Sarim Dawar Khan, Zahra Hoodbhoy, Mohummad Hassan Raza Raja, Jee Young Kim, Henry David Jeffry Hogg, Afshan Anwar Ali Manji, Freya Gulamali, Alifia Hasan, Asim Shaikh, Salma Tajuddin, Nida Saddaf Khan, Manesh R. Patel, Suresh Balu, Zainab Samad, Mark P. Sendak
Abstract
Research on the applications of artificial intelligence (AI) tools in medicine has increased exponentially over the last few years but its implementation in clinical practice has not seen a commensurate increase with a lack of consensus on implementing and maintaining such tools. This systematic review aims to summarize frameworks focusing on procuring, implementing, monitoring, and evaluating AI tools in clinical practice. A comprehensive literature search, following PRSIMA guidelines was performed on MEDLINE, Wiley Cochrane, Scopus, and EBSCO databases, to identify and include articles recommending practices, frameworks or guidelines for AI procurement, integration, monitoring, and evaluation. From the included articles, data regarding study aim, use of a framework, rationale of the framework, details regarding AI implementation involving procurement, integration, monitoring, and evaluation were extracted.
Introduction
The use of Artificial Intelligence (AI) tools has been exponentially growing, with several applications in the healthcare industry and tremendous potential to improve health outcomes. While there has been a rapid increase in literature on the use of AI in healthcare, the implementation of AI tools is lagging in both high-income and low-income settings, compared to other industries, has been noted, with fewer than 600 Food and Drug Administration-approved AI algorithms, and even fewer being presently used in clinical settings [1–4].
The development-implementation gap has been further assessed by Goldfarb et al., using job advertisements as a surrogate marker to measure technology diffusion patterns, finding among skilled healthcare job postings between 2015–2018, 1 in 1250 postings required AI skills, comparatively lower than other skilled sectors (information technology, management, finance and insurance, manufacturing etc.)
Methods Information sources and search strategy
We searched electronic databases (MEDLINE, Wiley Cochrane, Scopus, EBSCO) until June 2022. The search string contained terms that described technology, setting, framework, and implementation phase including AI tool procurement, integration, monitoring, evaluation, including standard MeSH terms. Terms that weren’t standard MeSH terms, such as “clinical setting” were added following iterative discussions. To capture papers that were methodical guidelines for AI implementation, as opposed to experiential papers, and recognizing the heterogeneous nature of “frameworks”, ranging from commentaries to complex, extensively researched models, multiple terms such as “framework”, “model” and “guidelines” were used in the search strategy, without explicit definitions with the understanding that these encompassing terms would capture all relevant literature, which would later be refined as per the inclusion and exclusion criteria.
Results
A total of 17,537 unique studies were returned by the search strategy, with 47 studies included after title and abstract screening for full text review. 25 studies were included in the systematic review following full-text review. 22 studies were excluded in total because they either focused on pre-implementation processes (n = 12), evaluated the use of a singular tool (n = 4), evaluated perceptions of consumers (n = 4) or did not focus on a clinical setting (n = 2). Fig 1. Shows the PRISMA diagram for this process.
Discussion
In this systematic review, we comprehensively synthesized themes emerging from AI implementation frameworks, in healthcare, with a specific focus on the different phases of implementation. To help frame the AI implementation phases, we utilized the broadly recognizable PDSA approach. The present study found that current literature on AI implementation mainly focused on Plan and Study domains, whereas Do and Act domains were discussed less often, with a disparity in the representation of LMICs/LICs. Almost all framework authors originated from high-income countries (167 out of 172 authors, 97.1%), with the United States of America being the most represented (68 out of 172 authors, 39.5%).
Conclusion
The existing frameworks on AI implementation largely focus on the initial stage of implementation and are generated with little input from LICs/LMICs. Healthcare professionals repeatedly cite how challenging it is to implement AI in their clinical settings with little guidance on how to do so. For future adoption of AI in healthcare, it is necessary to develop a more comprehensive and inclusive framework through engaging collaborators across the globe from different socioeconomic backgrounds and conduct additional studies that evaluate these parameters.
Citation: Khan SD, Hoodbhoy Z, Raja MHR, Kim JY, Hogg HDJ, Manji AAA, et al. (2024) Frameworks for procurement, integration, monitoring, and evaluation of artificial intelligence tools in clinical settings: A systematic review. PLOS Digit Health 3(5): e0000514. https://doi.org/10.1371/journal.pdig.0000514
Editor: Zhao Ni, Yale University, UNITED STATES
Received: September 4, 2023; Accepted: April 18, 2024; Published: May 29, 2024
Copyright: © 2024 Khan et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.
Funding: This work was supported by the Patrick J. McGovern Foundation (Grant ID 383000239 to SDK, ZH, MHR, JYK, AAAM, FG, AH, AS, ST, NSK, MRP, SB, ZS, MPS). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: MPS is a co-inventor of intellectual property licensed by Duke University to Clinetic, Inc., KelaHealth, Inc, and Cohere-Med, Inc. MPS holds equity in Clinetic, Inc. MPS has received honorarium for a conference presentation from Roche. MPS is a board member of Machine Learning for Health Care, a non-profit that convenes an annual research conference. SB is a co-inventor of intellectual property licensed by Duke University to Clinetic, Inc. and Cohere-Med, Inc. SB holds equity in Clinetic, Inc.
