CoChat
Sign Up
Cited by 0Artificial Intelligence in Healthcare and EducationEthics and Social Impacts of AIHealthcare Regulation

The AI Hospital Formulary: A Practical Governance Framework for Prescribing, Monitoring, and Deprescribing Artificial Intelligence in Hospitals

Open access
Epelde, Francisco·2026·Hospitals
DOI 10.3390/hospitals3030015
Open this paper in CoChat — free accountOpen PDF

Abstract

Background: Artificial intelligence (AI) is increasingly entering hospital practice through diagnostic, predictive, workflow, operational, and generative applications. Hospitals often govern these systems as procurement or information-technology projects rather than as clinical–organizational interventions requiring indication, evaluation, monitoring, accountability, and withdrawal. Objective: To refine the concept of an “AI Hospital Formulary” as an operational, proportional, and accountable framework for the safe, equitable, and sustainable adoption of hospital AI. Design and Methods: This is a perspective article using a structured, non-systematic narrative synthesis and conceptual framework development. Targeted literature and policy sources were identified through purposive searches and citation chaining through 20 July 2026. The synthesis compares the formulary with existing oversight approaches, maps its lifecycle gates to regulatory and risk management duties, and applies the framework to a worked example based on published evaluations of the Epic Sepsis Model. The EQUATOR reporting-guideline selection tool was consulted, and SANRA was used to strengthen the narrative synthesis component. Framework: The revised framework combines a hospital-wide AI register, a standardized formulary monograph, six lifecycle gates, proportional review pathways, governance-of-governance safeguards, cloud and data-sovereignty controls, continuous monitoring of technical and behavioral feedback loops, and explicit renewal or deprescribing criteria. The worked example shows how version-specific evidence can lead to local validation, controlled implementation, restriction, suspension, or renewal rather than automatic adoption. Conclusions: Hospitals should not merely purchase, install, and update AI systems. They should prescribe, monitor, audit, renew, restrict, and, when necessary, deprescribe them. The AI Hospital Formulary is proposed as a complementary institutional layer that converts external standards and existing governance approaches into documented portfolio decisions at the hospital level.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper proposes an “AI Hospital Formulary” as a hospital-level governance framework for prescribing, monitoring, renewing, restricting, and deprescribing AI systems. It combines a hospital-wide AI register, standardized monograph, six lifecycle gates, proportional review, governance safeguards, cloud/data-sovereignty controls, algorithmovigilance, and explicit deprescribing criteria, illustrated with published evaluations of the Epic Sepsis Model.

Key findings

  • The revised framework includes a hospital-wide AI register, a standardized formulary monograph, six lifecycle gates, proportional review pathways, governance-of-governance safeguards, cloud and data-sovereignty controls, continuous monitoring of technical and behavioral feedback loops, and explicit renewal or deprescribing criteria.
  • In the worked example, a published external evaluation of the original Epic Sepsis Model reported an AUC of 0.63, sensitivity of 33%, specificity of 83%, and positive predictive value of 12% at a commonly used threshold, with substantial alert burden.
  • A multicenter prospective validation of version 2 in 227,091 encounters across four health systems reported encounter-level discrimination of 0.82 to 0.92, but positive predictive value remained 0.13 to 0.26, alert burden was high, and thresholds and performance varied across institutions.
  • The paper argues that authorization must be version-specific, indication-specific, time-limited, locally validated, and paired with an exit plan rather than automatic adoption.

Methods

  • Perspective article using a structured, non-systematic narrative synthesis and conceptual framework development; the paper states it is not a systematic review, scoping review, consensus study, empirical validation study, or legal analysis.
  • Targeted literature and policy sources were identified through purposive searches and citation chaining through 20 July 2026; the EQUATOR reporting-guideline selection tool was consulted, and SANRA was used to strengthen the narrative synthesis component.
  • The framework is applied to a worked example based on published evaluations of the Epic Sepsis Model; no new data were created or analyzed.

