Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit
Open accessThe DOI matches its Crossref record; 63 of the 63 references it cites were found in a registry or an index; 3 of the 68 sources its claims cite were checked for support.
Abstract
PURPOSE: Due to the increasing demand for intensive care unit (ICU) treatment, and to improve quality and efficiency of care, there is a need for adequate and efficient clinical decision-making. The advancement of artificial intelligence (AI) technologies has resulted in the development of prediction models, which might aid clinical decision-making. This systematic review seeks to give a contemporary overview of the current maturity of AI in the ICU, the research methods behind these studies, and the risk of bias in these studies. METHODS: A systematic search was conducted in Embase, Medline, Web of Science Core Collection and Cochrane Central Register of Controlled Trials databases to identify eligible studies. Studies using AI to analyze ICU data were considered eligible. Specifically, the study design, study aim, dataset size, level of validation, level of readiness, and the outcomes of clinical trials were extracted. Risk of bias in individual studies was evaluated by the Prediction model Risk Of Bias ASsessment Tool (PROBAST). RESULTS: Out of 6455 studies identified through literature search, 494 were included. The most common study design was retrospective [476 studies (96.4% of all studies)] followed by prospective observational [8 (1.6%)] and clinical [10 (2%)] trials. 378 (80.9%) retrospective studies were classified as high risk of bias. No studies were identified that reported on the outcome evaluation of an AI model integrated in routine clinical practice. CONCLUSION: The vast majority of developed ICU-AI models remain within the testing and prototyping environment; only a handful were actually evaluated in clinical practice. A uniform and structured approach can support the development, safe delivery, and implementation of AI to determine clinical benefit in the ICU.
Overview
CoChat Summary
TL;DR
This systematic review of 494 studies found that most artificial intelligence models for the intensive care unit remain in retrospective testing and prototyping, with few clinically evaluated and none evaluated after integration into routine clinical practice. The authors conclude that a uniform, structured approach is needed to move ICU-AI safely from development to bedside benefit.
Key findings
- Of 494 included studies, 476 (96.4%) were retrospective, 8 (1.6%) were prospective observational, and 10 (2%) were clinical studies; the clinical studies comprised 5 non-randomized trials and 5 randomized clinical trials.
- Risk of bias was high in 378 of 467 (80.9%) retrospective development studies assessed with PROBAST, most often in the “participants” and “analysis” domains.
- On the readiness scale, 441 studies (89.3%) scored level 4 or below, 35 (7.1%) reached external validation (level 5), 8 (1.6%) integrated an AI model clinically without exposing staff to results (level 6), and 10 (2%) clinically evaluated model performance (level 8); no studies reported outcome evaluation of an AI model integrated in routine clinical practice (level 9).
- Among 10 clinical studies, 8 provided complete information on study characteristics and patient outcomes, and 7 showed significant improvement in patient outcomes.
Methods
- Systematic review prepared according to PRISMA and registered in PROSPERO, with searches in Embase, MEDLINE ALL, Web of Science Core Collection, Cochrane Central Register of Controlled Trials, and Google Scholar from March to July 28, 2020.
- Eligible studies included adults (≥17 years), assessed AI or ML algorithms for clinical impact, used data gathered during ICU stay, were original research, and were available in English full text.
- Data extracted included study design, aim, dataset size, level of validation, AI readiness level adapted from NASA technology readiness levels, and clinical trial outcomes; PROBAST was used to assess risk of bias in retrospective development studies, and no quantitative synthesis was conducted.
Limitations
- The authors state that the review focused specifically on AI models developed in the ICU, so models translatable from other specialties may have been missed; PROBAST was designed for conventional prediction models rather than AI-based models; the literature search may have missed eligible studies; and PROBAST did not assess discrimination, calibration, algorithmic bias, or racial and minority bias.
- The text also makes evident an inconsistency in the reported number of studies identified through the electronic search: the abstract states 6455, while the Results section states 8645.
