Artificial intelligence (AI) is becoming more prominent in today's evolving world. AI uses algorithms and machine learning techniques to improve and adapt, allowing it to make decisions and carry out tasks autonomously. Common examples of AI are virtual assistants such as Apple's Siri and Amazon's Alexa, speech and image recognition on our smartphones, and personalized recommendations on streaming platforms.
AI has even been utilized in medicine, particularly in radiology with AI-powered software assisting in the detection of disease. The field of headache medicine recently had perhaps its first encounters with AI; a study assessed the diagnostic accuracy of a computer-based diagnostic engine, and found its results were comparable to those obtained by headache specialists.[1] Another study used an AI-enabled prediction model assessing electrocardiograms (ECG) in patients with migraine. The model reported patients with migraine with aura had a higher prediction rate of atrial fibrillation (AF) compared to those with migraine without aura. The prediction model was developed by creating a convolutional neural network that assessed the probability of paroxysmal AF from a sinus rhythm ECG. Over 600,000 ECGs were inputted through the network and processed through each of its layers, such that the network would learn with every ECG.[2]
There are many instances in headache medicine where AI can potentially assist. One example could be AI-assisted triage of headache patients to the appropriate clinician. Improving the quality of consultations and reducing emergency room visits is an ongoing issue for the management of migraine.[3] With an AI-assisted triage model, it could be possible to facilitate and expedite the appropriate level of care for patients with migraine. AI could ease the process of submitting prior authorization for treatments, allowing clinicians and ancillary staff to focus more on clinical tasks.
ChatGPT is an exponentially growing application of AI that is becoming increasingly popular. GPT stands for Generative Pre-training Transformer and ChatGPT is classified as a general large language model (LLM). An LLM is a deep learning algorithm that can recognize, translate, predict, summarize, and generate a text response based on a given prompt. The model is trained on a dataset consisting of a diverse range of books, articles, and websites, and is exposed to a wide variety of language structures and styles to allow it to generate a human-like response. The variety of prompts are as limitless as your imagination; you can ask it to provide five different recipes for dinner, come up with a bedtime story for your child, or design a title for your manuscript about headache medicine and AI. There are reports of students inputting their homework assignments and exam-style questions, and ChatGPT has passed all three United States Medical Licensing Examinations (Step 1, Step 2 Clinical Knowledge, and Step 3).[4]
ChatGPT has already invaded the realm of academia and manuscript writing. Researchers tasked ChatGPT to provide 50 medical research abstracts based on a selection of articles published in The New England Journal of Medicine, The Lancet, Nature Medicine, JAMA, and The BMJ. The responses were then compared to their original abstracts through the use of a plagiarism detector, an AI output detector, and a group of medical researchers. The plagiarism checker judged the abstract to have a median originality score of 100% and the AI output detector was able to identify 66% of the abstracts as AI-generated. The human reviewers were only able to spot 68% of the AI-generated abstracts, and incorrectly identified 14% of the genuine abstracts as being AI generated. Many academic institutions and societies have begun to combat language modeling software, such as the Fortieth International Conference on Machine Learning banning any papers and abstracts written by AI language tools such as ChatGPT.[5] Several articles and abstracts have been published with ChatGPT credited as a coauthor. In response, multiple scientific journals such as Nature, JAMA, and Science (and their associated journals) have implemented policies stating that AI, language models, and machine learning entities do not qualify for authorship.[6] Additionally, Science prohibits submissions from including any text generated by an AI model.[7]
An AI model such as ChatGPT has great potential in headache medicine, for both clinicians and patients. I have used ChatGPT to create or augment, with proofreading, a variety of educational materials for patients, from explanations of various headache pathophysiology to the mechanism of actions and side effects for medications, and even helping to design a headache diary. I have also applied ChatGPT to assist in writing prior authorizations and appeal letters for rejected treatments. In addition to providing resources for clinicians, ChatGPT can be used by patients themselves. One of the patients I care for used ChatGPT to explain to their young child how topiramate treats their migraine (Figure 1).
Figure 1.
A screenshot of ChatGPT when given the prompt "explain how topiramate treats migraine to a child".9 [Color figure can be viewed at wileyonlinelibrary.com]
While ChatGPT and other AI models appear exciting and provide revolutionary changes to our society, there are many important concerns as well. Transparency and validation are critical components for these models, otherwise the responses provided can be obscure and imprecise. ChatGPT does not provide references or citations with its responses; therefore, it is uncertain what information is used to generate its responses. I asked ChatGPT to write up a summary of the epidemiology of migraine in penguins. It wrote a short manuscript, Migraine Under the Ice: Understanding Headaches in Antarctica's Feathered Friends, that describes the prevalence of migraine in penguins as higher in males than females, with the peak age of onset at 4–5 years, and emperor and king penguins being more prone to developing migraine than others. At the time of writing, there had been no published studies on migraine in penguins (a PubMed search consisting of the syntax "penguin" AND "migraine" was performed with no related results). This response epitomizes the danger that responses from ChatGPT not only could be wrong, but the AI model could generate and confabulate incorrect information.
To expand and moderate AI-assisted medical research, research guidelines such as CONSORT (Consolidated Standards of Reporting Trials) and SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) have added AI-specific additions. These additions include measures such as providing that the AI model has been validated, plus minimum requirements for a dataset and infrastructure, and transparency of the layers and mechanisms of the AI model.[8]
AI has rapidly evolved over the years, becoming more sophisticated and versatile. We have seen an explosion in the use and services of AI models, with ChatGPT being featured in the media on a near daily basis. Patients and I have found a variety of headache-related educational uses for ChatGPT. I predict that AI will be an integral part of medicine, including within headache medicine; however, we should be cautious or we may find ourselves in a precarious situation. If we are to use AI models, their machinations and mechanisms must be transparent, validated, and continuously reviewed and updated. I am excited to see what the future of AI and headache medicine have in store.
Abbreviations
AF, atrial fibrillation; AI, artificial intelligence; CONSORT, Consolidated Standards of Reporting Trials; ECG, electrocardiograms; LLM, large language model; SPIRIT, standard protocol items: recommendations for interventional trials.
Headache. 2023;63(5):694-696. © 2023 Blackwell Publishing
