IN THE DYNAMIC WORLD OF TODAY, artificial intelligence (AI) is becoming more sophisticated and gaining prominence. AI algorithms and methodologies enhance and self-adjust, enabling independent decision-making and task execution. Common examples of AI are virtual assistants such as Amazon’s Alexa and Apple’s Siri, recommendation systems used on platforms such as Netflix and Amazon, and social media algorithms such as those employed by Facebook. Additionally, more brazen ventures such as an endless AI-generated Seinfeld parody episode have been produced.1
AI has been employed in the medical field, with one of its more publicly notable uses being aiding radiologists in disease detection through AI-powered software.2 There have been several more recent examples of AI being used in medicine; deep neural networks have been demonstrated to have a greater positive predictive value in detecting arrhythmia from electrocardiograms (ECGs) than cardiologists.3 Another study has demonstrated neural networks to be effective in assessing 1-year all-cause mortality rates from ECGs.4 Deep learning models were employed to review video footage of laparoscopic cholecystectomies, resulting in models to aid surgeons in locating anatomical landmarks and establishing areas to dissect and not to dissect.5 Recently, a study evaluated the diagnostic precision of a computer-based diagnostic engine and determined that its outcomes were comparable to the results achieved by headache specialists.6
Numerous opportunities exist within medicine for AI to assist and support in a field so heavily overburdened. AI can help triage patients in an emergency department or help with contact tracing during a pandemic or epidemic. Models can be employed to aid in drug dosage optimization, such as in patients with renal impairments. Deep learning models can be utilized in early disease detection, such as identifying early signs of heart failure. Besides aiding in clinical decisions, AI can be employed for routine tasks traditionally burdening providers, like billing, patient scheduling, and prior authorizations.
Prior authorizations are a common hurdle patients face when prescribed new therapies and medications. Generally, certain requirements, such as specific clinical presentations, failed treatments, and contraindications to other therapies, must be documented. Although designed to aid health insurance companies in cost-effective care, prior authorization can lead to care delays because the approval process can be lengthy. Delays due to prior authorization have been a proverbial albatross not only in neurology but across various specialties.7-9 This is frequently encountered in headache medicine because of the emergence of numerous new treatments such as calcitonin gene-related peptide monoclonal antibodies and antagonists.10 Proposals have been suggested to implement AI in electronic health records to help facilitate prior authorizations.11 A recent retrospective analysis assessed 5 various machine learning algorithms expediting post–acute care discharge dispositions in an inpatient setting. The χ2 automatic interaction detector (CHAID) algorithm was shown to enhance early predictions of the post–acute care discharge process, projecting an average reduction of 22.22% in inpatient length of stay.12
In cases where prior authorization is denied, AI can generate appeal letters tailored to individual patient insurances to help overturn denials. This could streamline the process, saving providers valuable time that would otherwise be spent on appealing denials and scheduling peer-to-peer meetings with patients’ insurance, tasks known for their time-consuming nature. Likewise, this could also be implemented on the insurance company’s side, reducing administrative tasks and overhead costs.
AI’s potential is not limited to aiding providers; it can also be harnessed to close the gap between patients and providers. There are various scenarios where AI can assist in areas where patients need help. Patients frequently submit questions to clinical practices via texts, emails, and patient portals. Response times vary, and patients often anticipate an immediate reply. AI can assist in delivering prompt, precise answers to frequent questions and inquiries. AI can assess whether refills are suitable for low-risk medications, leading to quicker refill times. Reducing these tasks will help providers by lowering the volume of administrative tasks.
- AI's Growing Role in Medicine: AI, like deep neural networks, is proving its worth in medical applications. It is enhancing diagnostic accuracy, aiding in surgery, and even handling administrative tasks like billing and prior authorizations.
- Streamlining Prior Authorizations: Prior authorizations, often causing delays in patient care, could benefit from AI. It can expedite approval processes and generate tailored appeal letters when authorizations are denied, saving time and reducing administrative burdens.
- AI Empathy and Patient Advocacy: AI not only assists providers but also bridges communication gaps between patients and doctors. It can offer rapid and empathetic responses to patient inquiries, translate medical jargon into understandable language, and enhance patient-provider interactions.
