As part of our broader ethical analysis of patient values in AI-enabled healthcare, we conducted exploratory research into how AI might affect doctor-patient communication. This investigation forms one component of understanding the wider implications of AI in stroke neuro-rehabilitation.

AI-Assisted Communication in Healthcare

Recent studies suggest that AI systems, particularly large language models, can enhance various aspects of healthcare communication. Some research indicates these systems may in some cases generate higher-quality, more empathetic responses to patient questions compared to traditional physician responses. This raises intriguing possibilities for supporting communication during stroke rehabilitation, where patients may need frequent, patient-paced interactions to discuss their recovery journey.

Four Models of Doctor-Patient Communication

Our research explored how AI could emulate different healthcare communication styles, based on a widely-recognized framework developed by Emanuel and Emanuel. These four distinct approaches are:

- Paternalistic: The healthcare provider takes a directive role, recommending what they believe is best for the patient

- Informative: Focuses on providing detailed factual information, enabling patients to make their own choices

- Interpretive: Helps patients discover and understand their own values in light of their medical situation

- Deliberative: Engages patients in dialogue about what approach might be most suitable, while offering professional guidance

 

The Communication Style Dilemma

Currently, patients have little choice in how their healthcare providers communicate with them – it's essentially a lottery based on their provider's preferred style. While allowing patients to choose their preferred communication style could enhance care, it's challenging to implement with human providers who may not be able to easily switch between different communication approaches. This is what motivated our exploration of whether AI language models could offer a solution, potentially allowing patients to experience and choose their preferred style of medical communication.

Testing AI's Communication Abilities

To explore whether AI could effectively replicate these communication styles, we conducted a proof-of-concept study using GPT-4. We presented the AI with a hypothetical medical case involving a breast cancer diagnosis and instructed it to engage in conversations using each of the four communication styles. For each style, we provided the AI with specific guidance about the characteristics and objectives of that communication approach.

The results were promising: GPT-4 demonstrated the ability to adapt its communication style distinctly for each model. For instance, in the paternalistic style, it provided clear, directive recommendations, while in the interpretive style, it focused on helping the patient explore their values and preferences. We've since created custom versions of these AI communication styles that can be tested and refined for stroke rehabilitation contexts.

Challenges and Considerations

While AI-assisted communication shows promise for stroke rehabilitation, our research identified several important challenges that need careful consideration:

Echo Chambers and Bias

Allowing patients to choose their preferred communication style could have unintended consequences. Patients might gravitate toward styles that reinforce their existing beliefs and biases about their health, potentially limiting their exposure to different perspectives that could benefit their rehabilitation journey. For example, a patient who prefers highly directive communication might miss opportunities to develop their own decision-making capabilities during recovery.

Recent research has shown that AI systems can be remarkably persuasive - even more effective than humans at changing people's minds during conversations. While this could be beneficial for promoting healthy behaviors during rehabilitation, it also raises concerns about potential manipulation. We need to ensure that AI systems maintain a balance between supporting patient decisions and unduly influencing them.

Practical Challenges

Several practical issues need addressing before AI communication systems are in fact incorporate in stroke rehabilitation:

• Reliability and accuracy of AI responses
• Protection of sensitive patient information
• Integration with existing healthcare workflows
• Clear guidelines on when human healthcare provider involvement is necessary

Our exploration of AI communication styles is just the beginning. As part of the Artificial Intelligence in Stroke Neuro-rehabilitation project, we're now moving into several interesting directions:

• Stroke-Specific Communication Models: Developing AI communication approaches tailored specifically to different stages of stroke recovery, recognizing that communication needs may shift dramatically throughout rehabilitation
• Value Preference Learning: Creating tools to help us understand how patient values and communication preferences evolve during their rehabilitation journey, allowing for more dynamic and responsive AI support
• Ethical Framework Development: Building on this research to develop guidelines for the responsible integration of AI communication tools in stroke rehabilitation, with a focus on protecting patient autonomy and well-being

Risks, limitations, and responsible use of AISN technologies

The AISN project develops AI-based systems to support clinicians in decision-making and to optimise personalised stroke rehabilitation pathways. While these technologies offer important benefits, it is essential to clearly communicate their limitations and potential risks to patients, carers, clinicians, and other stakeholders.

