The DOI matches its Crossref record; 44 of the 44 references it cites were found in a registry or an index; 20 of the 25 sources its claims cite were checked for support.
A quantitative analysis of global AI medical studies: gaps in randomized controlled trials
Open accessAbstract
Medical artificial intelligence (AI) has advanced rapidly, yet a comprehensive quantitative overview of its clinical evaluation landscape remains lacking. We conducted a scoping review of 218 systematic reviews published between September 2023 and September 2024, from which 4667 primary studies were identified and classified by research stage, geography, specialty, and study design. Most studies were preclinical (88.2%, 4114/4667), while only 2.4% (113/4667) were randomized controlled trials (RCTs). Research was highly geographically concentrated: the top 10 contributing countries accounted for 75.5% of total country contributions, of which the United States and China contributed 47.5%. Among all primary studies, neoplasms (32.5%, 1518/4667) and musculoskeletal disorders (14.1%, 656/4667) were the most common, whereas digestive (38.1%, 43/113) and circulatory diseases (12.4%, 14/113) accounted for most RCTs. Among RCTs, 67.3% (76/113) were single-center, and 71.7% (81/113) did not report adherence to any reporting guideline. Overall, 82.3% (93/113) of RCTs reported favorable outcomes, although methodological concerns were common, particularly in allocation concealment and blinding. These findings indicate that medical AI research remains heavily skewed toward early-stage development, with limited high-quality clinical evidence. Strengthening trial design, multicenter collaboration, and reporting transparency will be critical to support the safe and equitable integration of AI into clinical practice.
Overview
CoChat Summary
TL;DR
This scoping review of 218 systematic reviews identified 4,667 medical AI primary studies and found a heavily preclinical evidence base: 88.2% were preclinical and only 2.4% were RCTs. RCTs were mostly single-center, often lacked reporting-guideline adherence, and had favorable outcomes but common allocation-concealment and blinding concerns.
Key findings
- Among 4,667 primary studies, 4,114 (88.2%) were preclinical development/offline validation, 440 were early live clinical evaluation, and only 113 (2.4%) were RCTs.
- Research was geographically concentrated: the top 10 countries accounted for 4,152/5,499 country contributions (75.5%), with the United States (1,393/5,499, 25.3%) and China (1,220/5,499, 22.2%) together contributing 47.5%.
- Neoplasms (1,518/4,667, 32.5%) and musculoskeletal disorders (656/4,667, 14.1%) dominated overall; among RCTs, digestive (43/113, 38.1%) and circulatory diseases (14/113, 12.4%) were most common, while neoplasms were only 12/113 (10.6%).
- RCTs were often single-center (76/113, 67.3%), had median sample size 278 (IQR 80–800), and 81/113 (71.7%) did not specify any reporting guideline; 93/113 (82.3%) reported favorable primary outcomes, but allocation concealment was low risk in only 28/113 (24.8%) and blinding of healthcare providers was probably high risk in 98/113 (86.7%).
Methods
- Scoping review following PRISMA-ScR, prospectively registered on OSF. A two-stage search first identified systematic reviews of medical AI in clinical settings published September 20, 2023–September 20, 2024 in PubMed, Embase, and Cochrane Library; primary studies cited in those reviews were extracted, de-duplicated, and screened, yielding 4,667 included primary studies from 2012–2024.
- Four researchers independently charted data on research stage, ICD-11 specialty, center type, sample size, country, AI role, and RCT-specific features including design, primary endpoint, comparator, reporting standards, and primary result. RCT risk of bias was assessed with ROBUST-RCT.
- Analyses were descriptive: counts and percentages for categorical variables, medians with interquartile ranges for continuous variables, subgroup analyses by research stage and specialty, and temporal trends; R version 4.2.2 was used.
Limitations
- Stated: extracting primary studies partly from recent systematic reviews may introduce selection/citation bias; publication bias may underrepresent negative or non-significant studies and overestimate favorable RCT results; English-language restriction may introduce language bias; categorization of reasons for non-favorable RCT outcomes relied on authors’ self-reported explanations and may be subjective; systematic-review lag and the time restriction may miss very recent research. The authors note the design likely underestimates total eligible studies more than RCTs, so the reported RCT proportion is likely conservative.
