September 13-19 2021
Using artificial intelligence to predict the condition of patients
The aim of artificial intelligence (AI) is that machines are able to imitate real intelligence. Artificial neural networks are made up of computer servers capable of processing complex calculations from large databases. AI and the resulting processed data are dynamic, since the machine is constantly learning thanks to the algorithms developed by researchers. AI holds promise for different kinds of application in the domain of health, such as forecasting the condition of patients with COVID-19. Its advantages when used for the treatment of medical data are speed, the possibility of sharing data between medical centres anonymously, and traceability. American researchers at Harvard and Santa Clara have developed a machine learning model that can help to predict the outcome for patients hospitalised with COVID-19.
The hypothesis of the researchers was that a machine learning model dealing with large data bases from several medical centres would give better results than when dealing with data from a single centre. They used only clinical data (vital signs, laboratory results, demographic data such as the age of the patients, thorax radiography) from 20 medical centres across the world without taking into consideration doctor evaluations, such as clinical impressions or observed symptoms. This model, called EXAM, provided a score to predict oxygen needs at 24 hours and 72 hours after hospitalisation. The data, concerning more than 16 000 individuals, was used to train the machines, and the researchers compared the results of their model with the results of each centre. They were able to show that the EXAM model enables a significant increase in prediction performance for patient oxygen needs at 24 hours and 72 hours.
In conclusion, this machine model proved to be successful. It enables large data bases from numerous medical centres to be used. Artificial intelligence can help the doctor, for example, to best orientate the patient from their arrival in the emergency unit, by forecasting their condition from the results of examinations.