acknowledged to be a multifaceted state that includes not only physical dimensions, but also social, cognitive, and psychological dimensions (5,6). Gaitrelated mobility is a key physical ability that has been linked to frailty and is often used as a predictor of future health outcomes (7,8).…”

A model for predicting physical function upon discharge of hospitalized older adults in Taiwan—a machine learning approach based on both electronic health records and comprehensive geriatric assessment

“…CGA has also been used in machine learning to better evaluate older patients with atrial fibrillation (30). Our previous work has also showed that CGA combined with EHR can predict fall risk among the older adults (16). Future studies are still warranted for both identification and intervention in the promotion of physical function during hospitalization after any machine learning prediction.…”

“…With advancing technology and improved medical informatics, some researchers have predicted adverse outcomes in hospitalized patients based upon electronic health records (EHRs), however data pulled from EHRs also have some limitations (13,14). Therefore, many scientists now use a machine learning model to predict adverse outcomes in older adults (15,16).…”

Predicting the risk of nodular thyroid disease in coal miners based on different machine learning models

“…Compared to the other four machine learning models, the XGBoost model has the following advantages: it has no explicit linearity requirement for the data distribution and automatically detects non-linear relationships and interaction effects using the relevant factors; it can make full use of missing data without filling in the data and can more realistically reflect the original results expressed by the data; the XGBoost model has the advantage of handling large samples of data (33), learning and remembering as it analyses and processes the data and improving its predictive power; the model trained with a large amount of data is more reliable than a machine learning model fitted to a small sample of individual tests (34). The XGBoost model has the advantage of being able to learn and remember as the data is analyzed and processed, and to improve its predictive power (35); a model trained on a large amount of data is more reliable than a machine learning model fitted to a small sample of individual tests, and has a high clinical application in predicting classification outcomes (36).…”