- Stated: the time restriction applied to the systematic-review search may narrow the scope of included primary studies.
- Evident: the review does not report a direct primary-study search or independent quality appraisal of the 218 included systematic reviews, so the mapped landscape is conditional on those reviews’ coverage and eligibility decisions; risk-of-bias assessment was performed only for RCTs, leaving methodological quality of the non-RCT majority uncharacterized.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
10 Valid · 0 Not valid · 2 Not checked · 111 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Second, there are also concerns about the quality and transparency of published clinical trials, which often lack the methodological rigor and detailed reporting necessary to inspire clinical confidence and enable replication9–11.”
§ A quantitative analysis of global AI medical studies: gaps in randomized controlled trials > Introduction
What the sources say
- Partly supportedcites [9] Martindale, A. P. L. et al. Concordance of randomised controlled trials for artificial intelligence interventions with the CONSORT-AI reporting guidelines. Nat. Commun. 15, 1619 (2024). sourceExperimental — accuracy not yet measured
The Discussion reports that several AI-specific details were poorly reported, including algorithm versions and code accessibility, and says these gaps may impede independent evaluation and replication. But it also finds generally high reporting concordance, while the Methods state that risk of bias was not assessed. The paper therefore supports concerns about specific reporting gaps, not the broader assertion that published trials often lack methodological rigor. (LLM, conf=0.87)
- Supportedcites [10] Liu, X., Cruz Rivera, S., Moher, D., Calvert, M. J. & Denniston, A. K. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. Lancet Digit Health 2, e537–e548 (2020). source
The Introduction says that AI trials have prompted concerns about their design and reporting. The Discussion says that details critical for independent evaluation or replication are often missing from trial reports, and links transparent reporting to trustworthy results. These passages support the citing sentence’s characterization of the concerns. (LLM, conf=0.89)
- Partly supportedcites [11] Sokol, K., Fackler, J. & Vogt, J. E. Artificial intelligence should genuinely support clinical reasoning and decision making to bridge the translational gap. NPJ Digit Med. 8, 345 (2025). sourceExperimental — accuracy not yet measured
The ‘Medical artificial intelligence adoption challenges’ section (page unavailable) links poor adoption to pervasive reproducibility issues and clinician mistrust. It does not say that published clinical trials often lack methodological rigor or detailed reporting, so the trial-specific part of the claim is unsupported. (LLM, conf=0.78)
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.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Most AI models therefore had not yet undergone the clinical scrutiny necessary for safe translation into practice13.”
§ A quantitative analysis of global AI medical studies: gaps in randomized controlled trials > Discussion
What the sources say
- Partly supportedcites [13] Muralidharan, V. et al. A scoping review of reporting gaps in FDA-approved AI medical devices. NPJ digital Med. 7, 273 (2024). sourceExperimental — accuracy not yet measured
The Results (no page provided) report substantial gaps in the public evidence for FDA-approved AI devices: 46.1% of approval summaries gave detailed performance-study results, and 1.9% linked to a published safety-and-efficacy study. But 70.4% contained a pre-approval performance study. The review establishes inadequate reporting and limits on assessing safety, not that most AI models had not undergone the necessary clinical scrutiny; its scope is also FDA-approved devices, not AI models generally. (LLM, conf=0.86)
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.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“indicating that the United States and China together accounted for approximately half of all AI medical research contributions worldwide29,30.”
