Artificial intelligent tools: evidence-mapping on the perceived positive effects on patient-care and confidentiality

Background Globally, healthcare systems have always contended with well-known and seemingly intractable challenges like safety, quality, efficient and effective clinical and administrative patient-care services. To firmly confront these and other healthcare challenges, the World Health Organisation proposed a full adoption of artificial intelligence (AI) applications into patient care to stimulate efficiency and guarantee quality in patient management. Purpose This review aimed to establish the extent and type of evidence of the positive effects of the use of AI tools in patient care. Thus, the review mapped evidence by using articles published between January 1, 2010, and October 31, 2023. Methods Consistent with the protocol by Tricco et al., a comprehensive literature search was executed from Nature, PubMed, Scopus, ScienceDirect, Dimensions, Web of Science, Ebsco Host, ProQuest, JStore, Semantic Scholar, Taylor & Francis, Emeralds, World Health Organisation, and Google Scholar. Upholding the inclusion and exclusion standards, 95 peer-reviewed articles were included in this review. Findings We report that the use of AI tools can significantly improve the accuracy of clinical diagnosis to guarantee better patient health outcomes. AI tools also have the ability to mitigate, if not eliminate, most of the factors that currently predict poor patient outcomes. Furthermore, AI tools are far more efficient in generating robust and accurate data in real time and can help ease and accelerate workflow at healthcare facilities. Conclusion If properly integrated into the healthcare system, AI will help reduce patients’ waiting time and accelerate the attainment of Sustainable Development Goals 3.4, 3.8, and 3.b. We propose that AI developers collaborate with public health practitioners and healthcare managers to develop AI applications that appreciate socio-cultural dimensions in patient care.


Introduction
The global healthcare system is challenged with the scarcity of critical healthcare professionals, changes in disease patterns, high cost of healthcare, adverse effects of climate change, pandemics, and access and equity issues, among others [1][2][3][4].Historically, the healthcare system has always contended with well-known and seemingly intractable challenges, including safety, treatment-diagnosis mismatch, misdiagnosis, under and over-prescription, inaccurate and incomplete patient records, inadequate resources and workforce to sustain the ever-stretched patient-care services [5,6].Given that the deadline set for the realisation of the Sustainable Development Goals (SDGs) is fast approaching, healthcare managers are adopting several strategies to fix the challenges sustainably [3,7].Considering that there is no quick fix to the numerous sets of healthcare challenges, the World Health Organisation (WHO) proposed, in addition to other interventions, a full integration of artificial intelligence (AL) tools in healthcare to stimulate efficiency and accelerate the realisation of the health-related SGDs [3,7,8].
AI tools are a set of technologies with computerised features that have the capacity to simulate intelligent human behaviours [5,8].These tools possess speed, huge data storage and processing capacity, are reliable, interoperable with other technological systems, and by far more accurate in their interpretations and patients' diagnoses [9,10].When effectively combined with human reasoning, AI tools have the ability to accurately establish patterns and complex correlations subtly in large and high-dimensional datasets that often escape the traditional techniques [8,11].
Though the full adoption of AI tools into healthcare is yet to be realised, there is evidence of the wide application of AI tools in patient care globally [10,12,13].So far, AI applications in patient care are getting more sophisticated, effective, and efficient in supporting clinical and administrative decisions [2,12].Regardless of the level of AI use in patient care across the globe, these intelligent machines appear to be super-supportive and could redefine the future of healthcare and change its face for the better [10,14,15].The utilisation of AI and other new technologies saw a surge globally during the Sars-Cov-2 pandemic [16][17][18].For instance, various AI and new technological platforms and devices were utilised to provide continuous and essential healthcare services to patients, including predicting mortality during the pandemic [16][17][18].Some of these mediums and devices include mobile-based self-care, video conferencing, virtual healthcare, tele-monitoring, tele-medicine, tele-consulting, tele-intensive care unit, e-consult, tele-radiology, virtual visits, and telesurgical services [16][17][18].At this point, the patient remains the ultimate "subject" in this whole discourse and is in whose best interest AI tools are deployed in the healthcare system [2,19,20].
While there is growing recognition of the utility of AI tools in patient care, their coverage in the developing world is rather on a small scale [21][22][23].So far, Asia, North America, and Europe appear to be the continents with the fastest AI coverage and with widespread application in patient-care services [13,22,24].Though these continents are far from realising full adoption of AI applications in all aspects of their healthcare systems, there are modest gains across these continents [25,26].For instance, funding for research projects in AI adoption in healthcare through the European Union Horizon 2020 scheme shot up between 2014 and 2020 [27].Moreover, the European Commission developed several ethicolegal instruments to regulate and guarantee the responsible design and use of AI systems in patient care and beyond [22,27].
In North America, for example, AI tools are currently being applied in the management of cancer, hypertension, cerebrovascular accidents and conditions, and in obstetrics as well as paediatric care services [6,9,27].The other continents, including Australia, South America, and Africa, have also recorded modest successes in the application of AI tools in their healthcare systems [7,13,22,28].Though records exist about AI applications in invasive and non-invasive procedures in these continents, especially in Africa, their use is more associated with smart devices aided by applications such as AiCure and a gamut of AI ChatBots [19,20,22].
Despite the growing application of AI in other fields worldwide, there seems to be inadequate empirical account, especially evidence synthesis, to clearly establish common themes and concepts across existing literature on the use of AI tools in patient care [13,15,20].Specifically, there is inadequate evidence mapping on how the use of AI tools in healthcare is positively impacting patient care globally [2].Meanwhile, these pieces of evidence are essential for developing policy and evidence-based integration of AI tools into healthcare for improved patient outcomes.Moreover, the role of AI in achieving SDG 3.4 (attainment of universal health coverage, including access to quality essential healthcare services, medicines and vaccines for all) by 2030 is not clear [3,7,8].Certainly, AI would be a critical resource in this global quest, and reviews collating such pieces of evidence are of urgent need.Therefore, this review aimed to fill this research gap by mapping the existing evidence of the positive effects of the use of AI tools in patient care.