Technical and clinical limitations

AI systems used in AISN rely on data-driven models to support clinical decision-making and rehabilitation planning. These systems have inherent limitations:

• Dependence on data quality and representativeness: The performance of AI systems depends on the data used for training and validation. Limited or biased datasets may affect the accuracy and fairness of recommendations.
• Uncertainty in the outputs: AI outputs are probabilistic and may not always be accurate, particularly in complex or atypical patient cases.
• Generalisation limits: Models developed in specific clinical contexts may not perform equally well across different healthcare settings or populations.
• Evolving system performance: AI systems may be updated over time, which can lead to changes in outputs and require continuous validation and monitoring.

For these reasons, AISN tools are intended to support clinical judgement, not replace it.

Risks related to clinical use

The integration of AI-based recommendations into clinical workflows introduces several risks:

• Over-reliance on AI outputs by clinicians or patients
• Misinterpretation of recommendations, especially without sufficient explanation or context
• Potential inaccuracies, which could impact treatment planning if not properly assessed
• Variability in patient response, to rehabilitation, as outcomes may differ between individuals
• Accessibility and usability challenges, particularly for patients with different levels of digital literacy

AISN mitigates these risks through clinical validation, iterative testing, and involvement of healthcare professionals and patients in system design and evaluation.

Ethical considerations

AISN is developed in line with European ethical principles for trustworthy AI, ensuring respect for fundamental rights, patient safety, and human dignity. Key ethical considerations include:

• Human oversight and responsibility: AI systems support, but do not replace, healthcare professionals. Final decisions remain under the responsibility of qualified clinicians.
• Patient autonomy and informed participation: Patients and carers should be informed about the role of AI in their care and be able to ask questions and express preferences.
• Fairness and non-discrimination: Efforts are made to identify and reduce biases in data and algorithms to avoid unequal outcomes across different population groups.
• Transparency and explainability: AISN promotes clear communication about how AI systems work, their purpose, and their limitations.
• Safety and risk management: Continuous monitoring, validation, and evaluation are carried out to ensure safe system performance.

These principles reflect the project’s commitment to ensuring that AI is developed and used in a way that benefits patients while minimising harm.

Legal and data protection framework

AISN operates in compliance with applicable European and national regulations governing data protection, medical research, and digital health technologies:

• Data protection (GDPR): Personal data is processed securely, with appropriate safeguards to ensure confidentiality and privacy.
• Data governance and access control: Access to data is restricted to authorised users, and data handling follows strict protocols.
• Regulatory compliance: AISN aligns with relevant frameworks for medical devices and AI systems in healthcare.
• Accountability and traceability: System use and outputs are documented to support auditability and responsibility.

These measures ensure that patient data is protected and that the use of AI complies with legal and regulatory standards.

Measures for safe and responsible use

To ensure that AISN technologies are used safely and effectively, the following measures are implemented:

• Clinical validation and continuous evaluation of system performance
• User-centred design, involving clinicians and patients in development and testing
• Training and guidance for healthcare professionals using the system
• Clear communication materials for patients and carers
• Defined roles and responsibilities, ensuring human oversight at all stages

AI outputs must always be interpreted within the broader clinical context, taking into account professional expertise and individual patient needs.

Communication of risks and limitations

A key objective of AISN is to ensure that patients, carers, and stakeholders are clearly informed about both the benefits and limitations of AI technologies:

• Information is provided through the AISN website and supporting materials
• Content is designed to be accessible and understandable for non-expert audiences
• Risks and limitations are communicated in a transparent and balanced way
• Stakeholders are encouraged to engage with the information and seek clarification where needed

This approach supports informed decision-making and helps build trust in the responsible use of AI in stroke rehabilitation.

