- More accurate and timely detection of diseases
- Highly realistic medical training and simulations
- Faster drug development
- Synthetic medical data could revolutionise medical research
As healthcare costs continue to escalate — PWC anticipates a 7 per cent rise in 2024 alone — the industry stands at a critical juncture. The reasons for this surge are manifold, from the burnout experienced by healthcare workers and the resulting workforce shortages to ongoing friction between insurers and healthcare providers and the relentless creep of inflation. Amid these challenges, the quest for efficient patient care that doesn’t break the bank has led to a growing interest in cutting-edge solutions. Generative AI, in particular, has emerged as a promising solution that promises to redefine healthcare delivery, offering a glimmer of hope for an increasingly strained healthcare system. This groundbreaking technology employs machine learning algorithms to sift through and make sense of vast amounts of unstructured data — patient health records, medical images, and audio from consultations — and then generate new insights and solutions from what it learns. Accenture’s research suggests that AI could boost the efficiency of healthcare providers by a staggering 40 per cent, while Forbes estimates that the adoption of AI could slash the US medical sector’s costs by more than $200 billion per year.
The enthusiasm for generative AI is widespread among healthcare executives. A Bain & Company survey shows that 75 per cent believe it could be a game-changer. Yet, strikingly, only 6 per cent have actually devised a plan to leverage this technology. The immediate uses for generative AI are compelling, from simplifying the intricacies of billing to organising patient data and streamlining complex workflows. But the true promise of generative AI lies in the future, as further advancements in technology enable it to predict patient risks, support clinicians in making decisions, and even offer diagnostic and treatment advice. The road to adoption of generative AI is, however, not without hurdles, with many pointing to a lack of resources and expertise, as well as a myriad of regulatory issues as the main roadblocks. In this article, we’ll unpack what generative AI means for healthcare. We’ll look at how it’s already making waves, where it’s headed, and what needs to happen to overcome the obstacles in its path.
“By bringing comprehensive generative AI and voice-first capabilities to our EHR platforms, we are not only helping providers reduce mundane work that leads to burnout, but we are also empowering them to create better interactions with patients that establish trust, build loyalty, and deliver better outcomes”.
Suhas Uliyar, senior vice president of product management at Oracle Health
More accurate and timely detection of diseases
One area where generative AI could prove particularly helpful is in the early detection and ongoing management of complex health conditions. Using advanced algorithms, these systems can generate intricate medical images, enhancing the detection of diseases at their earliest stages and facilitating the development of tailored treatment plans. For example, Oracle recently announced the launch of the Oracle Clinical Digital Assistant, an innovative tool that allows healthcare providers to harness generative AI and voice commands to minimise manual tasks and enhance patient care. Integrating seamlessly with Oracle’s existing electronic health record (EHR) systems, the assistant’s capabilities extend to patients as well, enabling them to schedule appointments, pay bills, and access clinical data through simple voice commands.
Most notably, the Oracle Clinical Digital Assistant addresses the disconnect patients often feel during consultations, where healthcare providers seem more preoccupied with their screens than with the patient. By automating note-taking and suggesting context-sensitive actions, the assistant permits physicians to give their patients their undivided attention while administrative duties are streamlined in the background. It can also respond to the physician’s conversational voice commands, seamlessly integrating patient EHR information into the clinical workflow. “The EHR should be a provider’s best ally in delivering engaging, personalised care to the patients they serve”, says Suhas Uliyar, senior vice president of product management at Oracle Health. “By bringing comprehensive generative AI and voice-first capabilities to our EHR platforms, we are not only helping providers reduce mundane work that leads to burnout, but we are also empowering them to create better interactions with patients that establish trust, build loyalty, and deliver better outcomes”.
