A branch of AI called generative AI, and in particular ChatGPT, a large language model, has taken the world by storm since its launch in November 2022. Some experts have compared this innovation to the most significant technological milestones of the past, such as the development of the personal computer, the internet, the transition to mobile, and the cloud. Others have called it the cornerstone of the fourth industrial revolution.(1) This fourth revolution has yet to be accelerated by other similarly advancing technological developments including robotics, blockchain technology, the internet of things, genetic engineering, and quantum computing.
Artificial intelligence (AI), first conceptualized in the mid-20th century, is a field of computer science that aims to create machines capable of intelligent behavior. The term “artificial intelligence” was coined by John McCarthy in 1956, marking the beginning of AI as a formal academic discipline.(2) Generative AI is a new branch of AI that is capable of generating new content, including text, audio, and images, by learning from vast data sets.
ChatGPT, developed by Open AI, belongs to a branch of generative AI called large language model (LLM). Large language models are designed to understand, generate, and interact with human language at a sophisticated level.(3) These models are trained on vast data sets encompassing a significant portion of the internet, enabling them to have a deeper understanding of language nuances and to perform a wide range of natural language-based tasks, from translation to content creation, revolutionizing how machines understand and generate human language.
The adoption of ChatGPT has been remarkably rapid. The speed of the adoption becomes clear when compared to previous, highly popular tech platforms. In 2010, it took Instagram approximately 2.5 months to accumulate its first 1 million users. It took Spotify almost half a year to reach 1 million users. ChatGPT reached 1 million users in only five days, and it garnered 100 million users within two months.(4)
Following ChatGPT’s success, other generative AI platforms have quickly emerged on the scene with breakneck speed, each with their own unique attributes. These include Google’s Bard (now rebranded as Gemini), Meta’s Llama, X’s (formerly Twitter’s) Gork, Anthropic’s Claude II, Inflection’s PI, and multiple other open-source platforms.
While these LLMs have been impressive in text generation, image generation platforms such as Midjourney, Dalle 3, and Stable Diffusion have also made impressive strides in visual AI. More recently, Google’s Gemini and Apple’s Ferret have landed with the ability to process text, image, and video inputs within a single interaction.
THE IMPACT OF AI ON HEALTHCARE
The impact of AI on healthcare has been studied for several years. Anecdotally, physician opinions on AI range from fears that AI will replace the physician workforce, to hopes that the inefficiencies and administrative workflows of medicine will be optimized by AI, to optimism that physician decision-making and personalized medicine will be perfected with AI.
Studies have shown that AI-based healthcare interventions can be cost-effective.(5) Surveys of physicians have indicated a generally optimistic stance toward AI in medicine, highlighting eagerness to adopt despite acknowledging potential disadvantages and challenges.(6) One study involving 54 healthcare stakeholders revealed a broad spectrum of opinions on AI’s impact on clinical skills, reflecting diverse views on the values driving healthcare work.(7)
In this article, I make a pragmatic prediction and recommendation about how physicians and leaders should welcome the safe and stepwise integration of AI into clinical medicine in the coming year. My prediction distinguishes two types of healthcare AI uses: clinical and operational AI.
Clinical AI is AI that participates directly in patient care by analyzing medical data, recognizing patterns, providing clinical recommendations, and personalizing treatment plans. Operational AI, on the other hand, automates organizational tasks and workflows, synthesizes large information into concise summaries and insights, and improves the inefficiencies of healthcare delivery without participating in clinical decision-making.
There are several unanswered questions about the integration of clinical AI into patient care. For one, the margin for error is almost zero. There are also several valid concerns regarding the accuracy, explainability, universal applicability, and ethics of clinical AI.
It will take time for patients, physicians, payers, and regulators to fully understand, measure, and accept clinical AI into practice. Before these tools can be safely integrated into healthcare, research and the development of robust oversight mechanisms are necessary to ensure their accuracy and reliability.(8)
The downside of generative AI is that it is by design a black box in how it works. There are no algorithms that can be explained and measured. Generative AI is trained without having to label all the training data in advance. This has led generative AI to be prone to the inherent biases and errors that exist in the data that it is trained upon, which is mostly the worldwide web. These are serious concerns that will take time to resolve.
