AI and healthcare: Big opportunities, bigger risks?

The use of artificial intelligence in medicine can be traced back to the 1960s, when researchers first began exploring the use of computers to help diagnose and treat patients. In the following decades, advances in computing power and machine learning algorithms enabled the development of more sophisticated AI systems, leading to the creation of applications such as predictive analytics, medical imaging, and clinical decision support.

In recent years, the availability of large amounts of healthcare data, along with advances in natural language processing and computer vision, have further accelerated the development of AI in the medical field, paving the way for new applications such as telemedicine, medical devices, and drug discovery.

The union of AI and healthcare has been growing in recent years, with the global healthcare AI market expected to reach $188 Billion by 2030.

AI refers to the development of machines that can perform tasks that normally require human intelligence, like reasoning, perception, problem-solving, and language processing. Meanwhile, machine learning is a type of AI that involves teaching machines to recognize patterns in data and make predictions based on that data. Put another way: AI is the overall goal of creating smart machines, while machine learning is like a tool that helps us achieve that goal by allowing machines to learn and improve from data.

Erkin Ötleş, MD/PHD at the University of Michigan has spent over a decade studying AI/ML for healthcare. His work has encompassed the whole healthcare AI/ML lifecycle, spanning development, evaluation, implementation, and maintenance of AI/ML tools used clinically. Ötleş points out that while there are technical distinctions, the term AI has come to be used broadly in everyday language.

“In common discourse, people use the term ‘AI’ to refer to any embodied knowledge that sits outside of the human mind. For instance, technology systems that algorithmically mimic logic and reasoning are often referred to as ‘AI’, even though they may be simple if-then statements. AI classically has had many subdomains focusing on problems ranging from perception, being able to interpret information from sensors (like cameras) to natural language processing, and knowledge representation and reasoning. Machine learning (ML) is one of these subdomains and has advanced very rapidly recently. Machine learning has the special distinction of being data-driven. ML systems attempt to encode broad knowledge in a data-driven way, by ‘learning’ models from data.”

While the potential benefits of AI in healthcare are widely recognized, the industry has been slow to adopt these technologies due to a variety of factors, such as concerns about data privacy and security, consumer distrust, regulatory hurdles, and a lack of infrastructure. Additionally, much of what is currently being described as “AI” in healthcare is actually just algorithmic decision-making, which is a more limited form of AI that involves rule-based systems rather than machine learning or neural networks.

That being said, there are many breakthrough applications of artificial intelligence in healthcare that have the potential to revolutionize the sector:

• Medical Imaging
• Clinical Decision Support
• Predictive Analytics
• Electronic Health Records
• Telemedicine
• Medical Devices
• Robotics
• Drug Discovery
• Operational Efficiency

AI and ML are being used to improve the accuracy and efficiency of medical imaging, which is critical for early detection and diagnosis of many conditions. Companies like Aidoc and Zebra Medical Vision (acquired by Nanox) have developed AI-powered tools that can detect abnormalities in medical images and alert clinicians to potential issues. Bringing together AI and medicine can help reduce diagnostic errors and improve patient outcomes, particularly in areas like radiology where image analysis is a central component of diagnosis and treatment planning.

AI-powered clinical decision support tools can help clinicians make more informed and data-driven decisions, improving patient outcomes and reducing the risk of medical errors. AI healthcare companies like Merative (formerly IBM Watson Health) and Caresyntax are developing AI-powered tools that can analyze patient data in real time and provide clinicians with personalized treatment recommendations. These tools can also help improve care coordination and reduce healthcare costs by identifying high-risk patients and directing resources to where they are most needed.

Artificial intelligence and machine learning in the medical field are being used to analyze large volumes of patient data and generate predictive insights that can help healthcare providers better anticipate patient needs and allocate resources more effectively.

For example, a study conducted at Johns Hopkins University used machine learning in healthcare to predict the likelihood of sepsis, demonstrating how medical AI can help physicians detect and prevent sepsis before it becomes life-threatening. However, there is much more work to be done to ensure that models like this are appropriately calibrated in order to adequately predict or identify cases in contemporary clinical practice.

Predictive analytics tools have the potential to identify patients who are at high risk of developing certain conditions or experiencing certain health events like an overdose, allowing for earlier interventions and more targeted treatments. Companies like Medial EarlySign and Health Catalyst are leveraging AI and ML to develop predictive analytics tools that can help healthcare providers better manage population health and improve outcomes for patients.

EHRs provide a wealth of patient data, but analyzing this data can be time-consuming and error-prone. AI algorithms have the potential to automate tasks such as transcription and coding to improve efficiency and accuracy.

