A team of Google researchers developed an artificial intelligence system that predicts medical outcomes like mortality and readmissions based on patient data held in EHRs, Quartz reports.
The team's results were published in a research paper submitted to the scientific preprint website arXiv.org Jan. 24. The research paper has not been peer-reviewed.
Here are five things to know about the research project.
1. To construct a predictive model, researchers typically extract select variables from standardized data. However, to avoid what Google researchers referred to as a "labor-intensive process," they attempted to use deep learning, an advanced machine learning technique that doesn't require standardized data.
2. The researchers developed a deep learning approach to analyze a patient's raw EHR record. The deep learning system evaluated data from thousands of patients to determine words and events associated with certain outcomes, while also identifying which data could be ignored.
3. To validate the deep learning approach, researchers used de-identified EHR data from 216,221 adult patients who were hospitalized for 24-plus hours at UC San Francisco Medical Center and University of Chicago Medicine. This volume of EHR data encompassed 46,864,534,945 data points, including clinical notes.
4. The deep learning models achieved high accuracy for predicting in-hospital mortality, 30-day unplanned readmissions, prolonged length of stay and all of a patient's final discharge diagnoses, according to the research paper.
5. The study authors concluded their deep learning approach outperformed existing "state-of-the-art traditional predictive models."
"We believe that this approach can be used to create accurate and scalable predictions for a variety of clinical scenarios, complete with explanations that directly highlight evidence in the patient's chart," the study authors wrote.