University of Oxford researchers were able to predict a patient’s risk of being admitted into emergency care by using machine-learning techniques with electronic health records (EHRs), according to a study published in PLOS Medicine.
“Our findings show that with large datasets which contain rich information about individuals, machine learning models outperform one of the best conventional statistical models,” lead researcher Fatemeh Rahimian, a former data scientist at the George Institute for Global Health, said in a statement. “We think this is because machine learning models automatically capture and ‘learn’ from interactions between the data that we were not previously aware of.”
For the study, researchers derived, validated and compared how well a conventional model and two machine-learning models could predict emergency hospital admission. To test the models, they used data from the EHRs of 4.6 million patients from nearly 400 practices in England between 1985 and 2015. They also used time and various variables, like patient demographics, lifestyle factors, lab tests, prescribed medications and previous emergency admissions.
When testing all of the models on just the variables, the conventional model achieved an area under the receiver operating characteristic curve (AUC) of 0.736, while the two machine-learning models achieved AUCs of 0.796 and 0.736. When adding additional factors in, like time and more variables, the conventional model achieved an AUC of 0.788, while the machine-learning models achieved AUCs of 0.826 and 0.810.
Though the models need further testing, researchers believe machine learning could be used to help physicians better monitor patients in order to avoid emergency visits.
“The use of machine learning and addition of temporal information led to substantially improved discrimination and calibration for predicting the risk of emergency admission. Model performance remained stable across a range of prediction time windows and when externally validated,” the study concluded. “These findings support the potential of incorporating machine learning models into electronic health records to inform care and service planning.”