Additionally, AI may be able to analyze EHR data to identify patterns and make predictions about patient outcomes. For example, hospitals are leveraging AI to identify their most at-risk patients. However, Ötleş also points out that EHRs were not originally designed for AI/ML so there are many issues that can arise in implementation, some of which are quite difficult to detect.

AI-powered chatbots can provide basic healthcare advice and support, reducing the burden on healthcare providers. AI can also help automate tasks such as triage and diagnosis, improving efficiency and accuracy.

This is particularly important for improving access to healthcare services for rural and underserved populations, as well as reducing healthcare costs by reducing the need for in-person visits and hospitalizations. For example, the telemedicine platform, 98point6 (recently acquired by Transcarent), uses AI-powered chatbots to provide patients with basic healthcare advice and triage services.

AI is also being integrated into medical devices, making them smarter and more efficient.

For example, companies like Abbott are developing AI-powered pacemakers that can learn from patient data to optimize treatment and reduce the risk of complications.

Similarly, AI-powered wearable devices like the Apple Watch can collect and analyze data on a patient’s health status in real time, providing valuable insights for clinicians and helping patients stay on top of their health. A study published in the Journal of Healthcare Engineering demonstrated that artificial intelligence in medical devices can predict ventilator failure, which can be life-threatening for patients who require ventilation support.

AI-powered robots are being used in healthcare to perform surgeries and other medical procedures with greater precision and accuracy. For example, the Da Vinci Surgical System is a robotic surgical system that uses AI and machine learning to assist surgeons in performing complex surgeries.

AI and healthcare has the potential to significantly speed up the drug discovery process by helping researchers identify promising drug candidates more quickly and efficiently.

Companies like BenevolentAI and Insilico Medicine are using AI to analyze vast amounts of data and generate new insights into disease mechanisms and potential drug targets. This approach has the potential to reduce the time and cost involved in developing new drugs, and could ultimately lead to more effective treatments for a wide range of conditions.

AI and machine learning can also be used to optimize operational efficiency in healthcare, reducing costs and improving patient experiences.

Companies like Olive are developing AI-powered solutions that can streamline administrative tasks, such as insurance verification and surgical documentation, allowing clinicians to spend more time on patient care.

Additionally, AI can be used to optimize supply chain management, reducing waste and ensuring that critical supplies are always available when and where they are needed.

With a near constant buzz about the potential for artificial intelligence to improve care, not nearly as much attention is paid to the challenges and risks of implementation. Ötleş explains that there are two main issues at play: The first is a lack of infrastructure—both technical and human—to successfully implement these systems. The second is a focus on big splashy applications of AI in clinical settings rather than more gradual application across lower risk workflows.

Ötleş explains that the infrastructure that has allowed for rapid application of AI in other sectors such as consumer technology is far less developed in healthcare.

“Right now, we really don’t have great tooling to do healthcare AI—actually, it’s almost non-existent. We have EHRs and other clinical information systems, but we don’t have tools to build a model, deploy, implement, and monitor it over time.”

He continues: “Most healthcare organizations right now don’t have team members with skillsets to be able to assess AI models. Validating and monitoring these tools requires a special combination of clinical and AI expertise. And for those organizations that have these skills, they’re still fighting against a very siloed technical and data environment. In the tech sector, companies like Facebook and Google own their whole technical stack. They have access to all of their data and can monitor these systems very carefully. Plus they’ve relied on rigorous evaluation of technology, like A/B testing, since day 1. Healthcare providers operate in a completely different environment, are relatively underequipped, and have to manage extremely high risks.“

Ötleş also points out that the current discussion around the potential of AI and healthcare often aims too high too quickly. With the noted lack of infrastructure, there needs to be a more gradual and low-risk approach to testing out these new technologies.

He continues: “One of the areas that I find most promising for initial AI traction is documentation. This is a huge problem in our current system and if you can improve documentation, you can reduce physician burnout, increase efficiency, and make people’s lives easier. Building tools that help physicians summarize charts or document faster are probably going to be much more beneficial than trying to build a new diagnosis prediction model. These tools don’t directly affect clinical decision making, so they reduce some of the risk and complexity and solve a pressing clinical need.”

Recent developments in AI in healthcare bring great promise for the industry but also immense risk if not treated with appropriate caution. It’s key that organizations have knowledgeable technical team members who can support the implementation of new AI-based systems. As infrastructure—both technical and workforce—develop, the industry will be better prepared to move toward a bright future of AI-enabled care.

HTD Health is a digital health consultancy specializing in next-generation software data engineering. To get in touch about a project or consult with a specialist, reach out to info@htdhealth.com.