In Spain, the Community of Madrid’s Department of Digitalisation recently launched a pilot project in collaboration with Microsoft and the non-profit organisation Fundación 29, which will provide primary care medical professionals in the area with a generative AI-based app that enables them to diagnose rare diseases more accurately and quickly. Historically, patients grappling with rare diseases have had to contend with an arduous journey to diagnosis, often spanning over five years and entailing consultations with numerous specialists. Even then, more than half of these patients end up with no definitive diagnosis or, worse, an incorrect one, which can lead to inappropriate, sometimes harmful treatments. The introduction of DxGPT, as the new app is called, promises to revolutionise this process. Based on Open AI’s GPT-4 model, the app functions as a conversational assistant and is designed to enhance, not replace, the physician’s role. When a doctor inputs a patient’s clinical information, DxGPT quickly suggests a list of potential diagnoses. The physician can then refine this list by adding further details, such as the patient’s comprehensive medical history or laboratory test results. This iterative approach allows for a more nuanced and informed diagnostic process, enabling physicians to make more accurate decisions and direct patients to the appropriate specialists more promptly.
Highly realistic medical training and simulations
Another area where generative AI could be of use is in medical training and simulations. The University of Michigan’s AI model for sepsis treatment epitomises this innovation by providing medical trainees with highly detailed simulations that adapt to their clinical decisions in real time. This technology marks a significant departure from the static models of the past, enabling a hands-on, risk-free approach to learning that mirrors real-life patient interactions with remarkable accuracy. These advanced simulations are designed to identify and respond to the various stages of sepsis, equipping healthcare professionals with the experience to make critical decisions at the right moment. The ability to simulate the administration of antibiotics, the initiation of intravenous fluids, or the management of septic shock is invaluable in preparing medical personnel for the complexities of patient care.
Meanwhile, the Manchester-based Re:course AI is aiming to revolutionise medical training by using generative AI to create highly realistic simulations with digital human avatars, offering doctors, nurses, and other medical professionals the opportunity to engage in authentic patient interactions in a virtual environment. Developed in collaboration with clinical experts, the platform boasts a diverse range of avatars, each representing different health conditions and demographics, which ensures a broad and unbiased training experience. The system acts as a virtual tutor, guiding users through a variety of medical scenarios, and enhancing their diagnostic and decision-making skills. Early results indicate that Re:course AI could be a powerful ally for medical professionals, with participants showing a marked improvement in accurately diagnosing and managing a wide spectrum of conditions, from rare cancers to mental health issues, during simulated patient encounters.
According to the Congressional Budget Office, the average price tag for developing a new drug is anywhere between $1 billion to $2 billion.
Faster drug development
The astronomical costs and lengthy timelines associated with bringing new drugs to market are well-documented headaches for the pharmaceutical industry. With the Congressional Budget Office placing the average price tag for developing a new drug anywhere between $1 billion to $2 billion — including the sunk costs of failed drugs — the search for efficiency gains is more urgent than ever. The advent of generative AI technology, with its potential to dramatically cut both the time and expense involved in bringing new drugs to market, offers a glimmer of hope. Estimates suggest that AI could potentially carve up to $26 billion off the industry’s annual drug design and screening expenditures while also shaving a further $28 billion off clinical trial costs each year. Generative AI’s role in healthcare is broad and transformative. It’s transforming the way we approach drug discovery, enabling the creation of molecules with tailor-made properties for testing in the lab and predicting the behaviour of new drug candidates and proteins. AI’s ability to generate compounds that show promise in computer simulations means we can now identify potential drugs without the immediate need for costly physical trials. Moreover, AI’s capacity to predict the side effects of new drugs by analysing their molecular makeup is a game-changer and raises the bar for patient safety.
