CoChat

Translational Potential and Explainability of Artificial Intelligence-Based Clinical Decision Support for Adults in Intensive Care: A Scoping Review

Open access
Pranata, Yodha, Mirwanti, Ristina, Bahtiar, Yusuf Achmad, Purwanti, Hesti · 2026 · Journal of Multidisciplinary Healthcare

The DOI matches its Crossref record; 77 of the 77 references it cites were found in a registry or an index; 1 of the 59 sources its claims cite were checked for support.

Abstract

This scoping review systematically mapped the translational potential and characteristics of explainable artificial intelligence (XAI) strategies in AI/ML-based clinical decision support tools for adult intensive care unit (ICU) settings. Methods: Following PRISMA-ScR guidelines, we searched Web of Science, PubMed, Scopus, and EBSCOhost up to January 2026. Translational potential was staged using an ICU-adapted, nine-level Technology Readiness Level (TRL) framework, and explainability strategies were classified as post-hoc or inherently interpretable (glass-box) to assess methodological transparency and clinical readiness. Results: A total of 808 records were identified, of which 29 studies met the inclusion criteria. The findings revealed a marked retrospective predominance (86.2%) and reliance on North American data, predominantly MIMIC (Medical Information Mart for Intensive Care). Tree-based ensembles (82.8%) and post-hoc SHAP explanations (86.2%) were dominant, with proposed clinical utility spanning three domains: therapeutic guidance, resource-allocation optimisation, and user-centric design. Most innovations were standalone, web-based prototypes requiring manual data entry (69.0%, TRL 4–5); a further 10.3% were shared only as open-source code, and only 17.2% reported integration with hospital systems. Only one study claimed clinical maturity (TRL 9), although its validation remained retrospective. Conclusion: Accuracy is no longer the primary bottleneck; the constraint has shifted to "last-mile" integration and external validity. Current XAI relies almost entirely on post-hoc methods that risk an "illusion of clarity", while inherently interpretable, glass-box models remain a rare but promising alternative. Future research should prioritise external validation in independent settings, prospective evaluation of clinical impact, and explicit comparison between post-hoc and interpretable approaches.

Overview

CoChat Summary

TL;DR

This scoping review mapped 29 AI/ML clinical decision support studies for adult ICU settings and found that most were retrospective, used tree-based ensembles, and relied on post-hoc SHAP explanations. The main bottleneck was not predictive accuracy but translational readiness: most tools were standalone prototypes, with only one study claiming clinical maturity and only 17.2% reporting hospital-system integration.

Key findings

  • From 808 records, 29 studies met inclusion criteria; 25/29 (86.2%) were retrospective, and sample sizes ranged from 237 to 28,819 patients.
  • Tree-based ensemble methods dominated the corpus at 24/29 (82.8%); deep learning appeared in 4 studies (13.8%), and only 1 study used a genuinely glass-box model (Explainable Boosting Machine).
  • SHAP was the dominant explainability method, used in 25/29 studies (86.2%); most explanations were post-hoc rather than inherently interpretable.
  • Translational maturity was low: 20/29 studies (69.0%) were web-based prototypes or online calculators (TRL 4–5), 3/29 (10.3%) shared open-source code, 5/29 (17.2%) integrated with hospital systems (TRL 7), and only 1/29 (3.4%) claimed clinical maturity (TRL 9), though it was validated retrospectively.

Methods

  • Scoping review following PRISMA-ScR, searching Web of Science, PubMed/MEDLINE, Scopus, and EBSCOhost up to January 2026, with backward/forward citation tracking.
  • Included primary quantitative studies on adult ICU AI/ML decision support that reported at least one translational indicator (eg, code, web tool, dashboard, calculator, or workflow/EHR integration); 2 reviewers screened/extracted, with additional review and supervision.
  • Narrative synthesis across four domains: model architecture, explainability strategy, translational potential mapped to a nine-level ICU-adapted TRL framework, and proposed clinical utility.

Limitations

  • Stated in the paper: no formal critical appraisal of study quality or risk of bias was performed; no meta-analysis was possible because of heterogeneity; no Cohen’s kappa/inter-reviewer agreement statistic was calculated.
  • Stated in the paper: eligibility required at least one translational indicator, so the review may understate early-stage, non-deployed AI research in ICU XAI.
  • Stated in the paper: the search strategy did not include explicit terms for inherently interpretable/glass-box models, so such studies may have been under-ascertained relative to post-hoc methods.
  • Evident from the text: the evidence base is concentrated in a narrow set of datasets and countries, heavily relying on MIMIC/eICU and Chinese studies, which the authors note limits generalisability.

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

Integrity signals

Registry

Crossref · 10.2147/jmdh.s621469

Cites

77 of 77 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Prisma Flow Diagram.47
Figure 1 Prisma Flow Diagram.47CC BY-NC · PubMed Central
Figure 2 The Translational Landscape of Explainable AI in Intensive Care.
Figure 2 The Translational Landscape of Explainable AI in Intensive Care.CC BY-NC · PubMed Central

Verification

Claims the paper makes

1 Valid · 0 Not valid · 44 Not checked · 89 Own results and interpretations

61 more claims

Create a free account to see all 134 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 28
  • It may state two separate claims as one. 51
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Citations

Cites

Metadata

Authors
Pranata, Y., Mirwanti, R., Bahtiar, Y. A., & Purwanti, H.
Affiliations
  • Indonesia Open University (ID)
  • Padjadjaran University (ID)
Corresponding author
Mirwanti, Ristina
Venue
Journal of Multidisciplinary Healthcare
Published
2026-07
Volume
Volume 19
Pages
1-16
ISSN
1178-2390 (ISSN-L)
OpenAlex
W7169497460
Cited by
1
References
69
Citations by year
2026: 1
FWCI
0.46
Type
journal article
Language
en
Open access
Locations
  • Journal of Multidisciplinary Healthcare (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by-nc)
  • DOAJ (DOAJ: Directory of Open Access Journals) (acceptedVersion)
  • Dove Medical Press (Taylor and Francis Group) (submittedVersion, cc-by-nc)
Funders
Lembaga Pengelola Dana Pendidikan (4303/ B3/DT.03.08/2025, 3927/UN6. RKT/HK.07.00/2025)
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
Keywords
explainable AI, clinical decision support, intensive care unit, technology readiness level, translational potential, post-hoc explanation, SHAP, tree-based ensemble models, external validation, clinical integration, resource allocation, adult critical care

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.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

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

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.