The Rhetorical Review

Student Spotlight: Nicholas Hom, Department of Business, University of Utah, Written for WRTG 2010

The amount of time a physician spends with a patient during an appointment is important. Studies have shown that “longer physician visits are associated with more attention to psychosocial problems, lower prescribing rates, better quality prescribing, lower referral rates, lower return consultation rates,” and greater patient satisfaction (Konrad et al., 2010, p. 2). As corporations take over more and more medical practices, the focus becomes more on profits than best clinical practices. Robertson (2023) noted in a recent survey of physicians who worked for a corporation that corporate ownership had a negative impact on the physicians’ relationship with their patients due primarily to decreased time and communication with patients. Studies show that physicians experience high levels of time pressure based on the amount of time they are scheduled versus the amount of time they feel they need with patients (Konrad et al., 2010). Corporations save money by decreasing the amount of time that a physician spends with each patient, which allows the physician to see more patients in a given time period. However, this practice negatively impacts both patients and providers. As Dr. Dan Moore stated in an article by Reid Abelson (2023), “You don’t become a physician to spend an average of seven minutes with a patient” (p. 1). According to Abelson (2023), almost seven in ten doctors are employed by a hospital or corporation. Given this trend, anything that can facilitate a part of the patient care process is beneficial and allows physicians to have more face-to-face time with their patients. With the advent of artificial intelligence (AI), medicine has the potential to undergo a transformational change. In this paper, I will argue that using a collaborative approach that combines human expertise with AI capabilities in medical diagnosis, treatment, and administrative tasks can help physicians do their jobs more effectively.

“In this paper, I will argue that using a collaborative approach that combines human expertise with AI capabilities in medical diagnosis, treatment, and administrative tasks can help physicians do their jobs more effectively.”

~Nicholas Hom~

To determine how AI can help doctors with their workload, we must first review what AI is. Artificial intelligence “refers to the development and application of computer algorithms that can perform tasks that usually require human intelligence” (Pannala et al., 2020, p. 599). It includes machine learning (ML), deep learning (DL), and natural language processing (NLP) (Alowais et al., 2023). Machine learning, a subset of AI, is used in medicine primarily for identifying and classifying images. Traditional machine learning uses supervised learning to train the model to recognize patterns and predict outcomes for a specific task and requires specialists to analyze and label all the data used for training. The disadvantage of traditional machine learning is the amount of time and expense needed to create a model that can only perform one specific task (Pannala et al., 2020). The continued evolution of AI resulted in the development of deep learning, which is “a type of machine learning algorithm” that can learn from unsupervised data, using multiple layers of data to create patterns that “can be used in decision making” (Ahmad et al., 2021, p. 2). Because deep learning did not require supervised learning or labeling of all data, it was less time-consuming and, therefore, less costly than traditional machine learning. The development of foundation models, which are large deep learning neural networks trained on massive unlabeled data sets through self-supervised learning, was a big breakthrough, as this model could be trained on one task but then could be adapted to perform many tasks (Lenharo, 2023; Pannala et al., 2020).

At this time, there are two models of deep learning that are used in medicine. They are Convolutional Neural Networks (CNNs) and transformer models. The “Convolutional Neural Network is the most prominent deep learning technique” used to recognize objects and patterns in various imaging in medicine to make a diagnosis (Pannala et al., 2020, p. 599). Through Convolutional Neural Networks, AI can “analyze large data sets and identify patterns that would be difficult for humans to detect” (Alowais et al., 2023, p. 2-3). The other deep learning model that is used in medicine is transformers. Large Language models, which are transformers, use deep learning “to understand, summarize, generate, and predict new text-based content” and are designed for natural language processing, focusing “on the interaction between computers and humans” (Alowais et al., 2023, p. 2). Transformers can “differentially weigh the importance of each part of the input data,” which made natural-language processing possible (Haug & Drazen, 2023, p. 1203). The latest Generative Pretrained Transformer 4 (GPT-4) with a chat interface “uses AI and natural language processing to understand questions and to automate responses to them, simulating conversation” (Haug & Drazen, 2023, p. 12A6). As a result, the GPT-4 is also capable of “clinical expert-level medical note-taking, question answering, and consultation” (Rajpurkar & Lungren, 2023, p. 1981). This model can keep track of content in a conversation, making it more useful and natural (Lee, Bubeck, & Petro, p. A233). Although these models are only trained with source data found on the internet and do not include any private health data, they were still able to answer questions from the U.S. Medical Licensing Exam with 90% accuracy (Lee, Bubeck, & Petro, 2023, p. A234-A235). As AI continues to evolve, the possibilities of how it can help physicians do their jobs are infinite.