- The review did not assess applicability because no specific therapeutic question was defined, and it conducted no quantitative synthesis.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures
![Fig. 1 PRISMA 2009 flow diagram of the study review process and the exclusion of studies; from [19]](https://figures.cochat.ai/paper-figures/1e2aaf4a728b47b1b036f5890b8f3dec/moving-from-bytes-to-bedside-a-systematic-review-on-the-use-of-artificial-figure-1.jpg)


![Fig. 4 Percentage risk of bias according to the domain of assessment for all studies with a retrospective study design; assessed using PROBAST [23]](https://figures.cochat.ai/paper-figures/ae559657e7f648e0a8fbd2c14748f57d/moving-from-bytes-to-bedside-a-systematic-review-on-the-use-of-artificial-figure-4.jpg)
Verification
Claims the paper makes
3 Valid · 1 Not valid · 48 Not checked · 57 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Sepsis-related in hospital mortality rate decreased by 60.24% (p < 0.001)”
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Discussion
What the sources say
- Partly supportedcites [28] McCoy A, Das R (2017) Reducing patient mortality, length of stay and readmissions through machine learning-based sepsis prediction in the emergency department, intensive care unit and hospital floor units. BMJ Open Qual 6:e000158 sourceExperimental — accuracy not yet measured
The Results section (Figure 1 passage) confirms the magnitude: 'Average mortality over the entire post-implementation period was 2.93%, for a 60.24% (p<0.01) average reduction,' echoed in the Abstract (Results). However, the source reports the significance as p<0.01, not p<0.001 as the citing sentence states, so the statistical claim overstates the source. Note also that the 60.24% figure is a relative pre- vs post-implementation reduction in a single-hospital SIRS-criteria population (n=1328, Cape Regional Medical Center), which the citing sentence does not carry. (LLM, conf=0.80)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Significantly reduced average LOS (2.7 days on average; p = 0.042)”
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Discussion
What the sources say
- Partly supportedcites [33] Shimabukuro DW, Barton CW, Feldman MD, Mataraso SJ, Das R (2017) Effect of a machine learning-based severe sepsis prediction algorithm on patient survival and hospital length of stay: a randomised clinical trial. BMJ Open Respir Res 4:e000234 sourceExperimental — accuracy not yet measured
The source confirms the direction, significance and p-value: Table 2 (Results > Outcomes) reports hospital LOS of 13.0 (SE 1.23) days in controls vs 10.3 (SE 0.912) in the experimental arm with p=0.042, and the Discussion restates a 20.6% decrease (p=0.042). However, the magnitude the citing sentence attributes is 2.7 days, whereas the paper's own stated 'amount of reduction' is 2.30 days (Table 2); no passage in our copy reports a 2.7-day LOS reduction (the only 2.7-figure is 2.76 hours earlier antibiotics / 2.79 hours earlier blood cultures, Results > Outcomes). So the significance and p-value are supported but the quoted effect size is not the source's number. (LLM, conf=0.78)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Sepsis onset was predicted in advance in realtime”
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Discussion
What the sources say
- Supportedcites [33] Shimabukuro DW, Barton CW, Feldman MD, Mataraso SJ, Das R (2017) Effect of a machine learning-based severe sepsis prediction algorithm on patient survival and hospital length of stay: a randomised clinical trial. BMJ Open Respir Res 4:e000234 source
The Results ('Patient inclusion and baseline characteristics') state that 142 patients were randomised (75 control, 67 experimental) in this trial of a machine learning-based severe sepsis prediction algorithm. The Introduction describes a prospective randomised study in which the algorithm was applied to EHR data and 'generated real-time telephonic notifications', and its 'ability to identify patients with severe sepsis before onset'; the Discussion adds that it was designed 'to predict severe sepsis up to 4 hours in advance' and contrasts it with systems that 'detect, not predict'. Together these support that sepsis onset was predicted in advance in real time in this n=142 RCT. (LLM, conf=0.90)
plainn = 142- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Of all studies which reported on external validation, most of them analyzed data on 1000-10,000 patients [13 of 35 studies (37.1%)] (Fig. 2).”
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Study design and purpose of AI
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“441 studies (89.3%) scored level 4 or below on the 'level of readiness' scale, 35 (7.1%) studies performed external validation (level 5), 8 studies (1.6%) integrated an AI model in the clinical setting without exposing the clinical staff to the results (level 6) and ten studies (2%) clinically evaluated model performance (level 8).”
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Level of readiness
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
61 more claims
Create a free account to see all 109 claims and to verify the 64 references still open.
- It may state two separate claims as one. 18
- The quote only partly states this claim. 1
- Probably not a substantial claim. 31
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“A recent meta-analysis analyzed publications concerning the use of AI in the ICU: out of the 494 articles studied, the majority were retrospective studies 476 articles (96.4%). In total, 10 articles (10/494, 2%) were clinical studies, of which only 5 (5/494, 1%) were randomized trials with a control arm.”