To date, the greatest advances in clinical AI have been in the areas of radiology and pathology image interpretation. The FDA has developed a regulatory framework for the clearance of AI and machine learning (ML)-based medical devices, focusing on safety, efficacy, and the software’s ability to improve patient care. This pathway assesses AI tools based on their intended use, risk, and accuracy, facilitating the integration of innovative technologies in healthcare. Given this framework, it is my belief that image interpretation will see quicker clinical implementation, which is further discussed below.
The ability of generative AI to organize data, facilitate workflows, and improve efficiency is quite remarkable and rapidly evolving. Operational AI can create new levels of efficiency and throughput in the delivery of healthcare and bring much-needed enhancements to the redundant and cumbersome inefficiencies that exist in the practice of medicine. My prediction is that as physicians develop familiarity and comfort with operational AI in the short term, we will be better equipped to navigate and integrate clinical AI into our daily practice at the right time.
Over the past 12 months, my adoption of operational AI tools has drastically improved my clinical, academic, and administrative productivity. Tools that streamline literature searches and provide summarization have enhanced my interactions with medical literature. AI has streamlined my meeting organization, quickly transforming meeting transcriptions into concise meeting minutes and action items. Generative AI’s ability to improve communications, organize email, and develop more eloquent and clear announcements by understanding nuances has streamlined many interactions.
My adoption of operational AI has required new learning and frequent experimentation in areas of prompt engineering, using chat-bots, using plug-ins, and understanding GPTs (general-purpose technologies). The learning curve has not required significant investment of time for the results it garners.
Most of the AI tools that I use are on third-party cloud platforms, are free or require monthly subscriptions, and do not require significant capital investment. The more I use these tools, the more rapidly I increase my understanding and ability to perform more complex tasks without needing to learn coding.
AI TOOLS TO REVOLUTIONIZE HEALTHCARE
I predict that the seven AI tools described below will revolutionize healthcare operations by 2025. As a physician engaged daily in the delivery of quality, safe, and best healthcare outcomes, as well as optimization of healthcare processes, I highlight here readily achievable areas for industry to engage with healthcare systems and with physicians to develop practical and quickly monetizable healthcare AI tools.
As the AI industry tries to figure out the healthcare AI market, custom tools targeted on specific task performance to monetize healthcare integration are being developed. Large amounts of money and human capital are being invested to solve specific challenges; thus, physician leaders need to communicate what they need and how they will use these AI tools.
Customized operational AI tools act as a super-resident, fellow, and/or administrator, available 24/7. These tools are aligned with the view of AI as augmented intelligence, suggesting a synergistic model where AI complements and supports medical professionals. These tools will have the capability to understand the domain of the sector and continually improve their usability and performance.
1. Streamlined patient encounters and rounds.
Today, documentation occupies 25-50% of a doctors’ time, yet much of the information collected is of dubious or unproven value.(9) Poorly designed EMRs, increasing complexity of comorbidities, and short time allocations for patient visits have all lead to the dissatisfaction of patients and providers.
A study published in 2020 showed that clinic visit times are now one-fifth of what they were decades ago. Clinicians may order diagnostic studies and imaging as a substitute for face-to-face time, as it is seen to save time and increase relative value units (RVUs). The medical interview is abbreviated, and the physical examination is disappearing. This is occurring at the expense of the physician-patient relationship.(10)
Clinical documents have a standardized structure that makes it feasible for AI to optimize patient and physician interaction by automating workflows pre-visit, during the visit, and after the visit.
Before the visit, chatbots can perform triage and execute intelligent scheduling much more efficiently than is done by virtual platforms today. Patients can complete automated intake forms that dive deep into the history of present illness and AI can trigger a targeted review of systems.