“As AI continues to evolve, the possibilities of how it can help physicians do their jobs are infinite.”

~Nicholas Hom~

AI has been used in a variety of medical applications. Currently, it is used primarily to assess skin lesions, interpret electrocardiograms, pathology slides, and ophthalmic images, and interpret diagnostic imaging from different imaging modalities, including X-rays, Magnetic resonance, Computed Tomography, and ultrasonography (Rajpurkar & Lungren, 2023, p. 1982). Its most useful application is to analyze large volumes of medical data and identify patterns that may be missed by the human eye and, in doing so, “help doctors make more accurate diagnoses and thus provide more targeted treatments for their patients” (King, 2023, p. 291). There are several different ways that AI can help physicians analyze lab results or images. According to Rajpurkar & Lungren (2023) in the New England Journal of Medicine, AI can help with “workflow triage” by rapidly reading scans or lab results and flagging or alerting the radiologist, pathologist, cardiologist, ophthalmologist, or dermatologist of possible positive findings, so that those cases can be reviewed right away (p. 1982).

AI can also help a radiologist or cardiologist view and interpret scans by enhancing the image through reconstruction or by diminishing image noise or artifacts (Rajpurkar & Lungren, 2023, p. 1982). Furthermore, AI can detect, localize, and classify abnormal lesions in the scans, and through its quantification algorithms, AI can segment and measure anatomical structures or abnormalities in scans (Rajpurkar & Lungren, 2023, p. 1982). For example, according to Rajpurkar & Lungren (2023), AI can measure breast density in mammograms, identify and measure breast abnormalities or pulmonary nodules, and determine cardiac flow (p. 1982).  Another way AI can help with the analysis of imaging is to “accurately predict clinical outcomes” based on CT scans “in cases of traumatic brain injury and cancer” (Rajpurkar & Lungren, 2023, p. 1982). In addition, AI is also able to determine coronary artery calcium scores through cardiac ultrasound imaging instead of CT scans, which helps reduce a patient’s exposure to radiation (Rajpurkar & Lungren, 2023, p. 1982). Finally, AI has also been used to “analyze images of the eye” to detect diabetic retinopathy and other retinal diseases (Becker, 2019, p. 201). AI has transformed medicine through its ability to analyze and interpret imaging studies.

“AI is also able to determine coronary artery calcium scores through cardiac ultrasound imaging instead of CT scans, which helps reduce a patient’s exposure to radiation (Rajpurkar & Lungren, 2023, p. 1982).”

~Nicholas Hom~

Besides its use in analyzing diagnostic imaging, AI has also been used in medicine in other ways to help physicians do their work more effectively. According to Ahmad et al. (2021), using a deep convolutional neural network, AI was able to identify skin cancers and demonstrated at least equivalent abilities when compared to “21 board-certified dermatologists” (p. 5). AI has also been used “to identify outbreaks of infectious disease that may have an impact on public health,” combine clinical, genetic, or other laboratory findings to identify rare conditions that may escape detection, as well as aid in “hospital business operations” (Haug & Drazen, 2023, p.12A4). The other primary use of AI in medicine is in gastroenterology to detect colorectal polyps. Deep learning has been used to detect colorectal polyps, classify them as adenomatous vs non-adenomatous, and then do a real-time histologic assessment to predict their invasiveness (Pannala et al., 2020, p. 599). The ability to detect whether a polyp is malignant in real-time is important because it determines the needed treatment. If the cancerous polyp is confined to the submucosal layer, it can be removed with an endoscopic resection. Still, if it is deeper, it may need surgical intervention, as there would be a higher risk of metastasis to the lymph nodes (Pannala et al., 2020, p. 605). In one study, Pannala et al. (2020) noted that with computer-aided diagnosis (CAD), the model was able to differentiate and identify invasive malignancy from nonmalignant adenomatous polyps with an accuracy of 94.1% (p. 606). Currently, through these applications, AI can help physicians do their work more effectively. As AI continues to evolve and develop, there are other ways it may be able to help physicians in the future. 

First, AI tools may soon be able to predict a tumor’s response to immunotherapy by looking at patterns between individuals who respond well to certain anti-cancer drugs and individuals who do not. Secondly, AI may be able to advance clinical laboratory testing by improving the “accuracy, speed, and efficiency of laboratory processes,” perhaps by allowing for culture and sensitivity testing to produce a result within 48 hours instead of weeks, thus enabling physicians to select the appropriate antibiotic more quickly to resolve the infection (Alowais et al., 2023, p. 4). 

