Google tool can predict when you'll die

Google is getting into healthcare and could determine the likely fate of patients. Photo / 123RF

Google is getting into healthcare and is touting the power of its artificial intelligence systems to determine the likely fate of patients.

With the enough data, the company thinks it can predict when a patient will die with up to 95 per cent accuracy.

In May, Google scientists published the account of a woman who came to hospital with late stage breast cancer and fluid building in her lungs. After the hospital equipment and computers took the woman's vital signs, it estimated that she had a 9.3 per cent chance of dying during her stay at the hospital.

Then it was Google's turn. Its neural network, a type of artificial intelligence that can analyse huge reams of data and automatically learn and improve, was fed 175,639 data points on the woman including past health records and her current vital signs. It delivered a decidedly more grim prediction, calculating the woman had a 19.9 per cent chance of dying during her stay.

She reportedly passed away a couple days later.

That heartbreaking anecdote was reported by Bloomberg this week after it was published in May by Google-backed researchers in the scientific journal Nature.

The initial findings of the proof-of-concept study suggest that Google's system is faster, and more accurate than other techniques at evaluating a patient's medical history and forecasting a range of patient outcomes such as mortality, hospital re-admission, prolonged hospital stay and discharge diagnosis.

"We were interested in understanding whether deep learning could produce valid predictions across a wide range of clinical problems and outcomes," the researchers wrote.

"These models outperformed traditional, clinically used predictive models in all cases. We believe that this approach can be used to create accurate and scalable predictions for a variety of clinical scenarios."

Where the system was able to add value was its ability to incorporate data not easily analysed by traditional systems such as clinical notes buried in PDFs or scribbled on old charts, while disregarding redundant data.

In total, Google has analysed 216,221 hospitalisations and 114,003 patients, which comes to more than 46 billion data points.

The proof-of-concept study found that the data-crunching algorithm could accurately predict risk of mortality, hospital re-admission, prolonged hospital stay and discharge diagnosis. As hospitals work to improve electronic health records of patients, Google's tools could be hugely useful.

According to the research, the algorithm was 95 per cent accurate at predicting patient mortality based on data from the University of California San Francisco (UCSF) health system and 93 per cent accurate using data from the University of Chicago Medicine system.

Google, or more accurately its parent company Alphabet, is arguably the world leader in developing systems of artificial intelligence. The emerging technology is tipped to reshape our world in unimaginable ways.

It's application to healthcare is already well and truly underway with self learning algorithms being used to detect things like skin cancer with greater accuracy than humans could hope to achieve.

But the data that underpins the neural network described in the journal paper does raise some questions if this technology is to be adopted. Speaking to Fox News, Dr Mikhail Varshavski raised some of the concerns around privacy and the quality of data.

"The thing that is worrying for me is what happens with this data and who owns this data?" he said. "I hope, as a doctor, that these companies use the data to benefit the patients, not the companies themselves."

Dr Varshavski also pointed out the huge risk in relying on predictions or assessments if the data used by the AI computer isn't as reliable as we think.

"Machines make mistakes and sometimes they make mistakes based on faulty data," Dr Varshavski added. "There needs to be oversight of what these things do."