DeepMind is a Google-owned AI research firm in the U.K. The Company has just mentioned its deep learning model in the journal, Nature, which predicts the possibility of a patient getting a life-threatening disease, like acute kidney injury (AKI). The Company also states that its model is able to predict whether a person will develop AKI, with 48 hours in advance.
DeepMind mentioned this research in a blog post. The Company stated that it is quite proud of making artificial intelligence, which can predict one of the main causes of patient harm, two days before it actually happens. The Company further states, “To date, this has been our company’s biggest research breakthrough. It doesn’t only show the ability to stop the deterioration effectively, but also predicts it two days before it happens.”
However, just the surface reading of the paper raises some limitations because the data used to train the model skews, was mainly male, above 90%. However, the reason behind it is that the researchers gave DeepMind’s AI data from the U.S. Department of Veteran Affairs (VA). The research paper mentions that females included only 6.4% of the training data. The paper also mentions that the overall model performance was somewhat lower for the females.
The data statistics from the paper indicate that around 19% of the patients were black. However, it doesn’t mention the proportion of black women. Simple logic tells us that they must be lower than 6.4%. The paper also doesn’t mention any other ethnicities.
The press asked the Company about the model’s ethnicities and the overall performance. The DeepMind spokesperson told the press, “In women, it predicted around 44% of all AKI earlier and 56% in men. The overall model performance was quite high in African American patients. It was around 60.4% of AKI’s detected early, compared to the 54% of all other ethnicities combined.”
“However, this research is just the beginning,” the DeepMind spokesperson confirmed, “We need more future research from the general population in order to reach a definite conclusion. We will also work with the VA to validate the model from prospective and retrospective studies. We hope to undergo a prospective interventional study that understands how the prediction influences the result in a clinical setting.”
The Spokesperson adds, “We need the right kind of data to do this. The VA makes use of the same EHR (electronic health record) systems in different sites and hospitals. This means that the data is quite well-structured and clean.” Therefore, DeepMind’s research paper underlies the relationships between the AI outputs and the training inputs.
DeepMind also mentions in the blog post that their app, Streams, improved the overall quality care for the patients by preventing missed cases and speeding up the overall detecting process. The blog post also claims that the clinicians responded to urgent AKI cases in less than 15 minutes. The existing system may take hours to complete this procedure.
It also leads to an overall reduction in the care cost to the NHS. The cost reduces from $14,300 to $11,800 for hospital admission of a patient with AKI. However, under the current NHS contract, DeepMind gives the Streams service free. Therefore, the cost reduction claims come with major caveats.
In a healthcare setting, the instructive output can make a difference between life and death. Therefore, the important thing isn’t the technology here. The main key is access to the representative data sets. Therefore, there is a big opportunity for countries with taxpayer-funded public healthcare systems to unlock value in medical data of their populations.
Oxford University’s Sir John Bell concluded this data discussion in comments to the Guardian Newspaper. He said, “Most of the value lies in the data. However, the worst thing is that we give the data away for free.”
In crux, the data provided by DeepMind along with the introduction of the app, is a big breakthrough in the medical field, just like Google’s SMILY. Let us hope that DeepMind’s further researches are successful so that patients get better healthcare.