Limitations

  • Stated by the paper: it is a conceptual Perspective based on a purposive, non-systematic narrative synthesis and does not provide exhaustive evidence identification, formal risk-of-bias assessment, consensus validation, or prospective testing of the complete framework.
  • Stated by the paper: the sepsis example uses published data and is not a new empirical evaluation or local implementation study; the regulatory mapping is illustrative and cannot replace jurisdiction-specific legal analysis; the framework may require adaptation for national health systems, hospital networks, and smaller institutions.
  • Evident from the text: the framework has not been piloted or prospectively tested; the paper states future research should pilot the formulary, measure feasibility and decision time, assess effects on procurement quality and incident detection, and compare implementation and deprescribing outcomes with existing fragmented governance.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/hospitals3030015

Cites

41 of 45 references verified0 citation errors · 4 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

12 of 21 cited sources checked · 8 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Translation of AI from technical performance to measurable clinical and organizational benefit remains difficult.

    “Yet the translation of AI from technical performance to measurable clinical and organizational benefit remains difficult [4].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [4] Kelly, C.J., Karthikesalingam, A., Suleyman, M., Corrado, G., and King, D. (2019). Key challenges for delivering clinical impact with artificial intelligence. BMC Med., 17. source
  • Interpretationcitation

    Evidence for Epic Sepsis Model version 1 must not be assumed to apply to version 2 or later versions.

    “Evidence for version 1 must not be assumed to apply to version 2 or later versions [52,53].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [52] Wong, 2021, External validation of a widely implemented proprietary sepsis prediction model in hospitalized patients, JAMA Intern. Med. source

      The Introduction warns that proprietary models are difficult to assess and may decline in performance over time, and the Discussion’s Limitations confines this validation to one medical center. That supports caution about transferring the study’s findings, but our copy does not identify the model as version 1 or discuss version 2 or later versions. (LLM, conf=0.75)

      Experimental — accuracy not yet measured
    • cites [53] Wong, 2026, Multicenter prospective validation of an updated proprietary sepsis prediction model, JAMA Netw. Open source
  • Premisecitation

    A multicenter prospective validation published in 2026 evaluated version 2 of the Epic Sepsis Model in 227,091 encounters across four health systems.

    “A multicenter prospective validation published in 2026 evaluated version 2 in 227,091 encounters across four health systems.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [53] Wong, 2026, Multicenter prospective validation of an updated proprietary sepsis prediction model, JAMA Netw. Open source
  • Premisecitation

    Existing hospital governance models, model-documentation tools, patient-safety approaches, and reporting standards provide important components for institutional oversight.

    “Existing hospital governance models, model-documentation tools, patient-safety approaches, and reporting standards also provide important components for institutional oversight [15][16][17][18][19][20][21][22][23][24].”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites [15] Bedoya, 2022, A framework for the oversight and local deployment of safe and high-quality prediction models, J. Am. Med. Inform. Assoc. source
    • cites [16] Reddy, 2020, A governance model for the application of AI in health care, J. Am. Med. Inform. Assoc. source
    • cites [17] Collins, 2024, TRIPOD+AI statement: Updated guidance for reporting clinical prediction models that use regression or machine learning methods, BMJ source
  • Premisecitation

    A published external evaluation of the original Epic Sepsis Model reported limited performance in one health system: an area under the receiver operating characteristic curve of 0.63, sensitivity of 33%, specificity of 83%, and a positive predictive value of 12% at a commonly used threshold, with a substantial alert burden.

    “A published external evaluation of the original Epic Sepsis Model reported limited performance in one health system: an area under the receiver operating characteristic curve of 0.63, sensitivity of 33%, specificity of 83%, and a positive predictive value of 12% at a commonly used threshold, with a substantial alert burden [52].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [52] Wong, 2021, External validation of a widely implemented proprietary sepsis prediction model in hospitalized patients, JAMA Intern. Med. source
  • May combine two claims. 12
  • Classified as probably not a substantial claim. 23
  • The quote only partly states this claim. 1

  • May combine two claims.

