Google AI Can Now Predict Death with 95% Accuracy

On: June 19, 2018

Google is leveraging the powers of Artificial Intelligence into domains other than its web services since a few years now. This time, the search giant is making use of its AI prowess for healthcare. It has now developed an artificial intelligence system that can predict when a patient will die. Moreover, the accuracy of the initial trials is at a staggering 95%.

Artificial Intelligence systems work in a way where they take up a lot of data points and use it to reach a desirable outcome. In this case, Google feeds the system with de-identified data of 216,221 adults in total. This data was procured from two US medical centers over a period of time.

The data included basic information like age, ethnicity and gender. Apart from this, the key piece of information included doctor notes in the form of graphs, charts or PDF’s. The latter data is hard to get for machines. This is further combined with data from the hospital that includes previous diagnosis, medications, lab results and much more. The combination of all this is what probably makes Google’s prediction data so accurate.

In total, the machine processes around 45 billion data points. According to Stanford professor Nigam Shah, “Around 80 percent of development time spent on predictive models goes on making the data look presentable for the AI”. But with Google’s systems, researchers can feed any type of data and it can process it with ease to reach desirable outcomes.

The predictions are for whether a patient will die after 24 hours of getting admitted to the hospital. Apart from predicting a patient’s chances of living or dying, it can also do a bunch of other things like figuring out the number of days a patient will be in the hospital or the chances of the patient being readmitted.

Google’s predictions on all of these fronts are quite impressive. Below are its scores in comparison with that of the hospital staff’s prediction.

  • Predicting death: 95% accuracy (Google) vs. 86% accuracy (Traditional methods)
  • Predicting patient’s length of stay in a hospital: 86% accuracy (Google) vs. 76% accuracy (Traditional methods)
  • Predicting readmission of a patient after being discharged: 77% accuracy (Google) vs. 70% accuracy (Traditional methods)

It must also be kept in mind that the technology used by Google is in a very nascent stage right now. As it gets better, like AI machines are meant to, the results shall be even better. It will be interesting to see how Google helps the healthcare industry with its innovations.



Aditya Mohanty

Previously co-founded a startup called WordKrowd, an online literature platform, which explains his love for writing and reading. Thinks about startups and technologies all day long, thus got himself to Pricebaba for a stint in writing. Stories, both fictional and real excite him. Other close recipients of love from him include good movies and good food.