At the time of the visit, real-time medical record summarization focused on the chief complaint can expedite physician chart review. Note taking can be automated by AI, and real-time translation AI tools such as those available on new Android phones can enhance language and cultural sensitivity. Post-visit, follow-up instructions and education can be customized by the AI and targeted for the patient’s level of interest and comprehension.
Early studies on using large language models for improving the patient experience are showing positive results. One study looked at the ability of ChatGPT to generate natural language text, summarize large amounts of data, and answer specific questions in the area of cardiovascular nursing and allied health, and it concluded that the quality of the generated text was hard to distinguish from a text written by humans.(11)
Nuance and other companies are currently testing AI for listening to patient-provider interactions and writing visit notes. The work of the physician will eventually be to read a well-organized summary, review recommendations and reminders from the AI, make informed clinical decisions, and have ample time to communicate with the patient.
2. Automated chronic disease management and follow-ups.
Chronic disease management refers to an integrated care approach to managing illness that includes screenings, check-ups, monitoring and coordinating treatment, and patient education. This remains a significant challenge for patients in terms of the logistics of keeping multiple appointments, understanding disease processes, managing medication, and following through. For clinicians, appointment scheduling and reminders, dissemination of test results, education, and counseling still require a significant amount of human effort and current virtual tools have many limitations.
Using AI for chronic disease management is not new, but its use to date has not been free of the need for active human engagement. Currently, AI’s knowledge of the disease state is based on learning algorithms with broad categorizations that omit specificity and may be limited by potentially biased sampling.(12)
AI and specifically large language models can analyze vast medical data sets and patient records that streamline and automate the process. LLMs can identify risk factors and predict disease progression and alert physicians, recommend customized lifestyle changes based on individual preferences, and provide culturally sensitive support.
Individualized AI tools, called GPTs, can be used to provide customized ongoing support as they learn what interventions a patient responds to. Babylon Health, an AI company, announced that it will invest $100 million to double its team of engineers and scientists and expand its capabilities to manage chronic diseases by interacting with the patient through a smartphone app, do a health assessment, and then create a personalized treatment plan, monitoring progress through multiple connected devices.(13)
Another feature of AI is its ability to automate a series of tasks by using what are called application programming interfaces or APIs. These APIs can be developed using natural language without the need for coding. Thus, physician offices can customize their workflows to call patients, send reminders, respond to emails, and provide information.
In addition, patients can interact with provider-approved chatbots to interpret their test results and plan subsequent steps, allowing the physician to concentrate on higher-risk patients. Wearable devices with healthcare data, which are increasingly popular, can also be integrated into chronic disease management. One study showed that wearable device monitoring combined with interpretable machine learning can objectively track clinical progression in a cardiac rehabilitation program.(14)
3. Simplified tools for staying up to date with the medical literature.
More than 1 million biomedical articles are published each year. Focusing on the most impactful updates in one’s area of expertise has become increasingly challenging. Major specialty societies have taken on the responsibility of developing guidelines that synthesize current evidence; however, these guidelines are sparce, slow to incorporate new literature, and cumbersome to put together. With the rapid development of personalized AI assistants, the development of such guidelines will be expedited.
One study that explored AI-assisted literature search in performing targeted literature reviews, systematic reviews, as well as answering specific questions demonstrated that AI categorized pertinent information efficiently in less time compared with manual review and extraction.(15)
New tools have recently emerged that use AI to enhance AI-assisted literature reviews. Elsevier’s Scopus has more than 94 million records and claims to improve search results, identify trends in research or emerging topics, and accelerate research using natural language interactions.
Consensus, a specialized GPT on the Open AI platform, contains 200 million academic papers and can act like a custom research assistant with the ability to analyze and cite multiple publications. Physicians will soon be able to curate their library of articles with the support of AI and develop summary recommendations or guidelines on the topics of their practice and interest.(16)
4. Reduction of medical errors.
Medical errors are a serious public health problem and a leading cause of death, with at least 0.7% of adult admissions involving harmful diagnostic errors, corresponding to approximately 249,900 harmful diagnostic errors yearly.(17) These errors are caused by a mix of factors in the processes of diagnosis, including both cognitive and system-related factors.
