Using Large Language Models to Make Medical Data Communication More Patient-Centered

Large Language Models (LLMs) represent a promising approach to addressing one of healthcare’s most persistent challenges: ensuring that patient communication is captured, understood, and acted upon in ways that respect individual differences in health literacy, language proficiency, and cognitive abilities. While the technology is still evolving, early evidence suggests that LLMs may offer new opportunities to transform how we capture and utilize patient voice in clinical care, with Natural Language Processing (NLP) serving as one component in the broader pipeline.

A recent JAMA Insights article by Zikmund-Fisher, Thorpe, and Fagerlin on communicating medical numbers highlights a significant challenge in healthcare communication: only 34% of US adults can perform simple numerical tasks, and 97% of adults at the lowest numeracy level also have the lowest literacy. The authors present five evidence-based recommendations for communicating numbers to patients, which could serve as a foundation for developing LLM-based systems that better support patient understanding. These recommendations include using numbers instead of verbal probability terms, maintaining consistent risk denominators, presenting absolute rather than relative differences, using part-to-whole visual displays, and providing contextual information for unfamiliar data. LLMs may offer opportunities to implement these recommendations more systematically and adaptively, though their implementation requires careful consideration of both opportunities and limitations.

The Role of NLP and LLMs: From Patient Voice to Personalized Health Information

NLP technology serves as the initial processing step, capturing patient communication in ways that may be more comprehensive than traditional clinical documentation. While current documentation relies primarily on provider interpretation and standardized formats, NLP systems could potentially capture additional context from patient communication, including emotional indicators, cultural expressions, and individual communication patterns. However, it’s the subsequent LLM processing that would adapt this information according to the JAMA recommendations for different patient needs. This two-step process—NLP for capture and LLMs for adaptation—is still in development and requires careful validation before widespread clinical implementation.

Consider the case of Maria, a Spanish-speaking patient with diabetes who has difficulty understanding numerical health information. Following the JAMA recommendations, traditional clinical documentation might record her HbA1c as “8.3%, target <7%,” but this format may not account for her language preferences, numeracy level, or cultural context around diabetes management. The JAMA authors emphasize the importance of providing contextual information for unfamiliar data, such as explaining that “normal is 4.0% to 5.6%” and that “our target is below 7%.” NLP systems could process her spoken concerns—“Me siento muy cansada y tengo mucha sed”—and extract clinical symptoms and emotional context. LLMs could then adapt this information to generate patient education materials that follow the JAMA recommendations for using numbers, consistent denominators, and contextual information, tailored to her language preferences and health literacy level. However, such applications would require extensive testing to ensure accuracy, cultural sensitivity, and appropriate clinical utility.

The Multimodal Potential: Written, Audio, and Visual Communication

The combination of NLP and LLM technology may offer opportunities to process multiple communication modalities and adapt outputs to better match patient capabilities and preferences. NLP would handle the initial processing of patient voice, while LLMs would generate the adapted content following the JAMA recommendations. This approach could represent an important step toward more personalized patient engagement, though the technology’s current capabilities and limitations must be carefully considered.

Written Communication Adaptation

For patients who prefer written communication or have hearing impairments, NLP systems could process spoken conversations, while LLMs could generate written summaries that better match individual health literacy levels. Following the JAMA recommendations, a patient with limited health literacy might receive a summary stating, “Your diabetes is getting better. Your blood sugar number went from 8.3% to 7.8%. This is good progress!” while a patient with higher health literacy might receive a more detailed summary with specific numerical targets and clinical terminology. The JAMA authors emphasize using numbers instead of verbal probability terms (e.g., “35%” instead of “common”) and providing contextual information for unfamiliar data. LLMs could potentially implement these recommendations more systematically, though the accuracy and appropriateness of such adaptations would need to be carefully validated.

Audio Communication Enhancement

For patients who prefer audio communication or have visual impairments, NLP could process written clinical information, while LLMs could generate spoken explanations that consider language proficiency and cultural context. A Spanish-speaking patient might receive audio instructions about medication timing that incorporate cultural references to meal patterns, while an English-speaking patient might receive more direct, time-based instructions. The effectiveness of such adaptations would require extensive testing across diverse populations.

Visual Communication Adaptation

LLMs may offer opportunities to generate visual representations of health information that could be adapted to patient cognitive abilities and cultural preferences. The JAMA authors specifically recommend using part-to-whole visual displays such as icon arrays or stacked bars when showing risk graphically, as these formats accurately represent the part-to-whole relationship. For patients with low numeracy, LLMs could potentially create icon-based visualizations of risk probabilities using these evidence-based formats, while patients with higher numeracy might receive more detailed charts and graphs. The system could potentially adapt color schemes and visual metaphors to match cultural preferences and accessibility needs, though these capabilities would need thorough evaluation for accuracy and cultural appropriateness.

Health Literacy and Language Proficiency: Potential for Personalization

The JAMA article highlights an important insight: effective health communication requires understanding not just what patients need to know, but how they can best receive and process that information. LLM technology may offer opportunities to assess and adapt to individual health literacy and language proficiency levels, though this capability would require extensive development and validation.

