Is it Possible to Predict Risk of Death through Google's AI Network?

Is it Possible to Predict Risk of Death through Google's AI Network?

Healthcare Tech Outlook | Friday, November 30, 2018

With Google's newly designed artificial intelligence (AI) network, predicting the risk of death may soon be possible with an accuracy of more than 90 percent. The multinational tech giant is exploiting the capabilities of machine learning (ML) and raw electronic health record (EHR) or electronic medical record (EMR) data. Clinicians are analyzing these ML and EHR/EMR data to envisage the course of disease and risk of death during the stays of patients in the hospitals.

The researchers are training deep learning models at the Universities of California, San Francisco, and Chicago. The training is focused on the risk of mortality rates, readmission, prolonged stays, and discharge diagnoses of patients. Based on ICD-9 code standards, the models are fed with algorithms incorporating the entire EMR, clinical notes, and volumes of other data for making the predictions easier.

The collected information is then processed and integrated into the fast healthcare interoperability resources (FHIR). An EMR data structure, FHIR is more flexible than the conventional processes of analyzing EMR data. The Google AI network produced an accuracy rate of 95 percent in anticipating the risk of death in patients and is hindered by a lesser number of false alerts. Using FHIR efficiently transfers the data to new facilities as compared to the traditional predictive models, which only focuses on organizing the data.  The outcomes realized from conventional tools and systems go beyond discharge diagnosis, length of stay, mortality rates, and readmission of patients. 

In addition, the AI network notifies clinicians of the sources from where data is extracted, including patients' medical record, or radiology findings decreasing reliance on neural networks for diagnosis purposes. Google's endeavor has empowered researchers and made them more confident in making correct predictions. The network ultimately reduces the healthcare costs.

Going forward, more researches and potential trials are needed in order to demonstrate the scalability of the healthcare sector.

Weekly Brief