 

AISN brings together different AI-based tools to support stroke rehabilitation, from assisting healthcare professionals in decision-making to enabling more personalised care across the recovery pathway. While these technologies can enhance care, it is important to understand their scope, limitations, and how they should be used responsibly.

The table below summarises the main types of AISN technologies described on the website and the key limitations and safe-use messages that patients and carers should be aware of. These examples are intended to support understanding, not to provide individual medical advice.

AISN technology or tool	What it supports	Main limitations	Safe-use message for patients and carers
AI-based clinical decision-support and prediction tools	Support healthcare professionals in analysing patient data, predicting recovery pathways, and selecting rehabilitation strategies	Outputs depend on the quality and completeness of available data; predictions may be uncertain and may not apply equally to every patient	These tools support clinical judgement but do not replace healthcare professionals
BrainX3 / brain simulation tools	Help explain and model aspects of brain function and recovery relevant to stroke rehabilitation	Simulations are simplified models and cannot provide certain predictions about an individual patient’s recovery	Simulation outputs should be interpreted by qualified professionals and used as one source of information
Rehabilitation Gaming System / digital rehabilitation tools	Support rehabilitation exercises and more personalised rehabilitation in clinical or home settings	Not every tool will be suitable for every patient; usability, access, fatigue, motivation, or individual response may vary	Use should be guided by healthcare professionals and adapted to the patient’s abilities and needs
AI-supported communication and information tools	Help improve how information is shared between healthcare professionals, patients, and carers	Information may need explanation, and patients may differ in how they understand or prefer to receive it	Patients and carers should be encouraged to ask questions and discuss unclear information with their care team

 

Understanding the limitations

AI systems in AISN support clinical care and decision-making, and are intended to be used with appropriate awareness of their scope and limitations.

• Their performance depends on the quality and completeness of available data
• Recommendations and predictions may not always apply equally to every patient
• Results may vary depending on the clinical setting and individual circumstances
• Some AI outputs may be complex and require professional interpretation
• System performance may evolve over time as technologies are refined

 

Possible risks

The use of AI technologies in healthcare requires careful consideration of potential risks:

• Over-reliance on AI outputs without appropriate clinical judgement
• Misinterpretation of recommendations if not properly explained
• Incomplete or uncertain outputs in some situations
• Differences in how patients respond to rehabilitation approaches
• Barriers to access or use, including differences in digital skills
• These risks are addressed in AISN through clinical validation, continuous evaluation, and close involvement of healthcare professionals and patients.

 

Safety, ethics, and data protection

AISN follows European principles for trustworthy AI and healthcare:

• Personal data is protected in line with data protection regulations (GDPR)
• Patients are informed about how AI supports their care
• Systems are designed to promote fairness and reduce bias
• Healthcare professionals remain fully responsible for medical decisions

 

Responsible use

To ensure safe and effective use of AISN technologies:

• AI outputs should always be reviewed and interpreted by qualified healthcare professionals
• Patients and carers are encouraged to ask questions and understand how AI is used
• Treatment decisions should combine medical expertise, AI support, and patient preferences

 

Understanding how these technologies work, and their appropriate use, helps ensure that AI is applied in a way that is safe, transparent, and beneficial for patients.

 

FAQ

What does AISN use AI for?
AISN uses AI to support healthcare professionals in planning and personalising stroke rehabilitation and improving patient care.

Can AI make decisions about my treatment?
No. AI provides support and recommendations, but healthcare professionals always make the final decisions.

Are AI recommendations always correct?
AI supports decision-making, but its results may vary depending on the data available and individual patient situations.

Is my personal data safe?
Yes. AISN follows strict European data protection rules (GDPR) to ensure your data is secure and used responsibly.

What are the main risks of using AI in rehabilitation?
Risks include misinterpretation of results, over-reliance on AI, and differences in how patients respond. These are managed through clinical oversight and system validation.

How can I use AISN technologies safely?
Always discuss AI-supported recommendations with your healthcare professional and ask questions if anything is unclear.