Artificial Intelligence for Improved Patient Outcomes: A Book Review

Am J Manag Care. 2024;30(1):17-18. https://doi.org/10.37765/ajmc.2024.89482

Takeaway Points

Artificial Intelligence for Improved Patient Outcomes: Principles for Moving Forward With Rigorous Science brings together the clinical and technical aspects of artificial intelligence (AI) in an easy-to-read, thoughtful, and comprehensive text.

• The book thoroughly covers all aspects of AI in health care in common language without bogging the reader down in technical details.
• This work will be a useful resource for health care leaders and AI developers seeking to implement AI tools due to its best practices “playbook” for AI development and evaluation.
• It will serve as an accessible resource for clinicians, researchers, and journal editors seeking to develop their skills as critical readers of AI research literature.

Artificial intelligence (AI) is a rapidly growing field that holds incredible promise to improve health care. Excitement for AI is high due to its potential to improve patient outcomes, reduce provider burden, optimize operations, and reduce costs. Application of AI in health care, however, is still limited, and evidence of its utility remains scarce. This can be attributed, at least in part, to misunderstanding the tool itself and its applications. Health care administrators and providers are not well versed in AI, making it seem like an enigma, and AI practitioners lack familiarity with the clinical environment, making it seem like a panacea. For AI to be successfully integrated into health care, AI developers need to focus on applications that are pragmatic and useful to health care systems and provide evidence of usefulness to clinical providers and administrators. This, however, is impossible to attain without a shared understanding between the clinical and the technical.

The book Artificial Intelligence for Improved Patient Outcomes: Principles for Moving Forward With Rigorous Science by Daniel W. Byrne brings together the clinical and technical aspects of AI in an easy-to-read, thoughtful, and comprehensive text. Byrne aims to dissolve barriers in the health care industry caused by misunderstandings about AI’s capabilities and limitations. Byrne is an established expert in this field as a senior associate in both the Department of Biostatistics and Department of Biomedical Informatics at Vanderbilt University Medical Center and director of artificial intelligence in the Advanced Vanderbilt Artificial Intelligence Laboratory.

Targeted toward AI developers and health care leaders, the book seeks to provide a comprehensive understanding of how AI can be integrated into health care workflows and address specific needs. This book offers a broad yet detailed exploration of AI in the health care context. Further, Byrne’s writing speaks to extensive experience communicating with multidisciplinary teams. The language is clear and easy to follow, and the structure of the book progresses logically. The book’s thoughtful organization greatly aids readers, whether they are delving into specific topics of interest or using the book as a comprehensive reference. Each section, broken down into well-titled chapters, facilitates easy navigation. The author enhances readability through strategic use of subheads, quotes, tables, and figures, creating a visually appealing layout that simplifies complex content. Combined with concise and clear language, this ensures readers of various academic backgrounds can easily grasp the material.

The book is structured into 5 distinct sections: “The Big Picture,” “The Mechanics,” “The Implementation,” “The Specific Applications,” and “The Future.” Within this framework, it provides readers with the information they need to consider when developing and critically evaluating AI implementations in the health care space. Beginning with “The Big Picture,” Byrne details how AI can improve health care in terms of patient safety, clinical decision support, equity and fairness, clinician work environment, and efficiency. He then explores issues with AI implementation and how to resolve them.

I was particularly pleased that Byrne discussed problems with AI evaluation. Often this does not occur at all, and when it does, the methods frequently are not robust. Byrne strongly recommends randomization as the overarching solution, calling it the “secret sauce” and highlighting it as necessary for accurate evaluation of AI’s utility. He then specifies evaluation approaches to achieve high-quality results and how to interpret them. Although I applaud Byrne’s advocating for randomization, I disagree that only results from randomized controlled trials should be believed. Observational studies, using existing clinical data, often serve as a necessary starting point for AI implementations, and this approach is entirely appropriate as long as researchers acknowledge the inherent limitations. Despite my minor disagreement with the author on this point, I found this section to be the most helpful in the whole book.

The consistent underpinning of connecting the clinical and the technical is a key factor in advancing AI in health care, yet due to the lack of common vernacular and experience between clinical and technical actors, it is rarely done. A particularly strong aspect of this section is Chapter 4, “Synergy—Building a Successful Clinician-Computer Collaboration.” Byrne notes that developers should be prepared to show the need for and expected improvements (institutional, clinical workflow, and patient outcomes) from any new AI tool. Importantly, he emphasizes that the goal of any AI tool should be to assist clinicians, never to replace or override them. Of equal importance, AI algorithms should be evaluated through an equity lens: “Be intentional about evaluating algorithms for bias, inequity, and potential harm.” Indeed, evaluating interventions in the clinical and greater societal context of their implementation is critical to using AI equitably.