AI can reduce simple errors such as errors of medication reconciliation and delays in reporting of critical diagnostic tests by improving appropriate data integration and communication. Most errors, however, have complex causes involving multiple factors. These require developing an error-reporting and investigation system to mitigate future medical errors. Multiple studies have identified that if error-prone situations are reported and managed by a modification of the system, the frequency of the error and concomitant errors associated with it will decrease.(18)
Investigating errors and developing accurate interventions are cumbersome tasks that require reviewing charts, understanding contexts, and assessing if existing policies were followed or if new policies are needed.
Teasing out human factors as causes of the error versus system errors also requires the ability to understand context. Large language models’ ability to understand context will eventually lead to an automated methodology for error investigation. Error reports can be automatically categorized based on emergency, severity, and classifications. AI can also support evidence-based safety measures by analyzing vast medical literature based on identified root cause.
5. Empowered bedside image interpretation.
Visual AI, a branch of artificial intelligence that focuses on the analysis and interpretation of visual data, has already made significant improvements in the clinical applicability of AI at the bedside.
While most of this article emphasizes operational AI, I predict that visual AI will stand out with earlier implementation of clinical AI, as the FDA has developed a pathway for FDA clearance of AI-enabled tools. Noteworthy FDA-cleared AI applications include IDx-DR for detecting diabetic retinopathy, Caption Health’s software for cardiac ultrasound analysis, and Paige.AI for cancer detection in biopsy slides.
One study showed that in a large representative data set of mammography images from the United States and the United Kingdom, AI demonstrated an absolute reduction of 5.7% and 1.2% (U.S. and U.K., respectively) in false positives and 9.4% and 2.7% in false negatives in accurate interpretation. In an independent review comparing the AI system to six radiologists, the AI system outperformed all of the human readers.(19)
While AI making independent clinical decisions is not ready for implementation, AI-augmented clinical image interpretations that are done in the office and at the bedside are already showing promise. AI tools that interpret radiology images, ECGs, and point-of-care ultrasounds are already available. Graphic data such as EEGs, sleep studies, ventilator waveforms, doppler studies, obstetric ultrasounds, and others will soon see preliminary readings by AI that will augment physician interpretation.
A study that compared the expertise of seven emergency physicians and three cardiologists against AI on interpretation of 187 ST elevation myocardial infarction (STEMI) ECGs demonstrated that image-based, an AI system outperformed clinicians in STEMI diagnosis and its performance was robust to change in image acquisition conditions.(20)
6. Simplified quality improvement initiatives.
Healthcare is increasingly an environment of value-based purchasing. Healthcare providers and healthcare systems must monitor, improve, and publicly report their quality metrics, and the outcomes affect their public profiles and reimbursements. In the past two decades, healthcare systems have made strides in organizing significant amounts of data, measuring quality metrics, and building dashboards to track improvements. However, the skills required, such as data analytics, project management, and change management, are not part of typical medical training.
Quality metrics are also dependent on appropriate documentation of diseases, comorbidities, and severity of illness, which utilize specific terminologies that are not always transparent for physicians. All of these create an environment where the interpretation of the data and the development of actionable insights toward meaningful quality improvement remains a challenge.
As LLMs are integrated into EMRs, AI can prompt appropriate documentation of diagnosis, comorbidities, and severity of illness so that hospitals can improve the accuracy of their quality metrics as well as mortality rates. In addition, data pulled into quality dashboards can easily be interpreted by front-line physicians using natural language with the use of chatbots.
LLMs can also improve prediction of avoidable patient safety indicators (PSIs) by identifying high-risk patients and prompting proven interventions. Using such AI tools will also bring financial rewards by reducing length of stay, unintended outcomes, and readmissions.
Implementation of quality improvement interventions can also be guided by AI using intelligence-enabled project management, which has been shown to be considerably useful in project management and for enhancing the planning, measurement, and uncertainty performance domains by providing promising forecasting and decision-making capabilities.(21)
AI can analyze and suggest appropriate quality interventions by reviewing published literature from other centers focused on similar challenges and predicting pitfalls that may be encountered.