Third, if AI can streamline administrative tasks that are so time-consuming, it would free physicians so that they would have more time to interact with their patients. For example, if AI can gather certain information while listening in on the physician-patient encounter to produce a chart note, generate preauthorization information, generate referrals, lab, or prescription orders, and then bill insurance automatically for treatment rendered, that could help health care providers tremendously (Lee, Bubeck, & Petro, 2023, p. A235). Rajpurkar & Lungren (2023) give another example of how AI could streamline the practice of medicine, “Given a medical image and relevant clinical information,” the model could generate “a complete radiologic report for the radiologist,” a patient-friendly report for the patient, recommendations for the best surgical approach for the surgeon, as well as “follow-up suggestions and tests for the primary care provider” (p. 1987). This is still a work in progress, but once it becomes available, it will revolutionize how physicians can care for their patients. 

A fourth way that AI could be helpful for physicians, as suggested by Alowais et al. (2023), is through the use of virtual assistants to educate patients about their diagnosis or treatment options, answer questions, provide medical advice, send reminders, schedule appointments, and monitor vital signs which can then be sent to their provider (p. 9). Although the use of AI to educate patients is still in its early stages, because so much of providing medical care involves educating patients about their condition, it is positioned to be a groundbreaking technology once it comes to fruition, further freeing healthcare providers to focus on other tasks. A fifth way that AI can help is by analyzing the images generated by wireless capsule endoscopy (WCE), where a patient swallows a capsule with a camera that takes images as it is passed through the gastrointestinal tract. According to Pannala et al. (2020), because WCE can generate “up to 8 hours of video and approximately 60,000 images in a typical examination” (p. 606), analysis of all the data generated is very time-consuming. It would be easier if AI could help facilitate the analysis by flagging abnormalities for the radiologist to review. Once again, this is a work in progress but has significant implications once it becomes available for clinical use. 

A sixth way that AI can aid medicine, as proposed by Alowais et al. (2023), is in oncology to categorize “cancers into clinically relevant subtypes” using transcriptomic profiling to help with diagnosis, prognosis, and treatment, which would allow a physician to provide precision or personalized medicine, based on the patient’s unique genetics, lifestyle, and biomarkers (p. 5). Alowais et al. (2023) further suggest that utilizing these genetic profiles could allow physicians to predict which chemotherapy drugs would be most effective to treat cancer, as well as to predict the patient’s response to a given drug, thus ensuring that they receive “the right drug, at the right dose, at the right time” to minimize side effects (p. 6).  A seventh way AI can help physicians, as put forth by Alowais et al. (2023), is by “examining extensive genomic data sets” to find intricate patterns that cannot be detected manually. Alowais et al. discuss a study that used a “deep neural network to identify genetic variants associated with autism spectrum disorder (ASD)” and then was able to use that information to predict “ASD status by relying solely on genomic data” (Alowais et al, 2023, p. 5).  An eighth way AI can help in a broader context is through medical imaging AI in places worldwide where radiologists are unavailable. This is still being evaluated, but according to Rajpurkar and Lungren (2023), “one study showed that an AI system for chest radiograph interpretation, when combined with input from a non-radiology resident, had performance values that were similar to those for board-certified radiologists” (p.1984). If an AI system could interpret an image and generate a report that a clinician who is not a radiologist could use to determine the treatment needed, it could be a game changer for areas where access to medical specialists is limited. Besides clinical applications, Lee, Bubeck, & Petro (2023) suggest that AI may also be able to help in research by reading medical research material, summarizing the content, providing analysis, identifying relevant prior work, and posing possible additional areas of research (p. A237). The possibilities are endless for AI to help physicians do their jobs more effectively.

In a collaboration between AI and physicians, AI can help physicians mitigate their areas of weakness, and physicians can ensure that AI is performing its job accurately. The key is to utilize AI for repetitive tasks prone to human error, where fatigue can affect the accuracy of the results or where pattern recognition is not possible with human eyes. Several studies show the superiority of AI in certain tasks, especially in interpreting images.

“The key is to utilize AI for repetitive tasks prone to human error, where fatigue can affect the accuracy of the results or where pattern recognition is not possible with human eyes. Several studies show the superiority of AI in certain tasks, especially in interpreting images.”

