A study has found that AI can be applied to routinely collected electronic health record (EHR) data to predict mortality of patients so that palliative care can be administered at the right time.

The study was conducted by six Stanford University scientists, including Andrew Ng, Director of the Stanford Artificial Intelligence Lab, Nigam Shah, Associate Professor of Medicine (Biomedical Informatics Research) and of Biomedical Data Science, and Stephanie Harman, Clinical Associate Professor, Medicine.

They developed a Deep Neural Network (DNN) model to analyse data from two million EHRs of adults and children treated for advanced illness at the Stanford Hospital or the Lucile Packard Children’s Hospital.

Data about each patient’s virtual past would be used to generate a prediction score for their survival 3-12 months into the future. The prediction date had to be at least 3 months prior to the patient’s death because of the time required to prepare and start palliative care.

Upon review of 50 randomly selected patients from the top 0.9 precision bracket of the test set, the research team found that all of them were appropriate for a referral to palliative care on their prediction date.

Even the false positives, who did not die within 12 months of their prediction dates, were often diagnosed with terminal illnesses or complicated medical needs.

Shah told CNBC, “Right now we miss most of [the patients who should receive palliative care], because clinicians over estimate survival…Less than one percent of individuals who die are offered palliative care more than six months before their passing,

“So, at this juncture, even if the AI assisted approach misses half the cases, we’d still be way better off than the current state of affairs.”

The model developed for the study is being piloted for daily, proactive outreach to newly admitted patients.