From Expectations of Intensive Care Physicians Regarding an AI-Based Decision Support System for Weaning From Continuous Renal Replacement Therapy: Predevelopment Survey Study (2025) · cites this paper as [3]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Systematic reviews show that most ICU AI studies remain retrospective, with limited external validation and scarce evidence on patient-centered outcomes after real-world deployment [35].”
From Artificial Intelligence to Facilitate SEP-1 Measure Compliance and Fluid Management in Sepsis (2026) · cites this paper as [35]
The cited source says what the claim attributes to it.
In the citing paper
“These AI-based CDSS have demonstrated potential across several domains in ICU, including sepsis prediction, early detection of patient deterioration, and management of mechanical ventilation [2-4].”
The cited source says what the claim attributes to it.
In the citing paper
“In addition, the majority of retrospective studies presented a high risk of bias [3].”
In the citing paper
“This manuscript has been prepared according to the Preferred Reporting Items for Systematic reviews and Metaanalysis (PRISMA) guideline and was registered in the online PROSPERO database (Record ID: 199,683; reference number: CRD42020199863) before initiation of the literature search [19,20].”
From Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit (2021) · cites this paper as [20]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Abstract
“Out of 6455 studies identified through literature search, 494 were included.”
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Identification of studies
“A total of 8645 studies were identified through our electronic search”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“intervention group 38.41 days ± 3.35 vs. control group 43.69 days ± 14.89, with a difference on average of 5.2 days [p < 0.01]”
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Discussion
Reported 5.2 · recomputed 5.28
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Study design and purpose of AI
“classifying sub-populations [57 studies (11.7%)]”
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Conclusion
“Classifying sub-populations | 58 (11.7%)”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Study design and purpose of AI
“improving prognostic models/risk scoring systems [86 studies (17.4%)]”
§ Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit > Conclusion
“Improving prognostic models/risk scoring | 91 (18.4%)”
Citations
Cites
Preferred Reporting Items for Systematic Reviews and Meta-Analyses: the PRISMA statement
Moher, D., Liberati, A., Tetzlaff, J., Altman, D. G., & for the PRISMA Group (2009) · BMJ · doi:10.1136/bmj.b2535 · Cited by 83,823
Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement
Moher, D., Liberati, A., Tetzlaff, J., Altman, D. G., & The PRISMA Group (2009) · PLoS Medicine · doi:10.1371/journal.pmed.1000097 · Cited by 64,042
Preferred reporting items for systematic reviews and meta-analyses: the PRISMA Statement.
(2009) · PubMed Central · Cited by 44,859
The Elements of Statistical Learning
Hastie, T., Tibshirani, R., & Friedman, J. (2009) · Springer Series in Statistics · doi:10.1007/978-0-387-84858-7 · Cited by 25,970
Dermatologist-level classification of skin cancer with deep neural networks
Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., & Thrun, S. (2017) · Nature · doi:10.1038/nature21056 · Cited by 14,362
CONSORT 2010 Statement: updated guidelines for reporting parallel group randomised trials
the CONSORT Group, Schulz, K. F., Altman, D. G., & Moher, D. (2010) · BMC Medicine · doi:10.1186/1741-7015-8-18 · Cited by 13,371
The Elements of Statistical Learning: Data Mining, Inference, and Prediction
John Lu, Z. Q. (2010) · Journal of the Royal Statistical Society Series A: Statistics in Society · doi:10.1111/j.1467-985x.2010.00646_6.x · Cited by 12,390
CONSORT 2010 Statement: updated guidelines for reporting parallel group randomised trials
Schulz, K. F., Altman, D. G., Moher, D., & for the CONSORT Group (2010) · BMJ · doi:10.1136/bmj.c332 · Cited by 9,329