Next, Byrne delves into “The Mechanics” with a high-level overview on key components of AI model development and evaluation. This includes a playbook for creating and testing an AI tool. The playbook is not a technical how-to, but rather guidelines, or a “strategic plan,” for health care leaders and AI developers to follow. Byrne goes on to address variable selection considerations, briefly describing different AI learning categories (supervised, unsupervised, and deep reinforcement) and explaining commonly used AI modeling techniques. Crucial to this section, Byrne details best practices—both formal best-practice guidelines and informal best practices. These guidelines will help health care leaders ensure that AI developers are building and evaluating AI models using the most rigorous standards.

Byrne also includes a chapter on electronic health records (EHRs), “EHRs—Exporting, Cleaning, Managing Datasets, and Integrating Models Into the Electronic Health Record.” Data management is the foundation of all EHR-based models (AI included), yet it is often omitted. Data extracted from EHRs to build and evaluate models are complex and time-consuming to transform into useful and usable data sets. This is compounded by the dynamic nature of EHRs, which are always changing (potentially affecting model function). The complexities of EHR data are difficult to understand for those who have not worked with them firsthand. Byrne explains EHR data management efficiently, hitting the main points a health care leader needs to know without needlessly bogging them down in the weeds.

After “The Mechanics” covers development and evaluation of AI models, “The Implementation” delves into how to successfully execute these models from an institutional point of view. Byrne begins this section with a chapter on resistance to AI and how to overcome it. He advocates that AI implementations be led by aligned clinicians—for example, physicians leading physician-based implementations and nurses leading nursing-based ones. Building on the principles in “The Big Picture,” Byrne describes how to effectively demonstrate the need for AI tools through an evidence-based approach. He then explains approaches to increase the odds of successful implementation, with an emphasis on team building (ie, which experts to include and in what capacity), pragmatism, and streamlining.

This leads directly into “The Specific Applications,” which goes into more granular detail about how to implement AI tools for specific purposes. The purposes include the following: complications (predicting and preventing), prevention, precision medicine, drugs and devices, medical imaging, and pandemics. Principles, which are used throughout the entire book, highlight important considerations and guidelines, and examples are provided for each purpose to demonstrate AI’s applicability in that realm. Byrne ends the book with a section titled “The Future” that contains a single chapter on building a career in AI. This section advises readers about resources such as reputable college programs, online courses, books, programming languages, and AI conferences to attend. This resource list would be most useful to non-AI experts seeking to learn more on the technical side of AI.

Overall, Byrne accomplishes connecting the clinical and the technical and providing “new, relevant, and practical information on what AI can do in health care and how to assess whether AI is improving health outcomes”—as described by the publisher—that is useful for both health care leaders and AI developers. This book would also be useful for clinicians, researchers, and journal editors seeking to develop their skills as critical readers of AI research literature. Byrne defines AI terms, placed in an easily referenced table in Chapter 3. He also describes statistical approaches for AI implementation evaluation and how to interpret results. The easy navigation of this book makes it an accessible resource for those wanting to better evaluate AI-based research publications. There are currently many books about AI, and the number is increasing as rapidly as the technology is advancing. The majority that I have found, however, are focused on either the technical aspects (eg, coding) of AI implementation or giving nonexperts a basic understanding of the topic. The uniqueness of this book is that it outlines the types of problems AI can address, a playbook of how to design AI implementations, strategies to successfully implement AI tools from institutional and end-user perspectives, and planning evaluations of their utilization—all from a dual clinical/technical perspective. 

Author Affiliation: Department of Emergency Medicine, Washington University School of Medicine in St Louis, St Louis, MO.

Source of Funding: None.

Author Disclosures: The author reports no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design; drafting of the manuscript; and critical revision of the manuscript for important intellectual content.

Address Correspondence to: Rachel M. Ancona, PhD, MS, Department of Emergency Medicine, Washington University School of Medicine in St Louis, 660 S Euclid Ave, Campus Box 8072, St Louis, MO 63110. Email: anconar@wustl.edu.

REFERENCE

Byrne DW. Artificial Intelligence for Improved Patient Outcomes: Principles for Moving Forward With Rigorous Science. Lippincott Williams & Wilkins; 2023.