Google has been working on training its AI to predict the risk of death among hospital patients and preliminary results show that the AI is “slightly” better than the hospitals. According to a report by Bloomberg, the company published the findings of its “Medical Brain” team to discuss more how AI was able to use the data from the notes in PDFs to the information scribbled on old charts to come at a conclusion. The interesting part of this system is that the AI also “showed which records led” the AI to reach a particular conclusion making it easier for the results to be checked.
According to the company, its newly trained models “outperformed traditional” predictive models that have been used by Clinics in all the bases. It went on to add that the company and researchers can be able to make more accurate prediction models that can be scaled according to a number of “clinical scenarios”. The company also talked about one of the primary case studies as part of this process.
Watch: Google Go: All you need to know
As part of the case study, the AI used 175,639 data points from the electronic medical records in addition to any handwritten notes to give a patient, a 19.9 percent chance of dying in the hospital in comparison to the 9.3 percent change of the same according to the computers in the hospital. The patient in question “passed away in a matter of days”.
According to a report by TheNextWeb, the company analyzed 2,16,221 hospitalizations of the 1,14,003 patients and the total 46 billion data points from their health records. The report goes on to add that this is not the first time when Google has used AI to apply “predictive healthcare” as its subsidiary, DeepMind teamed up with Department of Veterans Affairs to add 7,00,000 medical records to the AI in an effort to “predict deadly changes in patient condition”.
Google is also working on a voice recognition system for clinical notes so that doctors will no longer need to type the records as most of the problems arise from “the smallest mistakes” in the patient records. The report also pointed out that 80 percent of the time is spent in making the data presentable when it comes to training predictive models. Apart from the mere amount of time that it takes to use the data, the amount of time in paperwork is more than the amount of time spent in patient care. It was also noted that current systems for mining health data are time-consuming, costly, and cumbersome.
The company is working on moving the newly made predictive system to the clinics to ensure that it has an increased number of sources to get the health data from. The team working behind this is both excited as well as cautious. The reason for excitement is because it is the first time when the company has figured out a commercially viable application for AI and predictive technology and the reason for caution is because the company is no stranger to controversy of accessing patient data without their consent.