Study design and search strategy
We examined, synchronised, and analysed peerreviewed articles using Tricco et al. [29] guidelines.This includes developing and examining the purpose of the study, crafting, reviewing, and examining the research questions, and then identifying and discussing the search terms.The other guidelines include identifying and exploring relevant databases, downloading articles, data mining, organising and synthesising results and carrying out consultation.Five questions defined the review, including, "what are the potential positive effects of AI applications on (1) healthcare data, (2) diagnostic decisions, (3) patient-care, (4) medical errors, and (5) medical emergencies?".
To inject rigour and comprehension into the search process, we first explored PubMed for Medical Subject Heading (MeSH) terms on the topic (see Table 1).The MeSH terms were validated by a research librarian with over 10 years of working experience in the university.Search for articles was executed at two levels based on the MeSH terms.First, the search terms "Confidentiality" or "Artificial Intelligence" yielded a total of 4,512 articles.Second, the search was based on 30 MeSH terms and also produced a total of 1,688 articles (see Fig. 1 and Table 1).The search covered studies conducted between January 1, 2010, and October 31, 2023.The current study was executed between January 1 and October 31, 2023.This review was conducted in consistent with the Preferred Reporting Items for Reviews and Meta-Analyses extension for Scoping Reviews-PRISMA-ScR [29,30].Through a detailed and an exhaustive data screening process, all duplicate articles were kept in a folder and later deleted.Many of these deleted articles were considered incoherent with the inclusion standards (as described below).The first level screening was executed by five authors (FSA, SM, RVK, LAA, and IST).However, where the suitability of an article was in contention, that article was referred to four other authors (EWA, CES, VKD, and NNB) for further assessment until consensus attained.To ensure comprehension and rigour in the search process, citation chaining was done on all full-text articles that met the inclusion standards to identify relevant additional articles for further assessment.

Data sources
We searched for peer-reviewed articles from the following databases/publishers/search engines: Nature, PubMed, Scopus, ScienceDirect, Dimensions, Web of Science, Ebsco Host, ProQuest, JStore, Semantic Scholar, Taylor & Francis, Emeralds, World Health Organisation, and Google Scholar (see Fig. 1 and Table 1).Through a comprehensive and independent assessment of various databases/publishers/search engines conducted by five authors (EWA, VKD, CES, SM, and NNB), the sources mentioned above were found to contain a very good number of relevant articles on the subject under review.