7. Streamlined investigator-initiated research projects.
While large randomized controlled trials and meta-analysis will remain the highest quality of clinical evidence, the percentage of clinical medicine directly based on evidence from randomized controlled trials is relatively low, especially in fields like surgery.(22)
Investigator-initiated research studies play an important role in answering smaller clinical questions, measuring the impact of care modifications in local settings, and developing hypothesis for larger studies. Designing research projects and analyzing the results are a challenge for busy clinicians burdened by their clinical tasks.
AI can simplify and expedite several of the steps along the research pathway: literature review, research design, data aggregation, and statistical analysis. Current LLMs can also be useful in scientific writing and editing a manuscript, especially for non-native English speakers; they can also help the author accurately format the references.(23)
Some authors have started using rapid meta-analysis (RMA), which balances a quick time to production with reasonable data quality assurances, leveraging AI to strike this balance. They have shown that they were able to generate meaningful clinical insights in a much shorter time than traditional meta-analysis.(24) A platform called Laser AI claims to enable faster AI-based systematic literature reviews and even maintain living reviews that handle updates with ease.
PRE-REQUISITES TO IMPLEMENTATION
AI-based Improvements in healthcare delivery will be a welcome change. However, there are several prerequisites that must be carefully weighed by healthcare systems for this to become a reality in a short time.
It is imperative to remember the lessons learned from the implementation of other healthcare technologies of the recent past, most notably electronic medical records (EMRs). Locking in with a single vendor without the flexibility to cohort different vendor tools and the inability to personalize and optimize clinical tools have been grave mistakes of the past decade with immense consequences.
In the past 10 years, most healthcare systems have liberated their data by developing data warehouses that enable the internal control of data and the development of data-based metrics and intelligence. Similarly, AI implementation should be loosely affiliated with a vendor and measured by the delivery of reliable, progressive, and impactful outcomes.
The implementation of AI within healthcare delivery can follow multiple technical paths, and maintaining flexibility is important. Some healthcare systems may choose to plug in their data outside of the EMR with third-party cloud-based AI services, which avoids extensive in-house development or hardware investment. Others may integrate generative AI directly into the EMR, allowing real-time insights and recommendations. Large health systems may choose to develop their own AI solutions tailored to their specific needs enabling customizations, which will require significant resources.
While these business decisions work themselves out and cost benefit analysis is evaluated, it is imperative that systems maintain an ecosystem open to multiple modes of implementation.
It is also important that healthcare IT departments quickly incorporate AI-knowledgeable staff. They should make these experts available to front-line physicians and other providers so that clinicians can develop new ways of using and implementing AI in their daily practice.
As the advantages of operational AI increasingly contribute to the enhancement of healthcare delivery, physicians and patients will gain a deeper understanding of its functionality. This enriched comprehension facilitates the navigation of more complex transformations associated with clinical AI, allowing these advancements to be approached from a more enlightened and informed perspective.
The benefit of operational AI is that it is not bound to the EMR and it is not at risk for liability as clinical AI. Third-party plug-ins and web-based applications for operational AI are quickly getting developed. Productivity tools such as Microsoft Office, Google Workspace, and Slack have implemented AI operational tools called Co-Pilot, Duet, and Slack AI, respectively.
Leaders should make tools such as these available to physicians so they can use AI within familiar workflows. The use of LLMs and targeted AI assistants such as GPTs should be encouraged. Local and national conferences should include AI topics and interest groups should be developed to nurture the early adopters and garner common growth.
CONCERNS AND CAUTIONS
Currently, there are several reasons for skepticism toward clinical AI, including concerns about the accuracy and reliability of these technologies in making complex clinical decisions and fears that they may not always capture the nuances of patient care. Additionally, there’s apprehension about the potential for AI to disrupt traditional patient-physician relationships, as well as worries that an overreliance on technology could depersonalize care and erode the art of medicine.