Gulshan, V., Peng, L., Coram, M., Stumpe, M. C., Wu, D., Narayanaswamy, A., Venugopalan, S., Widner, K., Madams, T., Cuadros, J., et al. (2016) · JAMA · doi:10.1001/jama.2016.17216 · Cited by 7,697
SPIRIT 2013 Statement: Defining Standard Protocol Items for Clinical Trials
Chan, A.-W., Tetzlaff, J. M., Altman, D. G., Laupacis, A., Gøtzsche, P. C., Krleža-Jerić, K., Hróbjartsson, A., Mann, H., Dickersin, K., Berlin, J. A., et al. (2013) · Annals of Internal Medicine · doi:10.7326/0003-4819-158-3-201302050-00583 · Cited by 7,363
Cited by
Ventilator-associated pneumonia: pathobiological heterogeneity and diagnostic challenges
Howroyd, F., Chacko, C., MacDuff, A., Gautam, N., Pouchet, B., Tunnicliffe, B., Weblin, J., Gao-Smith, F., Ahmed, Z., Duggal, N. A., et al. (2024) · Nature Communications · doi:10.1038/s41467-024-50805-z · Cited by 162
van de Sande, D., Van Genderen, M. E., Smit, J. M., Huiskens, J., Visser, J. J., Veen, R. E. R., van Unen, E., BA, O. H., Gommers, D., & Bommel, J. V. (2022) · BMJ Health & Care Informatics · doi:10.1136/bmjhci-2021-100495 · Cited by 135
Self-citationCai, Y., Cai, Y.-Q., Tang, L.-Y., Wang, Y.-H., Gong, M., Jing, T.-C., Li, H.-J., Li-Ling, J., Hu, W., Yin, Z., et al. (2024) · BMC Medicine · doi:10.1186/s12916-024-03273-7 · Cited by 122
Peer review of GPT-4 technical report and systems card
Gallifant, J., Fiske, A., Strekalova, Y. A., Osorio-Valencia, J. S., Parke, R. L., Mwavu, R., Martinez‐Martin, N., Gichoya, J. W., Ghassemi, M., Demner‐Fushman, D., et al. (2024) · PLOS Digital Health · doi:10.1371/journal.pdig.0000417 · Cited by 181
Machine learning in perioperative medicine: a systematic review
Bellini, V., Valente, M., Bertorelli, G., Pifferi, B., Craca, M., Mordonini, M., Lombardo, G., Bottani, E., Del Rio, P., & Bignami, E. (2022) · Journal of Anesthesia, Analgesia and Critical Care · doi:10.1186/s44158-022-00033-y · Cited by 109
Metadata
- Authors
- van de Sande, D., van Genderen, M. E., Huiskens, J., Gommers, D., & van Bommel, J.
- Affiliations
- Erasmus MC (NL)
- Corresponding author
- van Genderen, Michel E.
- Venue
- Intensive Care Medicine
- Published
- 2021-06-05
- Volume
- 47
- Issue
- 7
- Pages
- 750-760
- ISSN
- 0342-4642 (ISSN-L)
- 1432-1238
- PMC
- PMC8178026
- OpenAlex
- W3164057767
- Semantic Scholar
- e7ea75d3a5ce6931a02ccc916b79234fa90168c1
- Cited by
- 279
- References
- 59
- Citations by year
- 2026: 54 · 2025: 68 · 2024: 62 · 2023: 47 · 2022: 38 · 2021: 9
- FWCI
- 3.25
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by-nc
- creativecommons.org/licenses/by-nc/4.0 (tdm)
- creativecommons.org/licenses/by-nc/4.0 (vor)
- Locations
- Intensive Care Medicine (publishedVersion, cc-by-nc)
- PubMed Central (publishedVersion, cc-by-nc)
- PubMed (publishedVersion)
- EUR Research Repository (Erasmus University Rotterdam) (publishedVersion, other-oa)
- RePub (Erasmus University Rotterdam) (submittedVersion, other-oa)
- EUR Research Repository (Erasmus University Rotterdam) (submittedVersion, other-oa)
- Topics
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
- Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
- Keywords
- artificial intelligence, intensive care unit, clinical decision support, prediction model, risk of bias, PROBAST, retrospective analysis, clinical implementation of AI, AI model validation
- MeSH terms
- Artificial Intelligence, Humans, Intensive Care Units, Retrospective Studies, Observational Studies as Topic
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-11-17, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-17 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (qwen3.8-omni-flash).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit. https://app.cochat.ai/papers/10.1007/s00134-021-06446-7
- The paper
- van de Sande, D., van Genderen, M. E., Huiskens, J., Gommers, D., & van Bommel, J. (2021). Moving from bytes to bedside: a systematic review on the use of artificial intelligence in the intensive care unit. Intensive Care Medicine. https://doi.org/10.1007/s00134-021-06446-7
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.