Language filter English
Time filter January 1, 2010 to October 31, 2023

Spatial filter Worldwide
MeSH terms used 1.Confidentiality -"Entry Terms" OR "Secrecy" OR "Privileged Communication" OR "Communication, Privileged" OR "Communications, Privileged" OR "Privileged Communications" OR "Confidential Information" OR "Information, Confidential" OR "Privacy of Patient Data" OR "Data Privacy, Patient" OR "Patient Data Privacy" OR "Privacy, Patient Data" 2. Artificial Intelligence -"Intelligence, Artificial" OR "Computational intelligence" OR "Intelligence, Computational" OR "Machine Intelligence" OR "Intelligence, Machine" OR "Computer Reasoning" OR "Reasoning, Computer" OR "AI (Artificial Intelligence)" OR "Computer Vision Systems" OR "Computer Vision System" OR "Systems, Computer Vision" OR "System, Computer Vision" OR "Vision System, Computer" OR "Vision Systems, Computer" OR "Knowledge Acquisition (Computer)" OR "Acquisition, Knowledge (Computer)" OR "Knowledge Representation (Computer)" OR "Knowledge Representations (Computer)" OR "Representation, Knowledge (Computer)"

Inclusion criteria
Articles must be AI and health-related, primary research and conducted in the English Language, applied either quantitative, qualitative, or mixed methods.In addition, the articles must provide details on perceived positive effects of AI use in healthcare, and conducted between January 1, 2010 and October 31, 2023.Again, articles must provide details on author(s), purpose, methods, country, and conclusion

Exclusion criteria
Articles on AI and health but did not touch on perceived positive effects of AI tools use in healthcare, reviewed articles on AI use in health, and articles on AI and health conducted in languages other than the English Language.Furthermore, abstracts, opinion pieces, short reports, incomplete articles, commentaries, grey literature, and media reports on AI and health were also ignored

Study selection
A random sample of 12 titles and abstracts screened independently by four authors (EWA, NNB, CES, and SM) was used to standardise the inclusion and exclusion criteria.Weekly virtual and in-person meetings were held to discuss and reconcile disagreements and clarify eligibility among the four authors (EWA, NNB, CES, and SM) and the fifth author (VKD).Before progressing to full-text screening, authors ensured that all differences concerning the selection of articles were resolved by five authors (EWA, NNB, CES, SM, and VKD).Three authors (EWA, NNB, and CES) independently screened the articles before data extraction commenced.

Data extraction and thematic analysis
All authors carried out data extraction independently.Four authors (CES, RVK, LAA, and IST) extracted data on "authors, purpose, methods, and country, " while five authors (EWA, VKD, FSA, SM, and NNB) extracted data on "perceived positive effects and conclusions" (see Table 2).
In consonance with Cypress [126] and Morse [127], thematic analysis was done by six authors (EWA, VKD, CES, SM, RVK, and NNB).Thus, data were coded until themes emerged directly from the data, in line with the stated research questions [127][128][129].Our analysis included reading over and over to familiarise ourselves with the data, identifying candidate codes, identifying and assessing emerging themes.Additionally, emerging themes were reviewed, clearly named and defined.However, where doubt occurred, we extensively discussed till consensus was established.Finally, a qualitative report was developed and extensively reviewed to guarantee internal and external homogeneity of the themes.

Quality rating
We conducted quality ratings on all candidate articles in line with the guidelines provided by Tricco et al. [29].That is, the shortlisted articles must have a research background, aim, context, clear method, sampling technique, data collection and analysis, reflectivity, value of research, and ethics.Therefore, all candidate articles were examined and scored according to the majority of the sections.Thus, articles that scored "A" had little or no limitations, "B" had some limitations, "C" had substantial limitations but carried some relevance, and "D" had substantial flaws that could undermine the validity of the study as a whole, so such articles were not used for this review [29].

Improvement in patient-care
Artificial intelligent tools are credited for their ability to mitigate, if not eliminate, most of the factors that currently predict poor patient outcomes (23; 24%) [32, 33, 36, 40, 45, 49, 54, 55, 60, 71, 82, 94-96, 100, 103, 108, 109, 113, 117, 121, 123, 125].These include errors in clinical diagnosis, long waiting times, poor staff attitudes, inaccurate and missing patient data, workload and staff burnout, and discrimination because a large number of patients need care [32,40,49,55,82,109,121].According to Fritsch et al. [55], Wang et al. [117], MacPherson et al. [82], and Ploug et al. [95], when AI tools that serve as machine clinicians are applied in combination with human clinicians, a rich and valuable context is provided to improve the quality of care provided to patients.Moreover, AI tools provide valuable opportunities for patients to receive needed care remotely [54,96,121,123].For instance, domestic caregivers could receive valuable guidance from machine clinician AI tools when confronted with difficult decisions regarding patients with chronic conditions such as cerebrovascular accident and related conditions, hypertension, and diabetes.This reduces the stressors associated with caring for patients with such chronic conditions and improves the needed quality of care [82,95,103], which may prevent or reduce medical emergencies.