The pharmaceutical landscape is already evolving, as evidenced by a growing number of strategic collaborations between biotech companies and AI startups. A prime example of this trend is Recursion Pharmaceuticals’ recent acquisition of two Canadian AI startups for $88 million. This move not only signifies the industry’s growing trust in AI but also opens up new avenues for drug discovery, especially with the integration of generative AI technologies capable of handling complex datasets. In the academic arena, the University of Toronto’s researchers have made significant progress by developing an AI system that can design entirely new proteins using technology similar to that employed by AI image-creation platforms like Midjourney and DALL-E. The system, known as ProteinSGM, analyses image-like representations of existing proteins and their structures and then applies a generative diffusion model to generate new proteins that don’t appear in nature. This breakthrough has been substantiated by crafting these proteins in the lab, confirming their potential for therapeutic use. “Our model learns from image representations to generate fully new proteins at a very high rate”, says Philip M. Kim, a professor in the Donnelly Centre for Cellular and Biomolecular Research at U of T’s Temerty Faculty of Medicine. “All our proteins appear to be biophysically real, meaning they fold into configurations that enable them to carry out specific functions within cells”.
The true potential of generative AI in healthcare is perhaps best illustrated by the recent breakthrough achieved by the Hong Kong-based Insilico Medicine. The company’s AI-generated drug candidate, INS018_055, which targets idiopathic pulmonary fibrosis (IPF), has made it to Phase II clinical trials — a groundbreaking moment for AI-driven drug discovery. “To the best of our knowledge, this is the first drug discovered and developed using generative AI to have reached this clinical stage of development”, says Sujata Rao, chief medical officer at Insilico. What’s particularly remarkable is the amount of time it took to go from the discovery phase to the start of trials: a mere 18 months, a pace unheard of in traditional drug development circles. This was made possible by Insilico’s Pharma.AI suite, which consists of three separate platforms: PandaOmics for identifying disease targets, Chemistry42 for predictive studies, and inClinico for clinical trial predictions. Positive results from the Phase I trial of INS018_055 only reinforce the capabilities of AI, aligning well with Insilico’s preclinical models and showing a favourable safety profile. Now, as the drug enters a 12-week Phase II trial, the focus will be on assessing its safety, tolerability, and preliminary efficacy in a more rigorous setting.
Synthetic medical data could revolutionise medical research
The prohibitive costs of data collection, coupled with stringent privacy laws governing the use of patient information, have historically restricted the scope and pace of medical advancements. Generative AI offers a potential solution to this problem by creating synthetic data samples that can enhance existing health datasets without breaching individual privacy. This innovation allows for the generation of electronic health records (EHR), diagnostic scans, and other types of data that are indistinguishable from real patient information. Illustrating the potential of this technology, a team of German researchers developed GANerAid, an AI-powered tool that utilises a Generative Adversarial Network (GAN) framework to produce synthetic patient data for use in clinical trials. Remarkably, GANerAid can create medical data that retains the desired characteristics even when the source dataset is relatively small, which is a common challenge in studies involving rare conditions. Echoing this approach, another group of scientists tackled the issues posed by restrictive data-sharing regulations between hospitals. They crafted the EHR-M-GAN model capable of synthesising mixed-type EHR data that includes both continuous and discrete values, accurately reflecting the complexity of patient health trajectories. These generative AI tools not only ensure compliance with privacy standards but also enable more robust and democratic access to vital health data, catalysing research and potentially accelerating the discovery of new treatments and therapies.
Closing thoughts
It’s clear that generative AI is reshaping the healthcare landscape in ways we’ve only begun to explore. The breakthroughs in drug development and the vast improvement of diagnostic tools are game-changers, particularly for those battling chronic illnesses. This technology is not just a minor upgrade; it’s a true revolution that promises to enhance the precision and efficiency of healthcare delivery, with real human impacts. But let’s not gloss over the hurdles. Generative AI is powerful, yes, but it’s also complex and not without its problems. The healthcare field is responding to this challenge with vigour, pouring resources into research to ensure that AI evolves in a way that’s both beneficial and responsible. So, as we stand on the cusp of this AI revolution in healthcare, we are left with a compelling question to ponder: How will we balance the scales between the unbridled potential of generative AI and the timeless values of patient safety and ethical practice? If we navigate this question with foresight and humanity, generative AI may well be remembered as one of the most significant milestones in our relentless pursuit of health and wellbeing. The answer to this question will shape the future of healthcare and, by extension, the future of human life itself.