Effective governance is a prerequisite to precisely address regulatory, ethical, and trust issues while advancing the acceptance and implementation of AI.(25) Several institutions such as the National Academies of Sciences, Engineering, and Medicine (NASEM), the Food and Drug Administration (FDA), and the Centers for Medicare and Medicaid Services (CMS) have produced documents addressing ethics of AI. But a clear absence of legal guidance at this time makes physicians reluctant to adopt clinical AI because of concerns about liability, data privacy, and ethical implications.
In addition to the legal risks, adopting clinical AI too soon has numerous other unanswered challenges. Technically, bias, security concerns, interoperability, reliability, and accountability need to be understood. Ethically, informed consent, equity of access, human oversight, impact on the patient-physician relationship, and the meaning of the definition of purpose need to be clarified. There are also practical risks of understanding cost, need for skill set, and infrastructure cost that need to be answered.
LOOKING FORWARD
Crucially, physicians must have a decisive voice in shaping both operational and clinical AI’s role in healthcare. Equally important is the demand for transparency from the tech industry, ensuring medical stakeholders are informed and involved from the beginning. This collaboration is essential for harnessing AI’s potential in transforming healthcare while preserving its core human values.
Generative AI is on its way toward a higher level of machine expertise called artificial general intelligence (AGI), which will have the same comprehensive capabilities as the human mind. This will present even more tremendous opportunities and dire risks.
With sage preparation and foresight, and a step-wise examination and clear understanding, AGI could usher in an even more unparalleled era of insight and invention for the betterment of all people. But without adequate safeguards and alignment, its disruptive potential could prove catastrophically destabilizing.(26)
References
Patrick E. Artifical Intelligence: Paving the Way for the Fourth Industrial Revolution. Cryptopolitan. msn.com. https://www.msn.com/en-us/news/technology/artificial-intelligence-paving-the-way-for-the-fourth-industrial-revolution/ar-AA1jHia0 .
Buchanan BG. A (Very) Brief History of Artificial Intelligence. AI Mag. 2005;26(4).
Brown TB, Mann B, Ryder N, et al. Language Models Are Few-Shot Learners. In: Advances in Neural Information Processing Systems. NeurlPS 2020:33.
Teubner T, Flath CM, Weinhardt C, et al. Welcome to the Era of ChatGPT, et al. The Prospects of Large Language Models. Bus Inf Syst Eng. 2023;65:95–101. https://link.springer.com/article/10.1007/s12599-023-00795-x
Vithlani J, Hawksworth C, Elvidge J, Ayiku L, Dawoud D. Economic Evaluations of Artificial Intelligence-Based Healthcare Interventions: A Systematic Literature Review of Best Practices in Their Conduct and Reporting. Front Pharmacol. 2023;14. https://doi.org/10.3389/fphar.2023.1220950
Pedro AR, Dias MB, Laranjo L, Cunha AS, Cordeiro JV. Artificial Intelligence in Medicine: A Comprehensive Survey of Medical Doctor’s Perspectives in Portugal. PLoS One. 2023;18(9). https://doi.org/10.1371/journal.pone.0290613
Aquino YSJ, Rogers WA, Braunack-Mayer A, et al. Utopia versus Dystopia: Professional Perspectives on the Impact of Healthcare Artificial Intelligence on Clinical Roles and Skills. Int J Med Inform. 2023;169. https://doi.org/10.1016/j.ijmedinf.2022.104903
Goodman RS, Patrinely JR, Osterman T, Wheless L, Johnson DB. On the Cusp: Considering the Impact of Artificial Intelligence Language Models in Healthcare. Med. 2023;4(3). https://doi.org/10.1016/j.medj.2023.02.008