Prompt detection of medical emergencies
Changes in the conditions of patients can sometimes be sudden and unpredictable, especially during emergency care [37,51,60,106,110].As reported in 16(17%) of the selected articles [31, 33, 35-37, 46, 51, 59, 60, 72, 90, 102, 106, 109, 110, 113], with the introduction of AI tools into healthcare, clinicians can now detect early and act swiftly in providing life-saving care to patients during medical emergencies.These are possible because AI tools have features that could trigger instantaneous alerts on imminent changes in patient conditions, such as seizures and strokes, and ensure timely medical interventions [31,36,51,59,107].The public [51,72,110,113] is hopeful that if well implemented, AI tools could become valuable in the management of medical emergencies.

Discussion
The utilisation of AI tools in governance, academia, manufacturing, security, entertainment, space and marine exploration, health, etcetera, is gaining popularity among researchers globally [15,24].There are several studies about the utility of AI tools in other fields [10,15,24], yet few studies exist on direct AI applications in patient care [13].Moreover, most of the studies that examined the application of AI tools in health were not conducted in the area of patient care [2,13].Affirming this, the current study found that out of 527 articles on AI use in health, only 95(31%) met the inclusion standards and formed part of this review.Besides, consistent with Khalid et al. [13] and Naik et al. [24], the current study reports that the articles reviewed were disproportionately concentrated in three continents (Asia, North America, and Europe), which account for over one-third of the total articles reviewed.We discuss our findings under the following themes: improvement in patient diagnosis, reduction in medical errors and facilitation of workflow, accurate and reliable data, improvement in patient care, and prompt detection of medical emergencies.

Improvement in patient diagnosis
Accurate and timely determination of a patient's diagnosis defines the patient-clinician relationship, which becomes a key prerequisite for administering treatment for improved patient outcomes [13,130].Center of Intellectual Property and Technology Law [12] explained that it is both a legal and moral obligation for a clinician to exercise due diligence in diagnosing patients' conditions and disclosing the same to the patient.The suggestion is that the design of AI tools provides a superior advantage of delivering accurate diagnoses in real time [14,19].Consistent with this, our review found that these machine clinician AI tools could significantly improve the accuracy of clinical diagnosis.Moreover, integrating AI tools into the care process will provide a robust and trustworthy context for clinicians to shape and improve their own diagnostic decisions.For instance, Al-Zaiti et al. [35] reported that an AI model outperformed both commercial interpretation software and experienced clinician interpretation in diagnosing underlying Acute Myocardial Ischemia in patients with chest pain.Again, Amarbayasgalan et al. [37] reported that the Reconstructive Error (RE) based Deep Neural Networks (DNNs) AI model outperformed other models in diagnosing coronary heart disease risk.Also, Ayatollahi et al. [38] found that Artificial Intelligence Neural Network (ANN) and Support Vector Machine (SVM) algorithms had higher sensitivity in diagnosing Cardiovascular Disease.
The caution, however, is that the current review disaffirms previous studies [10,11] that reported the possibility of AI tools committing serious errors in their diagnostic decisions based on factors like inaccurate and biased data and inadequate machine training and learning.This notwithstanding, the findings of the current review based on a large body of previous studies [12-14, 19, 130] suggest that AI tools in patient care significantly address shortcomings in the traditional diagnostic regime.Thus, if well implemented, AI tools in patient care could stimulate and accelerate the realisation of SDG 3.4, which calls for a decrease by one-third the avoidable deaths from non-communicable diseases by the year 2030 [3,7].