~Nicholas Hom~

Wang et al. (2020) found that up to 30% of adenomas might be missed during a routine screening colonoscopy, which would result in a higher risk of patients developing cancer in between screenings (p. 1252). Furthermore, they found that using Computer Assisted Diagnosis (CADe) AI technology reduced the overall miss rate of adenomas and especially for lesions that were less than 9mm in size (Wang et al., 2020, p. 1256). In his study, only 1.59% of visible adenomas were missed by CADe colonoscopy compared to 24.21% for traditional colonoscopy (Wang et al., 2020, p. 1256). In another study, Becker (2019) also noted that CAD assistance helped radiologists improve their diagnostic performance in interpreting three-dimensional magnetic resonance imaging scans of brain metastasis (p. 200). Sim et al. (2020) found that radiologists performed better with deep convolutional network software for detecting malignant pulmonary nodules on chest X-rays than without (p. 199). Although artificial intelligence’s capabilities are remarkable, it also has its limitations. 

AI can make mistakes. According to Lee, Bubeck, & Petro (2023) in the New England Journal of Medicine, research has shown that Chat GPT-4 can produce a false response or a “hallucination,” and these errors can be particularly dangerous when occurring in medicine (p. A233). For example, according to Becker (2019), it was found that of 2,298 electrocardiograms that AI diagnosed as atrial fibrillation, 442 of those from 382 patients were deemed incorrect upon review by two electrophysiologists (Becker, 2019, p. 201). Because AI can fabricate or create false information, data that is generated by AI needs to be closely monitored (Alowais et al., 2023, p.10). As such, although AI can be helpful, a clinician must follow up, review, and validate its findings. Although the “performance of these machine tasks is not perfect and often requires a skilled person to oversee the process…in many cases, it is good enough, given the need for relatively rapid interpretation of images…” (Haug & Drazen, 2023, p. 12A4). Neither AI nor physicians are infallible; both have limitations, but by working together and utilizing each other’s strengths, they can provide higher-quality care to patients.

“Neither AI nor physicians are infallible; both have limitations, but by working together and utilizing each other’s strengths, they can provide higher-quality care to patients.”

~Nicholas Hom~

There are other concerns regarding the use of AI in medicine. One of the primary concerns involves patient privacy. Because AI models require access to large amounts of personal information as they collect, store, and handle medical data, unauthorized access through cyberattacks could disrupt critical healthcare operations, jeopardize patient safety, and compromise patient data, resulting in fraud or identity theft (Alowais et al., 2023, p. 11). Another major concern with the use of AI in medicine is bias and discrimination. Because AI systems are only as good as the data they are trained on, if the data is biased or discriminatory, the results that the AI system produces will also be the same, which could result in bias or mistakes in its assessments, resulting in harm to patients (Ahmad et al., 2021, p. 5). There is also a concern that physicians’ reliance on and trust in AI systems could lead to the loss of their skills in those areas, resulting in “an inadequate or incorrect computer-aided diagnosis,” potentially resulting in the wrong treatment (Becker, 2019, p. 201). 

Finally, because AI models are “often not tested outside the setting where they were trained” or tested in multiple clinical settings, there is a lack of evaluation of AI systems in the real world (Rajpurkar & Lungren, 2023, p. 1984). Rajpurkar & Lungren (2023) feel that this “can pose a substantial risk to patients and clinicians” since the performance of AI systems worsens “when applied to patients who differ from those used” to develop the model (p. 1984). Their review of current commercial AI products also found that “scientific, peer-reviewed evidence of the efficacy of most commercial AI products is lacking” (Rajpurkar & Lungren, 2023, p. 1985). Furthermore, because the performance of AI models can degrade over time, continuous monitoring is needed, as well as “regular updates of the training data and retraining” to keep it current (Rajpurkar & Lungren, 2023, p. 1986). Despite these concerns regarding AI models, it is a tool that physicians can use to help them do their job better.

Physicians can do their jobs more effectively using a collaborative approach between AI and physicians. AI can be utilized in areas where physicians are less capable, such as tasks that are repetitive and tedious or require analysis of large data sets to find a pattern, as well as in administrative duties to take the burden off of them, allowing them time to focus on more complex cases or to interact with their patients. Because AI systems are not infallible and can make mistakes, physicians must follow up, review, and verify their findings. This collaboration with AI allows physicians to utilize the strengths of AI, as well as their own expertise, to provide a higher quality of care to their patients. As research progresses with AI, it will become an even more powerful tool to help physicians provide quality care to people globally, even in areas with limited resources. Once all of its capabilities are incorporated into medical practice, it has the potential to revolutionize and transform the practice of medicine.