    Premisecitation

    Clinicians may ignore alerts, over-trust outputs, misunderstand scores, or use the tool outside its intended scope.

    “Clinicians may ignore alerts, over-trust outputs, misunderstand scores, or use the tool outside its intended scope.”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites [24] Ratwani, 2024, Patient safety and artificial intelligence in clinical care, JAMA Health Forum source

      The safe-implementation discussion (page not available in our copy) identifies a tendency to overly trust AI and health care professional complacency. It does not specifically describe clinicians ignoring alerts, misunderstanding scores, or using tools outside their intended scope. (LLM, conf=0.87)

      Experimental — accuracy not yet measured
    • cites [31] Embi, 2021, Algorithmovigilance—Advancing methods to analyze and monitor artificial intelligence-driven health care for effectiveness and equity, JAMA Netw. Open source

      The JAMA Network section (page not specified) discusses users interpreting algorithmic outputs differently, lacking trust, and becoming overly reliant on outputs. It also describes variation in deployment and application. It does not specifically address clinicians ignoring alerts, misunderstanding scores, or using a tool outside its intended scope. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [32] Balendran, 2024, Algorithmovigilance, lessons from pharmacovigilance, npj Digit. Med. source

      The Post-approval evaluation section (page unavailable) says human factors, including how clinicians make decisions based on model predictions, matter to AI safety. Causality assessment also identifies operators’ skills and experience as possible contributors to errors. The paper does not specifically describe clinicians ignoring alerts, over-trusting outputs, misunderstanding scores, or using a tool outside its intended scope. (LLM, conf=0.78)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    In the multicenter prospective validation of Epic Sepsis Model version 2, encounter-level discrimination ranged from 0.82 to 0.92, but the positive predictive value remained 0.13 to 0.26, the alert burden was high, and thresholds and performance varied across institutions.

    “Encounter-level discrimination ranged from 0.82 to 0.92, but the positive predictive value remained 0.13 to 0.26, the alert burden was high, and thresholds and performance varied across institutions [53].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [53] Wong, 2026, Multicenter prospective validation of an updated proprietary sepsis prediction model, JAMA Netw. Open source

  • May combine two claims.

    Premisecitation

    The diagnostic AI literature has documented limitations in study design, reporting standards, risk of bias, and claims about performance compared with clinicians.

    “The diagnostic AI literature illustrates this gap: systematic reviews have documented limitations in study design, reporting standards, risk of bias, and claims about performance compared with clinicians [5,6].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [5] Nagendran, 2020, Artificial intelligence versus clinicians: Systematic review of design, reporting standards, and claims of deep learning studies, BMJ source
    • cites [6] Liu, 2019, A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: A systematic review and meta-analysis, Lancet Digit. Health source

  • Classified as probably not a substantial claim.

    Findingnone given

    The worked example shows how version-specific evidence can lead to local validation, controlled implementation, restriction, suspension, or renewal rather than automatic adoption.

    “The worked example shows how version-specific evidence can lead to local validation, controlled implementation, restriction, suspension, or renewal rather than automatic adoption.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    The AI Hospital Formulary framework may require adaptation for national health systems, hospital networks, and smaller institutions.

    “The framework may also require adaptation for national health systems, hospital networks, and smaller institutions.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Smaller hospitals may lack the expertise to operate a full committee.

  • Classified as probably not a substantial claim.

    Interpretationnone given

    The AI Hospital Formulary translates governance principles into a hospital authorization and portfolio management process.

    “The AI Hospital Formulary translates governance principles into a hospital authorization and portfolio management process.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductnone given

    This article is a conceptual Perspective based on a purposive, non-systematic narrative synthesis; it does not provide exhaustive evidence identification, formal risk-of-bias assessment, consensus validation, or prospective testing of the complete framework.

    “It is a conceptual Perspective based on a purposive, non-systematic narrative synthesis; it does not provide exhaustive evidence identification, formal risk-of-bias assessment, consensus validation, or prospective testing of the complete framework.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    The AI Hospital Formulary has economic value.

