Trainee Focus debate: Artificial intelligence will have a positive impact on emergency medicine

Abstract

Artificial intelligence (AI) and machine learning (ML) is a rapidly expanding technological field that presents great potential for EM and will infiltrate many facets of our future practice. This article outlines the positive arguments for the use of AI tools in EM. In order to benefit from AI, we need to understand the current capabilities. Narrow AI is the iteration that is being implemented in the current medical landscape. This requires AI tools to be trained on large data sets and can be used for automating the manual processing of writing content. It can also be used to summarise complex data into coherent narrative structures. Well-known general examples include: Amazon's Alexa and Apple's Siri. General AI is a more advanced theoretical form of AI.1 The clinical applications of AI tools are broad and include documentation, patient flow, diagnostics, point-of-care ultrasound (POCUS) applications and clinical decision-making. Clinical documentation comprises a large proportion of EM shifts, with clinicians spending approximately 43% of their shift on data entry.2 Streamlining and automation of documentation would allow clinicians to focus their time on clinical interactions, reducing documentation burden, and improving documentation quality. The majority of our clinical AI research comes from radiological studies. Computer vision ML can be trained to recognise radiological pathology, with computer-aided detection and diagnostic systems assisting the identification of abnormalities in imaging studies.3 One study showed that generative AI was able to produce radiology reports of similar clinical accuracy to in-house radiology reports.4 This has the potential to identify abnormalities sooner, improve accuracy, and expedite patient disposition. POCUS has already been revolutionary for EM. In 2020, the Food and Drug Administration approved an echocardiography software algorithm that uses ML to calculate the ejection fraction from auto-captured 3D video clips. It also uses ML software to assist in obtaining the optimal images. The estimates of ejection fraction agreed with human experts.5 Similar technology will help EM clinicians obtain high-quality POCUS images at the bedside, along with accurate interpretation. Inefficiencies of patient flow lead to poor patient outcomes. AI tools have the capacity to incorporate large volumes of data in real time augmenting patient flow. These combined elements will greatly increase patient time-to-diagnosis, disposition and clinician satisfaction. AI tools have been shown to triage ED patients more accurately than current common triage tools.6 Generative AI tools such as Chat GPT could be used to write practice questions for exams and generate large question banks that trainees can use to hone their exam technique. Generative AI could be used as a training assistant to help mentor students and provide more bespoke teaching feedback. Generative AI could also be used to synthesise large volumes of textbook information into more understandable and applicable content. AI tools have been used to produce virtual simulations for surgical training, improving surgical technical skills.7 This can be applied to high acuity, low occurrence procedures or situations in an EM context, allowing trainees to get regular exposure and develop competence. This technology could also be extended to help trainees navigate and simulate difficult communication scenarios. Clinical training requires extensive time, emotion and energy from trainees. The majority of doctors balance both clinical and personal lives alongside exams. AI tools have the potential to make this process more manageable, enjoyable and reduce clinician burnout. Indigenous health is an integral part of our clinical practice in Australasia; however, we still have a long way to go before equity is achieved. Current early evidence suggests AI tools are able to enhance equity by reducing the effect of human bias on care, lowering barriers to knowledge, and increasing the productivity of healthcare professionals.6 Data sovereignty refers to the rights and interests of an individual group in the collection, ownership, and application of their data. If systems are designed with First Nations and Māori involvement, then data sovereignty and equity will likely be achieved. With the use of the correct training data for generative AI systems we can ensure that AI systems are more equitable. AI can be used to assist in ethical and patient-centred decision-making. One study found that AI-initiated palliative care decisions were associated with a lower hospital readmission rate, and a higher referral rate to palliative care.8 This highlights that AI tools can be beneficial in facilitating difficult decision-making while working with patients. At the time of writing, both Australia and Aotearoa New Zealand have no specific laws relating to AI in a medico-legal context. However, the frameworks they do have in place provide a reasonable starting point. In both countries, governments and medical organisations are taking notice and recognising the need for AI governance. The most relevant legal frameworks in place in both countries are their respective Privacy Acts. They define how personal data is collected, used, stored and distributed. In Aotearoa New Zealand, the Privacy Commissioner has issued guidance for the use of generative AI.9 Specifically, they mention engagement with Māori about the potential impacts of generative AI. The federal government in Australia is seeking to amend the 1988 Privacy Act so that individuals will have the right to request meaningful information about how automated decisions are generated.10 Any technology that can allow us to document more succinctly and accurately will be beneficial from a medico-legal perspective. In reality it is unlikely AI tools will be implemented into common clinical practice or public hospitals until relevant regulation and legal protections exist. Decisions made using AI tools prior to this should be mindful of the potential medico-legal implications. When used thoughtfully, and with the correct legal and ethical frameworks in place, AI tools have incredible potential to make a positive impact in EM. With the appropriate training emergency clinicians can use AI as ‘augmented intelligence’ that can enhance their skill set. RM is a section editor for Emergency Medicine Australasia.

Ähnliche Arbeiten