References

Abelson, R. (2023, May 8). Corporate Giants Buy Up Primary Care Practices at a Rapid Pace. The New York Times. https://www.nytimes.com/2023/05/08/health/primary-care-doctors-consolidation.html

Ahmad, Z., Rahim, S., Zubair, M., & Abdul-Ghafar, J. (2021). Artificial intelligence (AI) in medicine, current applications and future role with special emphasis on its potential and promise in pathology: present and future impact, obstacles including costs and acceptance among pathologists, practical and philosophical considerations. A comprehensive review. Diagnostic pathology, 16, 1-16. https://doi.org/10.1186/s13000-021-01085-4

Alowais, S.A., Alghamdi, S.S., Alsuhebany, N. et al. Revolutionizing healthcare: the role of artificial intelligence in clinical practice. BMC Med Educ 23, 689 (2023). https://doi.org/10.1186/s12909-023-04698-z

Becker, A. (2019). Artificial intelligence in medicine: What is it doing for us today?. Health Policy and Technology, 8(2), 198-205. https://www.sciencedirect.com/science/article/abs/pii/S2211883718301758?via%3Dihub

Haug, C. J., & Drazen, J. M. (2023). Artificial intelligence and machine learning in clinical medicine, 2023. New England Journal of Medicine, 388(13), 1201-1208. https://www.nejm.org/doi/full/10.1056/NEJMra2302038

King, M.R. The Future of AI in Medicine: A Perspective from a Chatbot. Ann Biomed Eng 51, 291–295 (2023). https://doi.org/10.1007/s10439-022-03121-w

Konrad, Link, C. L., Shackelton, R. J., Marceau, L. D., von dem Knesebeck, O., Siegrist, J., Arber, S., Adams, A., & McKinlay, J. B. (2010). It’s About Time: Physicians’ Perceptions of Time Constraints in Primary Care Medical Practice in Three National Healthcare Systems. Medical Care, 48(2), 95–100. https://doi.org/10.1097/MLR.0b013e3181c12e6a

Lee, P., Bubeck, S., & Petro, J. (2023). Benefits, limits, and risks of GPT-4 as an AI chatbot for medicine. New England Journal of Medicine, 388(13), 1233-1239. https://www.nejm.org/doi/full/10.1056/NEJMsr2214184

Lenharo. (2023). An AI revolution is brewing in medicine. What will it look like? Nature (London), 622(7984), 686–688. https://doi.org/10.1038/d41586-023-03302-0

Pannala, Krishnan, K., Melson, J., Parsi, M. A., Schulman, A. R., Sullivan, S., Trikudanathan, G.,    Trindade, A. J., Watson, R. R., Maple, J. T., & Lichtenstein, D. R. (2020). Artificial intelligence in gastrointestinal endoscopy. VideoGIE : an Official Video Journal of the American Society for Gastrointestinal Endoscopy, 5(12), 598–613. https://doi.org/10.1016/j.vgie.2020.08.013

Rajpurkar, P., & Lungren, M. P. (2023). The current and future state of AI interpretation of medical images. New England Journal of Medicine, 388(21), 1981-1990. https://www.nejm.org/doi/full/10.1056/NEJMra2301725

Robertson, R. (2023, December 8). Corporate Ownership Worsens Patient Care, Surveyed Docs Say- Reduced autonomy, focus on financial incentives among top negative impacts cited by doctors. MedPage Today. https://www.medpagetoday.com/special-reports/features/107748

Sim, Y., Chung, M. J., Kotter, E., Yune, S., Kim, M., Do, S., … & Choi, B. W. (2020). Deep convolutional neural network–based software improves radiologist detection of malignant lung nodules on chest radiographs. Radiology, 294(1), 199-209. https://doi.org/10.1148/radiol.2019182465

Wang, P., Liu, P., Brown, J. R. G., Berzin, T. M., Zhou, G., Lei, S., … & Xiao, X. (2020). Lower adenoma miss rate of computer-aided detection-assisted colonoscopy vs routine white-light colonoscopy in a prospective tandem study. Gastroenterology, 159(4), 1252-1261. https://www.sciencedirect.com/science/article/pii/S0016508520348204

• ShareCopied to clipboard
• Facebook
• X
• LinkedIn
• Pinterest
• Threads
• WhatsApp
• Reddit
• Mail
• Bluesky
• Mastodon
• Tumblr
• Telegram
• Nextdoor
• Print