Artificial Intelligence in health care (focus on hospitals)

Duration: Mid- April to Mid- November 2024
Language: English (with German summary)
Publication: HTA Project Report No. 164: https://eprints.aihta.at/1546/

Background:
Artificial intelligence (AI) is a field of computer science that aims to imitate human cognitive abilities. AI is becoming increasingly important and is expected to have a significant impact on various social fields, such as education, research, and medicine [1]. Within AI, there are specific sub-areas such as machine learning and deep learning. Machine learning refers to algorithms and techniques that allow computers to learn from datasets and make predictions or decisions without explicit programming. Deep learning is a subset of machine learning that is based on functions of the human brain and is used to process more complex data [1-3].

AI is already being utilised in various areas within the healthcare sector [4]. Due to its capability for pattern recognition, AI is increasingly employed in radiology, particularly for the analysis of imaging procedures to diagnose diseases such as cancer, eye, and lung diseases [1]. AI has the potential to evaluate heartbeat patterns in cardiology and to predict diabetes risks in endocrinology. In addition, AI systems are expected to develop the ability to recognise speech patterns indicative of mental illnesses [5]. In Austria, 43 AI applications have been identified by the Austrian National Public Health Institute (Gesundheit Österreich GmbH - GÖG), either deployed as pilot projects or already in regular operation in hospitals [6]. These applications cover various areas, including screening and prognosis, diagnostics, big data analysis, as well as supporting therapy planning and administration.

Although AI offers promising opportunities in healthcare, there are also some risks and challenges associated with it [3, 7]. One of the biggest concerns is the security and privacy of sensitive health data that is processed by these systems. Another challenge is the lack of transparency and explainability of AI algorithms, which can make it difficult to understand how decisions are being made. Due to these challenges, AI systems used in healthcare are subject to market approval regulations for high-risk medical devices. Providers of these systems must implement a risk management and quality management system, along with a data governance protocol, and provide technical documentation and user instructions [8]. However, there is currently a lack of methodological guidance for assessing the benefits of AI systems in different fields of application in hospitals.

Project goals:
The project aims to provide an overview …

1.)    of AI systems in hospitals, categorised by their fields of application,

2.)    of HTA methods for evaluating the additional benefits of AI applications, and

3.)    of HTA reports in different application areas and extract the methods used

Non-objectives:
The project does not aim to…

• systematically evaluate the effectiveness of AI systems in various areas of application
• assess the benefit of individual AI-applications
• conduct a systematic search of AI systems (products)
• provide an overview of AI systems in other sectors, such as those outside the healthcare sector and outpatient settings
• analyse ethical and legal aspects

Research questions:
The following research questions (RQ) will be addressed:

RQ1: What are potential applications of AI systems in Austrian hospitals for medical purposes and what are the expectations for their additional benefit?

RQ2: How can hospitals assess the potential clinical added benefit of AI applications? What HTA methods and frameworks can be used for AI procurement and implementation decisions?

RQ3: What evaluation methods were used in previous HTA reports to assess the additional (clinical) benefits of AI applications in specific areas?

RQ4: What specific recommendations (e.g. requirements of evidence, quality assurance, and clinical applications) can be drawn from the collected information for the successful implementation of AI systems in Austrian hospitals?

Methods: Scoping overview of potential application areas of AI systems in hospitals, international HTA methods focusing on AI, previous HTA reports and exemplary AI use in Austrian hospitals, including expectations, risks, and potential methods for benefit assessments.

RQ1: mapping of application areas

• conducting a structured manual search, reaching out to relevant experts, documenting the findings in tabular form (incl. expectations and risks)
• based on the GOEG-report: categorising 43 AI systems implemented in Austria and exemplary description

RQ2-3: identification and overview of HTA methods and assessments of AI system in application areas, overview using vignettes

• conducting a systematic search to identify available HTA methods and assessments, such as INAHTA and institutional HTA websites, to find available methods and assessments related to AI
• documenting identified assessments in a table
• extracting the predefined data from HTA reports and assessments
• synthesising the findings
• creating vignettes to provide an overview

RQ4: elaborating conclusions and deriving recommendations of action

• narratively synthesising the results
• deriving recommendations for actions (e.g. requirements of evidence, quality assurance)

Inclusion criteria:

All steps including study selection, data extraction are performed by two researchers. The results will be reviewed by an AIHTA reviewer (internal review) and at least one peer reviewer (AI-expert).

Timetable and milestones:

References:

[1]       Al Kuwaiti A., Nazer K., Al-Reedy A., Al-Shehri S., Al-Muhanna A., Subbarayalu A. V., et al. A Review of the Role of Artificial Intelligence in Healthcare. J Pers Med. 2023;13(6). Epub 20230605. DOI: 10.3390/jpm13060951.

[2]       Martinez-Millana A., Saez-Saez A., Tornero-Costa R., Azzopardi-Muscat N., Traver V. and Novillo-Ortiz D. Artificial intelligence and its impact on the domains of universal health coverage, health emergencies and health promotion: An overview of systematic reviews. Int J Med Inform. 2022;166:104855. Epub 20220817. DOI: 10.1016/j.ijmedinf.2022.104855.

[3]       Fraser A. G., Biasin E., Bijnens B., Bruining N., Caiani E. G., Cobbaert K., et al. Artificial intelligence in medical device software and high-risk medical devices - a review of definitions, expert recommendations and regulatory initiatives. Expert Rev Med Devices. 2023;20(6):467-491. Epub 20230508. DOI: 10.1080/17434440.2023.2184685.

[4]       Bures D., Hosters B., Reibel T., Jovy-Klein F., Schramm J., Brendt-Müller J., et al. Die transformative Wirkung von künstlicher Intelligenz im Krankenhaus. Die Innere Medizin. 2023(64):1025-1032. DOI: https://doi.org/10.1007/s00108-023-01597-9.

[5]       Bedi G., Carrillo F., Cecchi G. A., Slezak D. F., Sigman M., Mota N. B., et al. Automated analysis of free speech predicts psychosis onset in high-risk youths. NPJ Schizophr. 2015;1:15030. Epub 20150826. DOI: 10.1038/npjschz.2015.30.

[6]       Degelsegger-Márquez A., Dick D. and Trunner K. Telemedizin und Künstliche Intelligenz im intramuralen Bereich Österreichs. Ereignisbericht.: 2022 [cited 3.4.2024]. Available from: https://jasmin.goeg.at/id/eprint/2443/1/Telemedizin_und_KI_in_Krankenanstalten_bf.pdf.

[7]       Prasser F., Riedel N., Wolter S., Corr D. and Ludwig M. Künstliche Intelligenz und sichere Gesundheitsdatennutzung im Projekt KI-FDZ: Anonymisierung, Synthetisierung und sichere Verarbeitung von Real-World-Daten. Bundesgesundheitsbl. 2024(67):171-179.

[8]       Future of Life Institute. EU Aritifical Intelligence Act. 2024 [cited 10.4.2024]. Available from: https://artificialintelligenceact.eu/de/high-level-summary/.