One study claims that an algorithm developed by Google in the United States can predict the mortality rate of patients after hospitalization, and the accuracy even exceeds the doctor’s prediction.
Using neural network artificial intelligence technology, Google’s “magic brain” team developed an algorithm that analyzes more than 170,000 data points for a medical record of a metastatic breast cancer woman and predicts that her pre-discharge mortality rate is 20%. The hospital assessed the patient’s mortality rate as 9%. As a result, the patient died in the hospital within two weeks.
Google’s algorithm can also predict the length of hospital stay and whether it will be re-admitted to hospital within 30 days after discharge. When doctors make such predictions, they usually rely on limited medical records. Google algorithm can account for almost all information obtained from the hospital, including extensive medical records and blood test reports.
Analysing the data of 216,000 hospitalizations in US hospitals, this technique predicts 95% accuracy of in-hospital mortality, compared with 86% predicted by traditional methods. Google technology predicts long-term hospitalization accuracy rate is 86%, the traditional prediction accuracy rate is 76; as for the accuracy rate of unexpected re-hospitalization treatment prediction, the two were 77% and 70%.
The researchers said that this neural network system, which is close to the human brain, can “understand” patient medical records without special training and can “understand the key factors and interrelationships of the data itself.” Google researchers said that the next study will introduce this technology into the clinic. However, some experts believe that this will be a more sensitive area because it involves patient privacy protection.