    “The AI Hospital Formulary also has economic value.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Purchase price is only one component of the total cost of an AI system.
    • Rests on an assumption: A nominally inexpensive AI system may be costly if it generates unnecessary work or creates vendor lock-in.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    SANRA was used to strengthen the narrative synthesis component.

    “SANRA was used to strengthen the narrative synthesis component.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    Governance-of-governance safeguards reduce the risk that the committee itself becomes opaque, commercially captured, or inaccessible to patients and professionals.

    “Governance-of-governance safe-guards reduce the risk that the committee itself becomes opaque, commercially captured, or inaccessible to patients and professionals.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The AI Hospital Formulary provides a familiar institutional container in which local predictive-model governance, model documentation, patient-safety infrastructure, cybersecurity, procurement, regulatory duties, ethics, and implementation science can produce one traceable decision about whether a system may be used, under which restrictions, and for how long.

    “Rather, it provides a familiar institutional container in which local predictive-model governance, model documentation, patient-safety infrastructure, cybersecurity, procurement, regulatory duties, ethics, and implementation science can produce one traceable decision about whether a system may be used, under which restrictions, and for how long.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    The AI Hospital Formulary creates institutional memory of what is in use, why it was approved, how it is performing, and when it should be reconsidered.

    “The formulary also creates institutional memory of what is in use, why it was approved, how it is performing, and when it should be reconsidered.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    The AI Hospital Formulary strengthens accountability by assigning named ownership, creating a decision record, and linking monitoring to the authority for restriction or suspension.

    “A formulary assigns named ownership, creates a decision record, and links monitoring to the authority for restriction or suspension.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Without a formulary, responsibility can be diffused among vendors, clinicians, information technology teams, innovation units, and executives.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Premisenone given

    Pre-implementation studies cannot capture all future contexts, workflow adaptations, or population shifts.

    “Pre-implementation studies cannot capture all future contexts, workflow adaptations, or population shifts.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The AI Hospital Formulary makes deprescribing planned, evidence-based, and accountable by defining triggers, authority, evidence, communication, and transition steps before deployment.

    “The formulary makes deprescribing planned, evidence-based, and accountable by defining triggers, authority, evidence, communication, and transition steps before deployment.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Hospitals often accumulate digital tools without systematic retirement.
    • Rests on an assumption: Accumulation of digital tools without systematic retirement leads to alert fatigue, redundant systems, hidden costs, outdated models, and unclear accountability.
    • Rests on an assumption: Withdrawal of an AI system may be technically, commercially, clinically, and politically difficult, particularly after users and workflows become dependent on a tool.
    • Rests on an assumption: Sudden removal of an AI system may itself create risk.

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    The AI Hospital Formulary becomes both a safety instrument and a strategic portfolio management tool.

    “The formulary thus becomes both a safety instrument and a strategic portfolio management tool.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Premisenone given

    Many AI systems are opaque or dependent on vendor-controlled updates, making subtle failure difficult to detect.

    “Fifth, many AI systems are opaque or dependent on vendor-controlled updates, making subtle failure difficult to detect.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The AI Hospital Formulary offers a practical institutional layer for safe, equitable, secure, and sustainable adoption of AI in hospitals.

    “The AI Hospital Formulary offers a practical institutional layer for safe, equitable, secure, and sustainable adoption.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    The sepsis example uses published data and does not constitute a new empirical evaluation or local implementation study.

    “The sepsis example uses published data and does not constitute a new empirical evaluation or local implementation study.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Hospitals need an operational mechanism that brings existing contributions together in an enforceable institutional authorization and renewal process.

    “Hospitals nevertheless need an operational mechanism that brings these contributions together in an enforceable institutional authorization and renewal process.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Hospitals often govern AI systems as procurement or information-technology projects rather than as clinical-organizational interventions requiring indication, evaluation, monitoring, accountability, and withdrawal.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The regulatory mapping in this article is illustrative and cannot replace jurisdiction-specific legal analysis.