Adaptive Health Literacy Matching

LLM systems could potentially analyze patient communication patterns to assess health literacy levels and adjust information presentation accordingly. For example, a patient who uses simple language and asks basic questions might receive explanations using everyday terms and analogies, while a patient who uses medical terminology and asks detailed questions might receive more technical explanations with specific numerical data. However, the accuracy of such assessments and the appropriateness of resulting adaptations would need careful validation across diverse populations.

Language Proficiency Optimization

For the 67 million Americans who speak languages other than English at home, LLMs could potentially ensure that clinical information is presented in their preferred language while maintaining medical accuracy. Additionally, LLMs might adapt the complexity and cultural context of information to match language proficiency levels, ensuring that translated information is not just linguistically correct but culturally appropriate and cognitively accessible. These capabilities would require extensive testing and cultural validation.

Cognitive Ability Adaptation

LLM’s potential may extend to cognitive ability adaptation. For patients with cognitive impairments, LLM systems could potentially generate simplified, repetitive explanations with visual cues and memory aids. For patients with high cognitive abilities, the system might provide detailed, analytical information that supports their desire for comprehensive understanding. However, such adaptations would need to be carefully validated for clinical appropriateness and patient safety.

Clinical Pathway Integration: Potential Applications

The potential value of combining NLP and LLM technology in healthcare may extend beyond processing patient voice to how this processed data could be integrated into clinical pathways to support more patient-centered care experiences. This integration could represent an important step toward better connecting patient communication, clinical decision-making, and care delivery, though significant development and validation would be required.

Real-Time Clinical Decision Support

NLP-processed patient voice data could potentially feed into clinical decision support systems, while LLMs could provide providers with additional insights into patient understanding, concerns, and preferences. For example, a provider might receive information indicating that a patient appears confused about their medication instructions, along with suggested communication strategies tailored to that patient’s specific needs and capabilities. However, the accuracy and reliability of such systems would need extensive validation.

Personalized Care Plan Generation

LLM systems powered by NLP-processed patient data could potentially generate care plans that better match individual communication preferences, health literacy levels, and cultural contexts. A care plan for a patient with low health literacy might include simple, step-by-step instructions with visual aids, while a care plan for a patient with high health literacy might include detailed explanations with specific numerical targets and clinical rationale. The appropriateness and effectiveness of such adaptations would require careful evaluation.

Adaptive Follow-Up Communication

LLMs could potentially enable more adaptive follow-up communication that evolves based on patient responses and engagement patterns. A patient who consistently responds well to visual information might receive follow-up materials with infographics and charts, while a patient who prefers audio communication might receive phone calls with personalized voice messages. However, the effectiveness and appropriateness of such adaptations would need thorough testing.

Potential Benefits: Supporting Healthcare Delivery

The integration of patient-facing LLM systems into clinical pathways could potentially represent an important step toward more patient-centered care that better honors individual differences in communication, cognition, and cultural context. However, this potential must be balanced against the need for careful development, testing, and validation.

Addressing Communication Barriers

LLM-powered systems could potentially help address communication barriers in healthcare by ensuring that patients receive information in formats that better match their capabilities and preferences. This could include not just language translation, but cultural adaptation, health literacy matching, and cognitive ability optimization. However, the effectiveness of such systems would need to be demonstrated through rigorous evaluation.

Enabling More Personalized Care

By processing patient voice into structured data through NLP, LLMs could potentially enable systems to deliver care that is more personalized to individual needs, preferences, and capabilities. This could go beyond demographic personalization to include communication style, learning preferences, and cultural context. The clinical value and appropriateness of such personalization would require careful assessment.

Supporting Provider-Patient Relationships

LLMs could potentially enhance the human elements of healthcare by providing providers with additional insights into patient needs and preferences. Providers could focus on the clinical aspects of care while LLM systems handle some of the complex tasks of ensuring that patients receive information in appropriate formats. However, the role of LLMs in supporting rather than replacing human judgment would need careful consideration.

The Future of Patient-Centered AI: Opportunities and Considerations

The potential of combining NLP and LLM technology to transform patient voice into structured data for AI-driven clinical pathways represents an important area of development in healthcare. As we consider how these systems might process and respond to the complexity of human communication, we have the opportunity to explore healthcare experiences that better honor individual differences and capabilities.

This future requires not just technological advancement, but a fundamental commitment to equity, accessibility, and patient empowerment. It requires building AI systems that can adapt to the full spectrum of human diversity, from health literacy levels to language proficiency to cognitive abilities. However, achieving this vision will require careful attention to evidence-based development, rigorous testing, and ongoing evaluation of clinical outcomes.

The journey toward this future begins with recognizing that every patient’s voice matters, and that technology should serve to amplify rather than diminish the unique needs and capabilities of each individual. Through careful development of NLP and LLM technology for patient-facing applications, we may have the opportunity to create healthcare experiences that are not just technologically advanced, but truly human-centered.

However, it is important to acknowledge that many of these capabilities are still in development and require extensive validation before widespread clinical implementation. Patient advocates and healthcare providers should approach these technologies with appropriate caution while remaining open to their potential benefits when properly developed and tested.