AI in Clinical Practice

Original Editor - Michael Rowe

Top Contributors - Stacy Schiurring, Jess Bell and Kim Jackson  

Introduction

The integration of artificial intelligence (AI) into rehabilitation and clinical practice presents both unprecedented opportunities and profound challenges that will reshape the future of our profession. Concerns about new technologies coming between the clinician and patient are increasingly relevant. These changes highlight a need for current and ongoing conversations and education for healthcare providers on the potential uses and benefits of AI in clinical practice.

Common concerns or hesitations about integrating AI into healthcare include:

• human interaction and relationships are a vital part of healthcare
• only humans can take on the role of care-provider
• clinical practice has a morality component, and AI cannot be moral
• clinical reasoning is too complex for AI

This article provides an overview of the evolving role of AI in healthcare and discusses the moral and ethical considerations behind human and AI collaboration.

Important Definitions

Generative AI: a type of artificial intelligence designed to create new content, such as text, images, music, and videos, based on input data. It uses algorithms to learn patterns and features from existing data, allowing it to generate original outputs, as directed by prompts, that resemble the input data but are not direct copies.

Large Language Models (LLM): a type of artificial intelligence designed to understand and generate human-like text. LLMs can perform a variety of tasks, including answering questions, summarising texts, translating languages, and engaging in conversational dialogue. In the context of this page, there are two types of LLMs:

• Generalist Models: LLMs trained on a broad and diverse dataset, enabling them to perform a wide range of tasks without being specifically optimised for any single task. They can handle various natural language processing tasks like text generation, summarisation, translation, and answering questions, but their performance may not be optimal for specialised tasks.
• Fine-tuned Models: derived from generalist models but have undergone additional training (fine-tuning) on a narrower, task-specific dataset to enhance performance on particular tasks. Due to their focused training, these types of models excel in specific applications, such as sentiment analysis, legal document summarisation, or medical data interpretation.

AI and Clinicians

Human versus AI Outputs

It is best not to judge AI outputs against the "best" human standard but rather against the "best available" human.

There are circumstances in healthcare where AI can be more efficient than humans in creating output or offer a time-saving option. Some current and potential uses of AI in healthcare include:

• literature searches
• consolidation of and summarising evidence-based practice research
• drafting empathetic responses and educational materials
• completing online routine patient queries
• using AI avatars for routine check-ins and motivation
• drafting discharge summaries
• drafting referral letters
• assisting with automated electronic medical record data capture via conversational AI

AI is a tool to assist and extend the abilities of the healthcare professional. For ethical and professional reasons, the healthcare professional must overview and supervise all AI-informed patient interactions and drafted documentation. This helps to ensure accuracy and patient safety.

Ethical Considerations

There are many ethical concerns around using generative AI in clinical practice. This is an ongoing and evolving issue which will require vigilance by individual providers and within the greater healthcare system. Major ethical considerations associated with the use of AI in patient care can include:

• privacy and data security:
  • protect patient information when using an AI system
  • do not upload protected patient information
  • if you are using a publicly available generative AI system, read the terms of service to ensure that what is being uploaded will not be integrated into the training data set
• bias and fairness:
  • AI can perpetuate human biases that are present in the training data
  • clinicians should reflect on how these biases could potentially influence their management choices
• transparency and explainability:
  • currently, it is difficult to explain why generative AI produces specific output - there is a lack of understanding of how the input becomes the output
  • it is important to make AI decision-making more interpretable within a clinical context
• clinical validation and regulation:
  • there is a need for rigorous testing and regulatory approval of AI systems
  • regulation is challenging due to the rapid changes and advances between AI versions
• professional responsibility and accountability:
  • determine where responsibility lies when AI is involved in clinical decision-making
  • the clinician is ultimately responsible for choices and outcomes in patient care

Professional Considerations

There are various professional challenges to consider when integrating generative AI into clinical practice, including:

• staff development: healthcare professionals may be unable to take advantage of generative AI because of:
  • low digital/AI literacy skills
  • lack of institutional support in understanding AI use
  • limited understanding of the potential of AI as a tool in clinical practice​
• integration of AI systems into existing healthcare workflows and infrastructure:
  • can be complex and disruptive
  • will be a challenging transition to negotiate
• de-skilling of clinicians: while AI may reduce clinician fatigue and workload and increase efficiency and patient safety, rehabilitation professionals must consciously practise and maintain core skills as AI becomes more involved in clinical practice
• technology:
  • lack of transparency and explainability
  • limited contextual understanding (e.g. of complex patient histories, social determinants of health, or nuanced clinical contexts commonly associated with patient care)
  • regulatory and legal challenges about liability, patient privacy, and compliance
  • errors and safety concerns: AI systems make mistakes and produce unexpected outputs, which could lead to incorrect diagnoses or treatment recommendations which will have a negative on patient outcomes

AI and Human Collaboration

The future of clinical practice may be one where human-AI collaboration produces better patient outcomes than human clinicians or generative AI working alone. Clinicians who learn to effectively leverage AI will likely have a major advantage. The integration of AI into clinical practice will require a shift in mindset and the development of new technology literacy skills.

Trending research is showing that AI literacy and implementation into clinical practice can have a positive impact on diagnostic reasoning, patient interactions, and empathy with patients. But how can rehabilitation professionals prepare for the integration of AI into clinical practice? Some options to consider include:

• rehabilitation education programmes:
  • integrate AI education into healthcare curricula to prepare future professionals for practice
  • new graduates need to have a basic understanding of how to effectively integrate AI into clinical practice
• individual rehabilitation professionals:
  • develop digital and AI literacy skills to effectively leverage this technology
  • develop skills in managing the human-AI interface and use AI to augment practice
  • be aware of the changing dynamics in patient-clinician relationships as patients gain access to AI-based health tools
• rehabilitation professionals at an organisational, state, or national level:
  • become involved in creating frameworks for transparency, explainability, and accountability in AI use
  • skilled insightful rehabilitation input will be needed in the creation of future AI interfaces into the electronic medical record and clinical practice

AI and Patients

With the integration of generative AI into healthcare, patients now have access to the same or similar tools as healthcare professionals:

• AI empowers patients to take a more proactive role in managing their health
• generative AI can function as a "second opinion" for patients who may otherwise be unable to afford or obtain one within the healthcare system
• generative AI can also provide support for mental health outside of the traditional healthcare system as it is easy to access, of no or low cost, and fills a societal need

Resources

Ethical Considerations around AI and Healthcare

• Ethical Considerations in the Use of Artificial Intelligence and Machine Learning in Health Care: A Comprehensive Review
• Legal and Ethical Consideration in Artificial Intelligence in Healthcare: Who Takes Responsibility?
• The Impact of Artificial Intelligence on the Doctor-Patient Relationship

This 2-minute video overviews the main ethical and legal challenges of AI and healthcare:

This 4:30 minute video discusses key human right implications around AI and healthcare: