Google claims that a new deep learning model designed by it and its UC San Francisco, Stanford Medicine, and The University of Chicago Medicine colleagues has predicted the 'inpatient mortality' with an accuracy of 95 percent. Machine learning, which was previously applied to actions like traffic predictions, translations and like has, in a recent attempt, been used for healthcare by a Google team. The clinical predictions by the computer system came out to be astonishingly accurate, as they predicted if the patient will stay long in the hospital with an 86 percent accuracy and further unexpected readmissions with a 77 percent accuracy, marking 'statistical significance' in the area.
In addition to making the predictions, the deep learning models were also used to recognise the patient's condition. Google gives an instance for this: "if a doctor prescribed ceftriaxone and doxycycline for a patient with an elevated temperature, fever and cough, the model could identify these as signals that the patient was being treated for pneumonia." In a blog, Google calls its program a "good listener" for its capability to gather the patient's data including their ongoing treatments and notes.
The new program aims to eliminate the discrepancies caused by different Electronic Health Records (EHR) found in individually customised EHR systems of hospitals. In essence, the patient data differs from one hospital to another. To solve this, the deep learning mechanism reads all the data points from the patient's EHRs and then decides which data can be used to predict the outcome. The program also identifies the specific data set which it used the make the prediction.
Google protected the data used for this observation with security measures including "logical separation, strict access controls, and encryption of data at rest and in transit".
As of now, Google says that the entire idea is still in its early age and that the test simply suggests how machine learning can be used to improve healthcare.