Machine learning in medicine was tested against a statistical model to see if it was more accurate in determining which patients need to be admitted to hospital.
Emergency hospital admissions are increasing and stretching what is often a system that is already at its limits. A large proportion of visits to the emergency room are not medically necessary. Policy makers have been looking into ways to help take the load off the emergency room by preventing the unnecessary visits from being admitted to hospital, and allowing medical professionals to provide patient care to those who really need it.
To identify which patients need emergency care, researchers have aimed to determine a list of risk factors. The list of risk factors would ultimately replace a hands-on examination by a doctor or nurse in determining whether a certain patient needs emergency care or not. Previously, statistical models where data is entered into a system have been used to determine the risk factors. However, these systems have been found to be inaccurate partly because there are so many different factors at stake and the models have no flexibility to their rules. To overcome these problems, researchers have started to look at machine learning in medicine.
Machine learning in medicine is a form of artificial intelligence. In much the same way as statistical models, data is entered into a system. However, in machine learning, the system can progressively improve its performance with the more data it receives. Researchers from Oxford, UK, used a big data set from 4.6 million patients on two types of machine learning models and compared it with a previously used statistical model, to see which was better at predicting which patients would end up in hospital. Their results were published in PLOS Medicine.
The machine learning methods outperformed the statistical model for predicting which patients would need hospital admission. The researchers expect that the information gleaned from this study could help guide policy and practice to reduce the burden currently placed on the emergency room. Ultimately they hope that physicians will be able to accurately monitor the risk score of a patient and take necessary actions in time to prevent an emergency admission.
Written by Nicola Cribb, VetMB DVSc Dip.ACVS
Reference: Rahimian F, Salimi-Khorshidi G, Payberah A, et al. Predicting the risk of emergency admission with machine learning: Development and validation using linked electronic health records. Plos Med. 2018;15(11):e1002695. doi:10.1371/journal.pmed.1002695.