The score is designed to help medical staff identify how serious the condition of a patient is, and has been shown to improve patients' safety and outcomes, especially for deteriorating patients.
Although it's a coarse evaluation, it helps the hospital predict how well a patient might do, and whether the odds are in their favour – essentially how likely they are to leave the hospital alive at the end of their treatment.
The challenge is that hospitals are busy places and patients are likely to be seen by several different medical professionals during their stay. Each of these people may record their findings on the patient in different places, varying from handwritten notes on the chart at the end of their bed to an online database filled with stats, vitals and medication history.
This scattering of patient information makes it difficult for one person to analyse a broad set of data about a patient at a glance.
Artificial intelligence systems have recently been trained to read handwritten documents (yes, even doctors' handwriting) and convert them to text using natural language processing.
A group of researchers at Google's Medical Brain Team used this capability to help collect huge amounts of data on individual patients that had been admitted to hospital.
This document reading system was then combined with a machine learning algorithm to interpret the patients' history quickly and easily.
Analysing 216,221 hospitalisations and 114,003 patients, and using more than 46 billion data points across two hospitals, the study published in the journal npj Digital Medicine showed that computers were significantly better than doctors at predicting patient outcomes.
In the first hospital, the doctors' early warning score was 85 per cent accurate, whereas the computer algorithm was 95 per cent accurate. In the second hospital, the doctors were 86 per cent accurate compared to the computer's 93 per cent.
This increased accuracy was due to the computer being able to collect and sift through a much larger amount of data about the medical history of the patient - including old notes that had been scribbled down on medical charts, and detailed medication history.
As awful as predicting death in patients may sound, the algorithm was also able to predict other factors, including the likelihood that a patient would be re-admitted, the likely length of their stay and their diagnosis on discharge.
This data could be used to help doctors plan patient outcomes more accurately, but also help hospital administration to better predict hospital bed occupancy and facility resources likely to be needed.
The study also opens up the controversial question as to whether or not patients should be allowed to know their death prediction score, and whether or not these highly accurate systems could be used in cost-cutting regimes to determine which patients are given more care to improve success rates.
With all these new decisions to think about, perhaps we should just crack open a fortune cookie to help us decide.
Dr Michelle Dickinson, creator of Nanogirl, is a nanotechnologist who is passionate about getting Kiwis hooked on science and engineering. Tweet her your science questions @medickinson
