AI is fueling major hopes in healthcare and medicine, especially in oncology. It is opening the door to more precise, personalized therapies and better patient care across all aspects of cancer treatment, said Fabio Dennstädt, MD, of the Department of Radiation Oncology at the University Hospital of Bern in Bern, Switzerland.

AI is already improving routine oncologic diagnostics and could expand the range of methods for the prevention and early detection of preneoplastic and malignant lesions. Furthermore, AI systems combined with clinical, histopathologic, and molecular data could advance precision medicine, emphasized Stefan Schulz, MD, and Sebastian Foersch, MD, from the Institute of Pathology at the University Medical Center Mainz in Mainz, Germany, and Digital Pathology & Artificial Intelligence Group.

Inga Berben, an entrepreneur and expert in digital innovations in healthcare, also sees AI as a major opportunity to improve medical care in a sustainable and fundamental way. For her, AI is absolutely essential in oncology, as Berben recently stated at an event hosted by MSD Sharp & Dohme GmbH in Wiesbaden, Germany: Treatment options for solid tumors have become increasingly complex. At the current rate of FDA approvals, treatment options could increase manyfold in 5 years. Overall, 59% of oncologists in the US experience burnout. No single person can keep track of all this information anymore. Not even the tumor board.

Personalized Therapies — Optimization With AI

AI, on the other hand, can analyze enormous amounts of data and precisely classify individual patient situations; as a result, it can help support clinical decisions and optimize personalized therapeutic approaches, said Dennstädt. AI is therefore playing an increasingly important role in each of the three classic pillars of oncologic therapy, namely, radiation therapy, surgery, and systemic therapy.

This development has been accelerated in particular by the emergence of large language models such as ChatGPT. These new AI tools have the potential to facilitate daily work not only in data analysis but also in decision-making and information management, explained Jan C. Peeken, PhD, Department of Radiation Oncology and Radiotherapy, TUM University Hospital Munich in Munich, Germany, and Jakob Nikolas Kather, MD, Technische Universität Dresden in Dresden, Germany.

The power of AI is demonstrated, for example, by a study published in Science by researchers led by the US internist Adam Rodman, MD, of Beth Israel Deaconess Medical Center in Boston. At the start of emergency care, when a patient arrived at the emergency room and provided only limited information about their symptoms, the OpenAI o1 language model integrated into ChatGPT provided an exact or very close diagnosis in 67% of cases — more than 10% points better than two physicians in the same cases. Although the gap narrowed slightly when more information was available, the large language model still outperformed the physicians by 2%-10% points later in the care process.

Impact on Doctor-Patient Communication

AI is not only becoming increasingly important for diagnosis and treatment but also playing a major and growing role in the physician-patient relationship and physician-patient communication. For example, patients today can use AI tools to access a wealth of medical information and acquire knowledge that might even surprise experienced physicians.

According to a study by the Boston Consulting Group, consumers are already using AI as a gateway to healthcare. “Nearly 60% of consumers already use AI for personal health,” said BCG expert Ben Keneally and his colleagues. In the first cross-national survey on consumer use of AI in healthcare, they surveyed more than 13,000 internet-savvy adults in 15 countries about how they use AI for their health and to what extent they are aware of and comfortable with physicians using AI. In addition to questions about their expectations and concerns regarding AI, participants were asked whether they had used or currently use AI-powered tools for their healthcare. These tools included, for example, AI chatbots for general advice; AI-powered health apps; AI-enabled wearables or fitness trackers; AI-based tools for mental health, nutrition, and fitness; as well as AI tools for sleep monitoring and for explaining medical findings or treatments.

According to the researchers, AI-enabled wearables (58%) and AI tools for sleep monitoring (49%) are used several times a week or daily, closely followed by AI chatbots for health advice (44%).

AI agents could drive this trend even further. They could be used to book appointments, check insurance coverage, or compare costs. However, an AI agent would require access to appointment data and authorization tools stored in healthcare administrative systems, as well as permission to share personal data. Consequently, consumer use of AI is currently largely focused on checking symptoms, asking questions, interpreting test results, and understanding treatment options, explained Keneally and his colleagues. Furthermore, consumers have concerns regarding data privacy and the reliability of the advice provided by AI. The most interesting finding is that patients do not view the use of AI and real physicians as an either-or choice. Most respondents indicated that they would prefer to work either with AI alone or with a human supported by AI rather than with a human alone.

Risks and Challenges

As is well known, where there is light, there is also shadow. AI tools are no exception. AI can support the care of patients with cancer in many ways, according to a team of authors led by AI specialists Jan Clusmann, MD, and Jakob N. Kather, MD. However, there are also risks, including false conclusions, hallucinations, and confirmation bias — a cognitive distortion in which people unconsciously seek, filter, and interpret information in ways that reinforce their existing beliefs. Other concerns include vendor lock-in; deskilling, which refers to the gradual loss of human skills and knowledge; and shadow use, meaning the unauthorized use of AI tools by employees in the workplace.

“Despite the immense potential of AI for oncologic therapy, its broad clinical implementation still faces significant challenges,” added Dennstädt. Many systems show promising results in research, “but only a few make their way into everyday clinical practice.” One of the biggest hurdles lies in data availability and quality. While there are a lot of data in oncology, they are often fragmented, incompatible, or inadequately annotated. Training robust AI models requires vast amounts of high-quality, well-structured data. Building large, interoperable oncology databases is one of the biggest challenges.

Other issues include regulatory and validation concerns. Before an AI system can be routinely used in patient care, it must undergo strict approval processes. This requires comprehensive clinical trials that demonstrate not only the technical performance but also the actual clinical benefit and safety of the AI applications. This is highly relevant given the complexity of oncologic disease progression and the potentially far-reaching consequences of incorrect decisions.

Despite these hurdles, Dennstädt believes the potential benefits outweigh the risks; AI holds immense promise for oncologic therapies. From automated radiation planning and AI-assisted surgeries to predicting responses to chemotherapy and immunotherapy, AI has the potential to improve the treatment of patients with cancer.

However, AI fuels not only hopes but also fears: There is growing resistance to AI, reports Swiss journalist Alain Zucker in the Neue Zürcher Zeitung. People have sabotaged robots, blocked data centers, and booed tech billionaires. One of the main reasons is the fear of job loss. But that’s a different story.

This story was translated from Univadis Germany, part of the Medscape Professional Network.