Artificial intelligence’s greatest potential for enriching medical knowledge and improving global healthcare systems arises from its ability to find useful insights inside big data sets. With access to de-personalized electronic health records (EHRs) of patients worldwide, deep learning can aid diagnosis, care and treatment strategies, and patient prognosis. The previous buzzword-packed sentences come down to this: as we gain the ability to gather, organize, and analyze ever-greater masses of relevant information, medicine will rely less on guesswork and more on accurate identification and prediction. We’ve already noted early returns on deep learning in fighting aging with bioinformatics, predicting heart disease from retina images, and detecting arrhythmia with an Apple Watch.
Google researchers published in npj Digital Medicine the results of a study in which deep learning outperformed traditional statistical methods for predicting hospital patient clinical outcomes, including death. Working with the University of California, San Francisco, University of Chicago Medicine, and Stanford University, the group applied deep learning algorithms to data from more than 200,000 adult patients hospitalized in two academic medical centers. In preparation, the Google staff developed technology to transform EHR files into a standard format. The Google algorithm accurately predicted death, readmission, prolonged hospital stay, and discharge diagnosis with greater accuracy than any previously published model.
Predictions such as these can be powerful tools for improving outcomes and lowering hospital costs. It’s still early days for deep learning and other forms of artificial intelligence in medicine, but the horizon for wider applications is approaching rapidly.
