The Potential for Reducing Medical Mistakes through Machine Learning in Health Care

Consideration has been given for the editing and publishing of this post

Over the course of the last decade, artificial intelligence, sometimes referenced as machine learning, has become a buzzword in technology companies of all shapes and sizes. The use of AI in businesses cases is purported to help create more efficient processes through data interpretation that does not require human interaction. Instead, an algorithm mines information looking for trends it can build upon and respond to with speed and accuracy when the time comes. Tech firms focused on developing machine learning software and tools are at the forefront of countless industries throughout the world, both in consumer and enterprise driven fields. Even healthcare has been folded into the artificial intelligence mix, but with cautious optimism.

One of the world’s largest technology powerhouses, Google, owns a subsidiary called DeepMind which is focused on delivering AI research to meet the needs of complex organisations and their business problems. DeepMind Health, or DMH, has recently been in the news because of its partnership with the Royal Free hospital trust of London and its intent to improve the interaction between patients and medical staff in certain instances. DeepMind Health is not explicitly an AI program in the work it is doing with the NHS, but the future remains open as to what can be done in the healthcare environment with the help of machine learning. However, the first step into building specific technology to benefit the patient population has been met with serious opposition, given the fact that the Royal Free Trust shared confidential patient data with DeepMind Health in the development of the mobile application, Streams.

Understanding AKI and DeepMind’s Streams Application

The development of the Streams app was intended to help with the early detection of acute kidney injury, or AKI. Throughout the UK, 40,000 deaths are connected to AKI each year, due mainly to the abrupt damage that takes place in the kidneys with the onset of the medical condition. When the organs do not function properly, whether that is a minor loss or complete failure, temporary or lifelong dialysis may be required to bring the patient back to health. AKI is not a simple condition to identify as it can be the direct cause of a myriad of issues, including major surgery, severe burns, sepsis, or heart and another organ failure. Certain patients also experience AKI due to interaction with medications or pre-existing conditions with the kidney or urinary tract.

Because AKI shows few warning signs or symptoms, clinicians, doctors, and nurses must perform a slew of laboratory tests to determine the issue is present. In some cases, lab results can take hours to come back with definitive results, leaving patients at great risk of developing further medical problems because of a delayed AKI diagnosis or treatment. The Streams app offers alerts based on a standardised algorithm to nurses and clinicians that helps improve patient outcomes through speed and accuracy. The app does not utilise machine learning in its current state, but DeepMind has shared that AI-driven alerts could be part of the future of apps like Streams in hospital and critical care settings.

The partnership between the Royal Free Trust and DeepMind Health in the creation of the Streams app is promising in the diagnosis and treatment of AKI, but there are concerns among several healthcare advocates surrounding the development of the tool. In 2016, it was found that the NHS delivered more than 1 million patient records to DeepMind Health in an effort to help design Streams and its functionality. Those records, which now sit idle on Google-owned servers, include identifiable patient information, and the individuals whose information was shared did not give prior consent. Issues like these in the realm of healthcare privacy and the distribution of sensitive information are causing some to pause about the realistic implications of technology in the healthcare field moving forward.

Broad Implications for Technology in Medicine

With Streams, healthcare professionals have a more efficient way to view data, receive intelligent alerts, take clinical takes, and manage tasks throughout the patient care experience. While DeepMind Health has strongly stated its mission to improve patient outcomes through these technology-infused tools, the issues involving the compromise of data from the Royal Free Trust paints a slightly different picture. A legal team of medical negligence specialists shared that the wealth of data provided by Streams could be used to monitor and measure organisational responses to serious medical issues like AKI, as well as in events where patient harm occurs because of a delayed response. However, the use of technology in healthcare cannot come at the price of the deterioration of fundamental privacy rights afforded to all patients.

As more technology firms and app developers embrace the power of machine learning in the healthcare sector, there are several questions doctors, policymakers, and patients must seek to answer so that data compromises and other issues do not arise. First, it is necessary for each party involved to have an understanding of who or what will control and have access to individual patient data. The ability of AI to sort through millions of data points in a short period of time is beneficial in creating alerts and predictions as to who may be at risk for serious medical conditions, like a heart attack or a stroke. There is some evidence that points to similar promise in determining high-risk individuals who could develop diabetes, schizophrenia, or even certain cancers. None of these advanced detections should override informed consent or privacy issues, however.

Similarly, understanding AI in terms of how it justifies one diagnosis or course of treatment over another is important. Having an explanation as to why a machine learning tool came to a specific conclusion is necessary for ensuring patient care is delivered at its highest possible level of quality. Finally, machine learning in healthcare must be understood regarding what it is truly optimised for. Better patient outcomes, greater organisation-wide performance, or reduction in mortality rates may all be viable endpoints when AI is implemented throughout a healthcare system. However, being able to measure its effectiveness while working to safeguard the information used to improve efficiencies is a necessary part of the process. Overall, the trend toward personalised medicine can be catapulted forward with the help of machine learning, but only when these considerations are taken into account from the start.

October 4, 2017 · Tim Kevan · Comments Closed
Posted in: Uncategorized