    “The regulatory mapping is illustrative and cannot replace jurisdiction-specific legal analysis.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The AI Hospital Formulary approach can avoid two extremes: uncritical enthusiasm and regulatory paralysis.

    “This approach can avoid two extremes: uncritical enthusiasm, in which tools are adopted because they are innovative or commercially attractive, and regulatory paralysis, in which uncertainty prevents potentially useful innovation.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Premisereasoning

    Statistical subgroup evaluation is necessary but insufficient: similar performance can still lead to unfair consequences if outputs alter access to beds, specialist review, diagnostic testing, discharge planning, or follow-up.

    “Statistical subgroup evaluation is necessary but insufficient: similar performance can still lead to unfair consequences if outputs alter access to beds, specialist review, diagnostic testing, discharge planning, or follow-up.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The EQUATOR reporting-guideline selection tool was consulted.

    “The EQUATOR reporting-guideline selection tool was consulted”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Interpretationreasoning

    The formulary can create the local evidence trail and assignments needed to operationalize many duties under the EU AI Act and other regulatory and risk management frameworks.

    “Nevertheless, the formulary can create the local evidence trail and assignments needed to operationalize many duties.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The distinctive contribution of the AI Hospital Formulary is the combination of existing governance elements in a hospital-wide authorization and portfolio management process modeled on a familiar clinical institution.

    “The distinctive contribution of the formulary is the combination of these elements in a hospital-wide authorization and portfolio management process modeled on a familiar clinical institution.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Existing regulations and guidance emphasize risk management, transparency, post-market monitoring, human oversight, lifecycle evaluation, and trustworthy AI governance.

    “Existing regulations and guidance increasingly emphasize risk management, transparency, post-market monitoring, human oversight, lifecycle evaluation, and trustworthy AI governance [7][8][9][10][11][12][13][14].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [7] World Health Organization (2021). Ethics and Governance of Artificial Intelligence for Health: WHO Guidance, World Health Organization. Available online: https://www.who.int/publications/i/item/9789240029200.

      Not checked in the free run.

    • cites [8] World Health Organization (2023). Regulatory Considerations on Artificial Intelligence for Health, World Health Organization. Available online: https://www.who.int/publications/i/item/9789240078871.

      Not checked in the free run.

    • cites [9] World Health Organization (2025). Ethics and Governance of Artificial Intelligence for Health: Guidance on Large Multi-Modal Models, World Health Organization. Available online: https://www.who.int/publications/i/item/9789240084759.

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    The AI Hospital Formulary can strengthen procurement.

    “The formulary can strengthen procurement.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The evidence supports version-specific local validation for the Epic Sepsis Model; mixed evidence and low positive predictive value justify caution rather than automatic rejection or adoption.

    “The evidence supports version-specific local validation; mixed evidence and low PPV justify caution rather than automatic rejection or adoption.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The Epic Sepsis Model findings demonstrate why an authorization must identify the version and why improved external evidence still does not remove the need for local validation and workflow evaluation.

    “These findings demonstrate why an authorization must identify the version and why improved external evidence still does not remove the need for local validation and workflow evaluation.”
    Quote in the paperSubstantial claim
30 of 30

  • The paper relies on this without support.

    AI is increasingly entering hospital practice through diagnostic, predictive, workflow, operational, and generative applications.

    “Artificial intelligence (AI) is increasingly entering hospital practice through diagnostic, predictive, workflow, operational, and generative applications.”

  • The paper relies on this without support.

    Hospitals often govern AI systems as procurement or information-technology projects rather than as clinical-organizational interventions requiring indication, evaluation, monitoring, accountability, and withdrawal.

    “Hospitals often govern these systems as procurement or information-technology projects rather than as clinical-organizational interventions requiring indication, evaluation, monitoring, accountability, and withdrawal.”

  • The paper relies on this without support.

    Hospitals increasingly use AI in domains that may fall outside classic medical device regulation but still affect care quality, safety, equity, professional work, resource allocation, or institutional decisions.

    “Hospitals increasingly use AI in domains that may fall outside classic medical device regulation but still affect care quality, safety, equity, professional work, resource allocation, or institutional decisions.”