Clynch N, Kellett J. Medical Documentation: Part of the Solution, or Part of the Problem? A Narrative Review of the Literature on the Time Spent on and Value of Medical Documentation. Int J Med Inform. 2015;84(4):221–228. https://doi.org/10.1016/j.ijmedinf.2014.12.001
Drossman DA, Ruddy J. Improving Patient-Provider Relationships to Improve Health Care. Clin Gastroenterol Hepatol. 2020;18(7):1417–1426. https://doi.org/10.1016/j.cgh.2019.12.007
Moons P, Van Bulck L. ChatGPT: Can Artificial Intelligence Language Models Be of Value for Cardiovascular Nurses and Allied Health Professionals. Eur J Cardiovasc Nurs. 2023;22(7):e55–e59. https://doi.org/10.1093/eurjcn/zvad022
Bardhan I, Chen H, Karahanna E. Connecting Systems, Data, and People: A Multidisciplinary Research Roadmap for Chronic Disease Management. MIS Q Manag Inf Syst. 2020;44(1). https://doi .org/10.25300/MISQ/2020/14644
Sandle P. Babylon Healthcare Targets Chronic Disease in $100 Million Expansion. Physicians Weekly. September 13, 2018. https://www.physiciansweekly.com/babylon-healthcare-targets-chronic
De Cannière H, Corradi F, Smeets CJP, et al. Wearable Monitoring and Interpretable Machine Learning Can Objectively Track Progression in Patients During Cardiac Rehabilitation. Sensors (Switzerland). 2020;20(12):3601. https://doi.org/10.3390/s20123601
Pashos CL, Michelson M, Chow T, et al. AI Literature Search to Support Rapid Analytics in Epidemiology. Pharmacoepidemiol Drug Saf. 2022;31.
Lanzagorta-Ortega D, Carrillo-Pérez DL, Carrillo-Esper R. Artificial Intelligence in Medicine: Present and Future. Gac Med Mex. 2022;158 (Suppl 1):17–21. https://doi.org/10.24875/GMM.M22000688
Gunderson CG, Bilan VP, Holleck JL, et al. Prevalence of Harmful Diagnostic Errors in Hospitalised Adults: A Systematic Review and Meta-Analysis. BMJ Qual Saf. 2020;29(12):1008–1018. https://doi.org/10.1136/bmjqs-2019-010822
Rodziewicz TL, Houseman B, Hipskind JE. Medical Error Reduction and Prevention. Treasure Island, FL:StatPearls Publishing. 2024.
McKinney SM, Sieniek M, Godbole V, et al. International Evaluation of an AI System for Breast Cancer Screening. Nature. 2020;577(7788):89–94. https://doi.org/10.1038/s41586-019-1799-6
Choi YJ, Park MJ, Ko Y, et al. Artificial Intelligence Versus Physicians on Interpretation of Printed ECG Images: Diagnostic Performance of ST-Elevation Myocardial Infarction on Electrocardiography. Int J Cardiol. 2022;363:6–10. https://doi.org/10.1016/j.ijcard.2022.06.012
Taboada I, Daneshpajouh A, Toledo N, de Vass T. Artificial Intelligence Enabled Project Management: A Systematic Literature Review. Appl Sci. 2023;13(8). https://doi.org/10.3390/app13085014
Kao LS, Tyson JE, Blakely ML, Lally KP. Clinical Research Methodology I: Introduction to Randomized Trials. J Am Coll Surg. 2008;206(2):361–369. https://doi.org/10.1016/j.jamcollsurg.2007.10.003
Santos de Oliveira R, Ballestero M. The Future of Pediatric Neurosurgery and ChatGPT: An Editor’s Perspective. Arch Pediatr Neurosurg. 2023;5(2). https://doi.org/10.46900/apn.v5i2.191
Michelson M, Chow T, Martin NA, Ross M, Ying ATQ, Minton S. Artificial Intelligence for Rapid Meta-analysis: Case Study on Ocular Toxicity of Hydroxychloroquine. J Med Internet Res. 2020;22(8):e20007. https://doi.org/10.2196/20007
Al Kuwaiti A, Nazer K, Al-Reedy A, et al. A Review of the Role of Artificial Intelligence in Healthcare. J Pers Med. 2023;13(6):951. https://doi.org/10.3390/jpm13060951
Obaid OI. From Machine Learning to Artificial General Intelligence: A Roadmap and Implications. Mesopotamian J Big Data. 2023:81–91. https://doi.org/10.58496/MJBD/2023/012