Reduction in medical errors and workflow
There is a correlation between increased workload, stress, and burnout of clinicians and the occurrence of medical errors in patient care [9][10][11].Some of these medical errors result in serious negative health outcomes (including death) of patients.Our review found that AI tools can help minimise the incidence of medical errors associated with hospital-based stressors.In one example, Catho et al. [45] found that Computerised Decision Support Systems (CDSSs) model could actually reduce medical errors and support the clinical decisions of physicians.Another finding [51] suggests that the Coronary Heart Disease (CHD) prediction AI model could reduce medical errors and shape the clinical decisions of physicians.Thus, IA tools can enhance the trust that patients have for healthcare professionals.Additionally, the reviewed studies [5,11,15,131] reported that AI tools will serve the best interest of patients.This may mean that these machine clinician AI tools seem to provide a great way forward in healthcare presently.

Accurate and reliable data
The role of robust, accurate, and reliable data in the decision-making process regarding patient care cannot be overemphasised [13,15].Several studies [11,14,35,80] recognised the super-abilities of AI tools in procuring, organising, and preserving large volumes of datasets for use in both clinical and administrative decisions concerning patient care.For example, Huang et al. [67] reported that the Major Adverse Cardiac Events (MACE) prediction AI model can leverage a large data-mining-based approach in predicting acute coronary syndrome (ACS) in patients.Again, Kayvanpour et al. [73] reported that the Silico Neural Network AI model can utilise multimodal data points in accurately predicting Acute Coronary Syndromes in patients.However, the findings from the current study oppose some earlier ones [10,12], suggesting that AI tools could seriously compromise data privacy.In addition, AI tools are vulnerable to attack by computer hackers who could misuse patient records [9,19].Although these concerns are legitimate, some previous studies [2,15] suggest that AI tools have inherent security mechanisms against data leaks and theft.Moreover, in the long term, the odds are high that AI tools will guarantee more reliable, accurate, and timely data in patient care.Thus, with proper design and implementation, including an effective ethicolegal framework, AI tools used in patient care would help in achieving SGD 3.b, which calls for research and development of vaccines and medicines for communicable and non-communicable diseases by 2030 [3,7].

Improvement in patient-care
The traditional patient-care regime is largely characterised by delays in receiving care, discrimination, poor care-provider attitude, staff fatigue and stress, inadequate staff, misdiagnosis, treatment-diagnosis mismatches, under-and over-prescription, missing patient records, etcetera [8,15].These factors contribute to the everincreasing incidence of mortality and complications in patient conditions recorded in most healthcare facilities worldwide [14].However, our review found that AI tools have the capacity to significantly reduce, if not eliminate, most of the factors that currently undermine patent care.For example, Pattarabanjird et al. [93] reported a general improvement in the health of patients with severe Coronary Artery Disease when the Single Nucleotide Polymorphism (SNP) AI model was applied.Also, Uzir et al. [110] reported that AI-enabled smartwatch applications are promoting health democracy and personal healthcare.Meanwhile, Walter et al. [115] found that the Automatic pain recognition (APR) AI system had a high sensitivity to pain in patients, which helps in pain management.
AI tools have more effective and efficient data storage and protection ability, data interpretation ability, accurate diagnostic ability, and exceptionally expeditious and reliable service to patients [8,11,14].For instance, previous studies [7,13], as corroborated by the findings from our current review, indicated that AI tools have the ability to provide needed care remotely to patients who may not necessarily be present physically at the hospital.This will help reduce large patient numbers at the hospital and improve the overall turnaround time for care.However, our review contradicts other previous studies [20,132], which raised concerns over AI tools providing care in a discriminatory manner.Thus, such findings question the utility of AI tools in mental health services and are sceptical about their ability to provide non-pharmacological care.Regardless, a large body of previous studies [7,[13][14][15] suggests that AI tools could significantly improve the total quality of care to patients.Ultimately, AI tools would contribute significantly to the realisation of the universal health coverage provided in SDG 3.8.

Prompt detection of medical emergencies
Critically ill patients are constantly under close monitoring for even the slightest change in condition [15,130].So far, AI tools have proven very helpful in saving the lives of many patients [11,131].For instance, with the introduction of AI tools into healthcare, clinicians can detect and act swiftly to provide life-saving care to patients during medical emergencies [5,15].One evidence is that the Automated Delirium-Risk Assessment System (Auto-DelRAS) has a high level of validity in predicting delirium risk in patients in the intensive care unit [85].Also, Hu et al. [65] reported that the Rough Set Theory (RST) and Dempster-Shafer Theory (DST) have higher sensitivity in predicting the incidence of Major Adverse Cardiac events during medical emergencies.Further evidence [76] suggests that deep neural network AI application has a better sensitivity in predicting mortality among patients with Spontaneous Coronary Artery Dissection (SCAD).Thus, given its several algorithms, these intelligent applications could detect imminent changes in patient conditions and trigger instantaneous alerts for quick interventions [9].Moreover, we reported that there are AI applications that could significantly mitigate errors associated with AI tools [9,14,130].So far, there seems to be no better or more competent alternative to AI tools in the management of medical emergencies.

