AI tools have the potential to change the way brain tumor patients access and understand critical care information, but without careful oversight, the same technologies can introduce new risks and uncertainties.
Research: Large-scale language models in brain tumor patient education: Opportunities, risks, and ethical considerations. Image credit: Nanci Santos Iglesias/Shutterstock.com
Brain tumor patients suddenly have to make sense of a lot of information related to their condition and medical care, coupled with emotional conflicts and cognitive overload. review of Frontiers of oncology concluded that large-scale language models (LLMs), if properly supervised, can be a useful tool for improving patient understanding and participation in care.
Brain tumors overwhelm patients with sudden cognitive and emotional burden
Brain tumors are life-changing for both patients and their families, often appearing suddenly with alarming symptoms such as seizures and cognitive impairment. As the disease progresses, personality changes, memory loss, and paralysis may occur, worsening both psychological and functional distress. This burden is compounded by poor outcomes. For example, the five-year survival rate for glioblastoma is less than 10%.
Despite existing tensions, these patients and their families need to be educated about this disease, the types of multidisciplinary care involved, the risks of each treatment approach, prognosis, and available support. Often these are people who lack significant health literacy.
Current patient literature on brain tumors typically requires at least a high school education, sometimes more, and access to that information is limited for those who need it most. The doctor’s explanations are thorough, but there is too much to understand at once, and the consultation time is limited. Anxiety and cognitive overload make it difficult for patients and caregivers to understand and remember this important information and to obtain new information as their condition changes. You can’t get the answers you’re looking for, so you go online or join a support group.
Recognizing the challenge of providing patient education in a manner that satisfies patients and their families, the authors of this paper considered LLM for its ability to fill this gap. As a narrative review, themes considered important were selected by the authors, possibly introducing selection bias. However, their selection was directed by experts based on a curated body of literature.
LLM is an artificial intelligence (AI) system trained on large amounts of data to provide human-like answers, simplifying when asked, and clarifying in certain situations. Unlike healthcare providers who can only see one patient at a time and a limited number of patients per day, you can handle multiple chats at the same time.
AI tools may support understanding, but lack true clinical insight
LLMs are trained to respond politely and reassuringly, and to convey empathy. This may provide psychological support to suffering patients, but evidence of lasting real-world effects remains limited. LLM can be integrated with other platforms to explain complex procedures, test results, and the effects of various treatments at an individual level. This potentially allows patients to feel heard and supported. LLMs can enhance medical advice by providing ongoing patient education outside of the treatment setting.
Overall, it has the potential to provide patients with the clear and timely answers they need to common diagnostic and treatment questions. If you don’t design your prompts carefully, they can produce output that is too specialized or at an advanced reading level.
LLM can be very helpful in explaining preoperative cognitive testing to such patients. These tests are key to planning your surgery, but they take a lot of time to explain. they could do it, Although not yet consistent, it transforms structured radiology reports into explanations that are understandable in real-world neuro-oncology settings.. Currently, LLMs are poor at interpreting advanced neuroimaging results, such as magnetic resonance imaging (MRI), and apparent success often involves illustrating reports produced by radiologists rather than directly analyzing raw imaging data. Although they may try to simplify such reporting, it can be misleading and in some cases may raise data privacy concerns.
Traditional metrics in evaluating LLM performance include the accuracy, completeness, brevity, and security of the information provided. However, the authors point out that in addition to usability, other qualities should also be evaluated, such as readability, cultural appropriateness, anxiety, and empathy.
Privacy, accountability and bias pose challenges to safe clinical deployment
Although LLM shows promise for patient education, its use has significant potential drawbacks. They generate responses to medical questions based on statistical analysis and computational manipulation of the data used for training. This can lead to inaccurate or non-existent information (“AI hallucinations”), for example about treatments and outcomes. To minimize this, recent research has focused on search augmentation generation (RAG), where LLM is restricted to preselected knowledge sources.
Additionally, LLM provides fluent and seemingly authoritative answers, which can induce overconfidence among patients and impede shared decision-making with clinicians. It can also create an emotional bond that can later lead to disappointment when expectations are not met. These aspects remain understudied despite being essential for use as a patient-facing tool.
Although seemingly empathetic, AI systems lack true insight and accountability, raising ethical concerns. This can lead to recommendations for impersonal care. Patient privacy is also an important concern.
Default LLM output is typically at or above the undergraduate reading level, emphasizing the need to structure prompts appropriately. This requires clinician training in the use of LLM.
The way LLM reaches its conclusions is difficult to interpret, especially in more sophisticated multimodal systems that use both visual and textual data simultaneously. The authors suggest that these should be applied carefully in clinical situations. Given the basic probabilistic design, there is a tendency to favor comprehensive coverage over strict clinical inference. As a result, unwarranted extrapolations and inferences are made, and the output tends to fluctuate. The output of the LLM needs to be validated by the neuro-oncologist at the level of decision-making information such as tumor characteristics and other diagnostic possibilities. An incorrect answer here can increase patient distress.
This highlights the need for careful monitoring, transparent output, technical guardrails such as RAGs, and clinician validation to balance the benefits of this new platform with appropriate safety measures. An example of a new regulation is the Valmed professor system. This is the first clinical decision support tool to receive EU (European Union) medical device CE approval. This heralds the formal regulation of these tools in healthcare. The EU is moving towards mandating the use of LLMs within human-in-the-loop architectures, a framework that ensures that LLMs act as assistants rather than agents in their own right.
Other pressing needs include using better models trained on better datasets. A secure framework for integrating an LLM into clinical practice includes multiple areas.
- Define the intended use
- Set clear boundaries
- Use structured prompts and mandatory uncertainty disclosure statements
- Ensure readability
- Require clinician verification
- Secure patient portal to ensure data privacy
- Establish safety metrics such as hallucination thresholds and accuracy goals
- Train clinicians and patients to use AI safely
According to these authors, legal responsibilities for LLMs in brain tumor patient education may include three areas: the manufacturer’s responsibility for system performance, the organization’s responsibility for regulating system implementation, and the clinician’s responsibility for validating the final decision.
Secure integration requires oversight, regulation, and improved models
Although LLM may be frequently used to educate brain tumor patients, future studies are essential to validate the outcomes of LLM across tumor subtypes, especially when tumors have a poor prognosis or are relatively rare. Evidence to date varies by tumor type, with more data available for some tumors (such as pituitary adenomas and meningiomas) than others, and some subtypes remain to be investigated.
Interactions between patients and LLMs, such as patient understanding, anxiety, decision-making, and overdependence, should also be studied. Robust real-world validation of patient outcomes remains limited, and increasing health literacy, improving multimodal LLM, and accountability are important future goals that will help limit LLM as an assistant rather than an autonomous tool in current practice.
Click here to download your PDF copy.
