AI and Communication Styles in Healthcare

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