We’re living in a world full of data today. Everything we do is, or can be, captured digitally, which means that each of us are potentially generating terabytes of data every single day. The natural question is how can we use this data to make our lives better, especially in healthcare, where data includes electronic health records, doctor’s notes, pharmacy prescriptions, insurance claims, genetic information and more.
In fact, we are now seeing a paradigm shift where medicine is fast becoming a data science supported by clinicians, rather than a clinical science supported by data. Unlocking the hidden value in healthcare data has the potential to greatly improve patient care and costs. We can do this through integrating and connecting with data in an effective way, and leveraging on artificial intelligence (AI).
AI, or machine learning, in healthcare is the use of complex algorithms to analyze large volumes of complicated medical data and discover patterns and correlations within the data that would not have been revealed otherwise. This has become feasible in the last few years as more and more data is now stored on the cloud which makes them accessible, as well as the availability of cheap computing power to analyze the data.
There are many ways AI can benefit healthcare organizations and stakeholders, in clinical decision support, predictive analytics and research. AI can allow clinicians to make better and more accurate diagnoses for patients. For example, in radiology, there are now algorithms that can help radiologists uncover brain, lung, liver and cardiovascular disease in CT scans in an automated fashion, some of which may otherwise be missed by the human eye. At Holmusk, we are using AI on large longitudinal mental health datasets to predict the response of treatments for individual patients with major depressive disorder and identify patients who are likely to have treatment-resistant depression. When identified early, such patients can benefit from a tailored treatment instead of having to experience years of failed therapy.
AI can improve efficiency for healthcare providers by automating repetitive tasks and making them more convenient for the patient and doctor. In 2018, the US Food and Drug Administration (FDA) approved the first AI medical device to detect diabetic retinopathy, which is the most common cause of vision loss among diabetic patients. This device is able autonomously to detect disease on retinal images, with a sensitivity and specificity comparable to that of human clinicians. This reduces the burden for ophthalmologists to read and interpret these images.
AI can also help in triaging patients and providing diagnostic advice on common ailments without human interaction. This potentially saves money and produces better health outcomes for patients. For example, Babylon Health in the UK has developed an AI system that claims to be capable of reasoning on a space of >100s of billions of combinations of symptoms, diseases and risk factors, to help identify conditions which may match the information entered by the patient, and allow them access to the right doctors through a telemedicine platform. It is currently being used by Britain’s National Health Service for some of its patients.
In research, AI has great potential in drug discovery: to shorten the time to identify new drugs. The typical success rate in traditional drug discovery is low. An AI-based approach can reveal relationships among drugs, genes, and diseases by analyzing multiple data sets including genomic, phenotypic and clinical data, and sifting through huge public life science knowledge bases. In this field, data is often too plentiful that it is difficult for the human mind to make sense of it, and that is where AI is critical. Pharmaceutical and biotechnology companies can leverage on this to find better novel targets for early-stage drug discovery. Another excellent use case for AI in research is in clinical trials. Clinical trials are expensive, time-consuming and often face problems in recruitment of patients. AI can identify the right patient profiles that would most likely benefit from a new treatment, enabling rapid patient recruitment and trial execution.
For AI to achieve its full potential in healthcare, we need to overcome several challenges, such as technological, regulatory and clinical barriers to data sharing and usage. Different sources of data are still mostly kept in silos by multiple stakeholders (insurers, pharma, hospitals, clinics) which means it remains fragmented. There are data privacy and security issues in bringing these data together. Often, health data is captured in a disorganized and unstructured manner, requiring a significant effort and cost to clean and re-organize in a meaningful way.
Another challenge is that AI tools are developed using historical data and need to prove their applicability in the real-world clinical setting. Hence, they will require prospective clinical studies to validate their effectiveness and safety, few of which have been published so far. A limited evidence base makes it more difficult for clinicians to adopt and apply them in their practice, especially since many of the algorithms are ‘black boxes’.
AI is not a panacea. AI is excellent at processing huge amounts of data but fails miserably at mimicking the care, empathy and human touch of a physician. It will enhance the capabilities of doctors rather than replace them. Thus, healthcare organizations with proper digital strategy and commitment to overcome the challenges will be able to reap the benefits of AI to generate better clinical outcomes and financial returns. Data and AI is powering the “Fourth Industrial Revolution” that is happening today and will be the key driver of innovation in healthcare in the coming years.