A scenario patients frequently face is deciding when to ask their doctors for a further detailed explanation of a diagnosis or to clarify a complicated medical term. Some patients might resort to search engines or look for scientific articles, potentially causing misunderstanding. Generative AI and large language models (LLMs) can help translate complex medical terms into simpler, more comprehensible language. An LLM is a deep learning algorithm capable of recognizing, translating, predicting, summarizing, and producing text responses based on given prompts. It is trained on a diverse data set spanning books, articles, and websites, acquainting it with a broad spectrum of language patterns and styles, enabling it to create humanlike replies.13 A previous published example was the use of ChatGPT, a popular LLM, to explain to a child aged 5 years how topiramate works in treating migraine.14
Doubts have been raised about AI in health care settings, questioning whether it can match a human’s capacity for understanding and empathy. Although medicine is an art and AI may not fully replace providers, AI has been shown to have the ability to project empathy when patients require it. A recent study selected 195 questions obtained from social media and responses from a physician and ChatGPT. The responses were randomized and given to blinded licensed health care professionals from a variety of specialties such as pediatrics, internal medicine, and geriatrics. The evaluators preferred ChatGPT’s responses to the physicians’ responses 78.6% of the time, rated ChatGPT’s responses as being of significantly higher quality than the physicians’ responses, and found ChatGPT’s responses to be more empathetic than those of the physicians.15
The rapid advancement of AI technologies is affecting various aspects of society, the medical field being no exception. In fields like radiology, AI has become commonplace. Numerous other medical fields have implemented a multitude of AI research models to improve and enhance workflow. AI has great potential to assist and expedite clinical and administrative tasks such as prior authorizations and appeal requests. Furthermore, AI could revolutionize patient advocacy by bridging communication gaps between patients and providers. Moving forward, the synergy of medicine and AI offers a promising avenue for enhanced patient care and operational efficiency.
1. Winslow L. AI-generated Seinfeld-like Twitch ‘TV show’ is peak absurdity. Kotaku. February 1, 2023. September 13, 2023. https://kotaku.com/twitch-nothing-forever-ai-generated-seinfeld-dalle-show-1850061075
2. Moawad AW, Fuentes DT, ElBanan MG, et al. Artificial intelligence in diagnostic radiology: where do we stand, challenges, and opportunities. J Comput Assist Tomogr. 2022;46(1):78-90. doi:10.1097/RCT.0000000000001247
3. Hannun AY, Rajpurkar P, Haghpanahi M, et al. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat Med. 2019;25(1):65-69. Published correction appears in Nat Med. 2019;25(3):530.
4. Raghunath S, Ulloa Cerna AE, Jing L, et al. Prediction of mortality from 12-lead electrocardiogram voltage data using a deep neural network. Nat Med. 2020;26(6):886-891. doi:10.1038/s41591-020-0870-z
5. Madani A, Namazi B, Altieri MS, et al. Artificial intelligence for intraoperative guidance: using semantic segmentation to identify surgical anatomy during laparoscopic cholecystectomy. Ann Surg. 2022;276(2):363-369. doi:10.1097/SLA.0000000000004594
6. Cowan RP, Rapoport AM, Blythe J, et al. Diagnostic accuracy of an artificial intelligence online engine in migraine: a multi-center study. Headache. 2022;62(7):870-882. doi:10.1111/head.14324
7. Wallace ZS, Harkness T, Fu X, Stone JH, Choi HK, Walensky RP. Treatment delays associated with prior authorization for infusible medications: a cohort study. Arthritis Care Res (Hoboken). 2020;72(11):1543-1549. doi:10.1002/acr.24062
8. Wirrell EC, Vanderwiel AJ, Nickels L, Vanderwiel SL, Nickels KC. Impact of prior authorization of antiepileptic drugs in children with epilepsy. Pediatr Neurol. 2018;83:38-41. doi:10.1016/j.pediatrneurol.2018.03.006
9. Choi DK, Cohen NA, Choden T, Cohen RD, Rubin DT. Delays in therapy associated with current prior authorization process for the treatment of inflammatory bowel disease. Inflamm Bowel Dis. Published online January 30, 2023. doi:10.1093/ibd/izad012
10. Cohen F, Yuan H, Silberstein SD. Calcitonin gene-related peptide (CGRP)-targeted monoclonal antibodies and antagonists in migraine: current evidence and rationale. BioDrugs. 2022;36(3):341-358. doi:10.1007/ s40259-022-00530-0
11. Lenert LA, Lane S, Wehbe R. Could an artificial intelligence approach to prior authorization be more human? J Am Med Inform Assoc. 2023;30(5):989-994. doi:10.1093/jamia/ocad016
12. Choudhury A, Perumalla S. Using machine learning to minimize delays caused by prior authorization: a brief report. Cogent Engineering. 2021;8(1):1944961. doi:10.1080/23311916.2021.1944961
13. Singhal K, Azizi S, Tu T, et al. Large language models encode clinical knowledge. Nature. 2023;620(7972):172-180. doi:10.1038/s41586-023-06291-2
14. Cohen F. The role of artificial intelligence in headache medicine: potential and peril. Headache. 2023;63(5):694- 696. doi:10.1111/head.14495
15. Ayers JW, Poliak A, Dredze M, et al. Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Intern Med. 2023;183(6):589-596. doi:10.1001/ jamainternmed.2023.1838