  • The paper relies on this without support.

    Many professionals already use public tools for summarization, translation, drafting, coding, teaching, literature review, and administrative tasks.

    “Many professionals already use public tools for summarization, translation, drafting, coding, teaching, literature review, and administrative tasks.”

  • The paper relies on this without support.

    Hospitals often accumulate digital tools without systematic retirement.

    “hospitals often accumulate digital tools without systematic retirement”

  • The paper relies on this without support.

    Accumulation of digital tools without systematic retirement leads to alert fatigue, redundant systems, hidden costs, outdated models, and unclear accountability.

    “This leads to alert fatigue, redundant systems, hidden costs, outdated models, and unclear accountability.”

  • The paper relies on this without support.

    Withdrawal of an AI system may be technically, commercially, clinically, and politically difficult, particularly after users and workflows become dependent on a tool.

    “Withdrawal may be technically, commercially, clinically, and politically difficult, particularly after users and workflows become dependent on a tool.”

  • The paper relies on this without support.

    Sudden removal of an AI system may itself create risk.

    “Sudden removal may itself create risk.”

  • The paper relies on this without support.

    Purchase price is only one component of the total cost of an AI system.

    “Purchase price is only one component of the total cost.”

  • The paper relies on this without support.

    A nominally inexpensive AI system may be costly if it generates unnecessary work or creates vendor lock-in.

    “A nominally inexpensive system may be costly if it generates unnecessary work or creates vendor lock-in.”

  • The paper relies on this without support.

    Without a formulary, responsibility can be diffused among vendors, clinicians, information technology teams, innovation units, and executives.

    “Without a formulary, responsibility can be diffused among vendors, clinicians, information technology teams, innovation units, and executives.”

  • The paper relies on this without support.

    Smaller hospitals may lack the expertise to operate a full committee.

    “Smaller hospitals may lack the expertise to operate a full committee”

  • The paper relies on this without support.

    Vendors may resist transparency.

    “Vendors may resist transparency.”

  • The paper relies on this without support.

    Clinicians may perceive review as bureaucracy unless timelines and requirements are proportional.

    “Clinicians may perceive review as bureaucracy unless timelines and requirements are proportional.”

  • The paper relies on this without support.

    Monitoring may be constrained by weak data infrastructure, delayed outcomes, and incomplete demographic data.

    “Monitoring may be constrained by weak data infrastructure, delayed outcomes, and incomplete demographic data.”

  • The paper relies on this without support.

    Cloud concentration and vendor lock-in can limit institutional leverage.

    “Cloud concentration and vendor lock-in can limit institutional leverage.”

  • The paper relies on this without support.

    Hospitals already maintain governance mechanisms for medicines, devices, infection control, transfusion, radiation safety, clinical pathways, and patient safety.

    “hospitals already maintain governance mechanisms for medicines, devices, infection control, transfusion, radiation safety, clinical pathways, and patient safety.”

  • The paper relies on this without support.

    Cloud-based AI can distribute data and control across hospitals, vendors, hyperscale cloud providers, and sub-processors.

    “Cloud-based AI can distribute data and control across hospitals, vendors, hyperscale cloud providers, and sub-processors.”
18 of 18

Cites

Flagged first, then by citations

Authors
Epelde, F.
Affiliations
Universitat Autònoma de Barcelona (ES)
Corresponding author
Epelde, Francisco
ORCID
Epelde, Francisco
Venue
Hospitals
Publisher
MDPI AG
Published
2026-07-22
Volume
3
Issue
3
Pages
15
ISSN
2813-4524 (ISSN-L)
DOI
10.3390/hospitals3030015
OpenAlex
W7170037529
Cited by
0
References
39
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
Hospitals (publishedVersion, cc-by)
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Health Sciences › Health Professions › Health Information Management › Healthcare Regulation
Keywords
AI governance, AI lifecycle management, data sovereignty, cloud security, AI risk management, generative AI in healthcare

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-27.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-27.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review