Strengths and limitations
The review attempts to explore evidence of the perceived positive effects of AI applications in patient care from a global perspective.To ensure reproducibility, reliability, and trustworthiness of our findings, there was strict adherence to the following: first, all authors independently searched for and screened articles using the MeSH terms.Moreover, guided by the inclusion and exclusion guidelines and a checklist, all selected full-text articles were subjected to a quality rating.Additionally, to establish validity and replicability, all authors participated in a thorough data extraction process and review.
This review also has some limitations.First, relying on only peer-reviewed articles and selecting articles written in the English language limit the literature sample used because we might have excluded other relevant articles written in other languages.Moreover, we recognise that the study may have carried weaknesses and biases contained in the reviewed articles.Therefore, the generalisability of our findings may be limited.Finally, it is important to acknowledge that regardless of the positive outcomes reported for AI tools by several randomised trials, the generalisability of the utility of these intelligent tools is yet to be established, given that the overall utility of AI tools is dependent on the quality of data and the training provided to the tools.

Recommendations for policy and research directions
We propose that governments leverage AI applications to aid and accelerate the realisation of the health-related SDGs in their jurisdictions.Specifically, governments in developing countries can provide financial support for the effective and efficient adoption of AI tools in healthcare.Secondly, there would be the need to introduce academic courses on the application of AI tools in healthcare for all categories of health professionals, especially in developing countries.Thirdly, governments in developing countries need to sponsor Biomedical Engineers to be trained in the use of AI tools in healthcare.Such training should, among other specialties, cover cyber Intelligence.In addition, public health experts and healthcare managers need to collaborate with AI developers to develop applications that can efficiently provide both pharmacological and non-pharmacological care to patients.Furthermore, we propose that AI developers need to collaborate with healthcare managers to develop AI applications that are super sensitive to socio-cultural dimensions in patient care.Lastly, we encourage WHO and other agencies to provide sponsorship for research into AI applications, patient care, and SDGs.Future research may explore how AI applications are promoting public health interventions, such as the fight against pandemics and epidemics.Also, assessing the potential negative effects of the use of AI tools in patient care is warranted.
While making these recommendations, we envisage some potential barriers.These may include non-commitment from political decision-makers to successfully implement projects incorporating AI tools into healthcare, especially in developing nations.First, budgetary constraints and increasing competing demands on governments in most developing countries may limit the implementation of these recommendations.Second, unstable power supply, cultural sensitivity and potential hesitancy to new technology, conspiracy, and prevalence of cyber Intelligence globally could undermine the smooth implementation of AI tools used in patient care in less-resourced countries.

Conclusion
AI applications are steadily and rapidly shaping the relationship between clinicians and patient care globally.This development has attracted some criticism, including potential breaches of privacy, data fraud, bias and discrimination, and decline in humanity during patient care.However, AI applications are demonstrating enhanced capacity that can change the course of our collective future for the better.Thus, AI tools have the ability to improve the accuracy of clinical diagnosis significantly and, guarantee better health outcomes for patients and mitigate, if not eliminate, most of the factors that currently predict poor patient outcomes.Furthermore, AI tools are far more efficient in generating robust and accurate data in real time and could help ease and accelerate workflow.Additionally, AI devices and applications are contributing largely to the management of medical emergencies.
Furthermore, if properly integrated into the healthcare systems and used in patient care, AI tools could accelerate the realisation of SDGs 3.4, 3.8, and 3.b.So far, there seems to be no going back on this journey of AI use, including their use as machine clinicians in patient care.Thus, the focus should be on ensuring that these tools are responsibly applied to provide the needed results-improvement in health outcomes.This study is a significant addition to existing evidence on the use of AI tools in healthcare.

Fig. 2 Fig. 3
Fig. 2 Yearly distribution of included articles in percentages