§ A quantitative analysis of global AI medical studies: gaps in randomized controlled trials > Discussion
What the sources say
- Partly supportedcites [29] Lin, M., Lin, L., Lin, L., Lin, Z. & Yan, X. A bibliometric analysis of the advance of artificial intelligence in medicine. Front Med. (Lausanne) 12, 1504428 (2025). sourceExperimental — accuracy not yet measured
The Results (§3.2) list 708 US and 371 Chinese publications, and the Abstract reports 1,811 publications in total. Together, those counts are about 60%, rather than approximately half. The Conclusion (§5) also limits the analysis to papers indexed in WoSCC from 2019 to 2023, not all AI medical research worldwide. (LLM, conf=0.87)
- Not checked — no full text of the sourcecites [30] Jimma, B. L. Artificial intelligence in healthcare: A bibliometric analysis. Telemat. Inform. Rep. 9, 100041 (2023). source
hedged- 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.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“biomedical research is more typically concentrated across multiple leading countries rather than two alone32–35”
§ A quantitative analysis of global AI medical studies: gaps in randomized controlled trials > Discussion
What the sources say
- Supportedcites [32] Fontelo, P. & Liu, F. A review of recent publication trends from top publishing countries. Syst. Rev. 7, 147 (2018). source
The Discussion reports that 30 countries accounted for an average of 93.9% of PubMed publications over 20 years. The Results also identifies a top 10 publishing countries in 2015, supporting the claim that publication output is concentrated across multiple leading countries rather than just two. (LLM, conf=0.90)
- Partly supportedcites [33] Wondimagegn, D. et al. Faster, higher, stronger - together? A bibliometric analysis of author distribution in top medical education journals. BMJ Glob. Health 8, e011656 (2023). sourceExperimental — accuracy not yet measured
The Results report that authors from five countries held 70% of first or last authorship positions in the medical education journals studied. The Introduction also reports a comparison with biomedical sciences, but gives no country count for biomedical research generally. The source therefore supports concentration across more than two countries in medical education, not the broader claim about biomedical research. (LLM, conf=0.82)
- Partly supportedcites [34] Nunes, A. R. & Dale, J. Global trends in sustainable healthcare research: A bibliometric analysis. Future Healthc. J. 12, 100251 (2025). sourceExperimental — accuracy not yet measured
The Results (Geographical distribution of publications) identify the USA, England and Australia as the three leading contributors to sustainable healthcare research, supporting the point that leadership is not confined to two countries. The paper does not establish that this pattern is typical of biomedical research generally, or compare it with a two-country pattern. (LLM, conf=0.83)
- Partly supportedcites [35] Wang, X. et al. From past to future: Bibliometric analysis of global research productivity on nomogram (2000–2021). Front. Public Health 10, 2022 (2022). sourceExperimental — accuracy not yet measured
The Results report nomogram articles from 74 countries and regions and identify contributors beyond China and the United States, including South Korea. That supports a multi-country pattern for nomogram research, but the paper does not establish what is typical of biomedical research generally or compare it with research concentrated in two countries. (LLM, conf=0.83)
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.
- 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
“440 as Early Live Clinical Evaluation”
§ A quantitative analysis of global AI medical studies: gaps in randomized controlled trials > Results > Distribution of studies across research stages
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.
76 more claims
Create a free account to see all 123 claims and to verify the 1 reference still open.
- It may state two separate claims as one. 32
- Probably not a substantial claim. 21
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
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
“| Others | 25 | 22.1% |”
§ A quantitative analysis of global AI medical studies: gaps in randomized controlled trials > Results > Distribution of AI application roles
Reported 22.1 · recomputed 17.857143
Citations
Cites
PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation
Tricco, A. C., Lillie, E., Zarin, W., O'Brien, K. K., Colquhoun, H., Levac, D., Moher, D., Peters, M. D. J., Horsley, T., Weeks, L., et al. (2018) · Annals of Internal Medicine · doi:10.7326/m18-0850 · Cited by 43,565 · cited 1×
“This scoping review followed the PRISMA-ScR guidelines 42 , and the protocol was prospectively registered on the Open Science Framework (OSF) Registry (https://doi.org/10.17605/OSF.IO/YF6XJ).”
InfluentialHigh-performance medicine: the convergence of human and artificial intelligence
Topol, E. J. (2019) · Nature Medicine · doi:10.1038/s41591-018-0300-7 · Cited by 10,450 · cited 1×
“This has fueled optimism about AI ’ s potential to revolutionize diagnostics, treatment, and patient care 5 .”
A guide to deep learning in healthcare
Esteva, A., Robicquet, A., Ramsundar, B., Kuleshov, V., DePristo, M., Chou, K., Cui, C., Corrado, G., Thrun, S., & Dean, J. (2019) · Nature Medicine · doi:10.1038/s41591-018-0316-z · Cited by 5,312
Rajpurkar, P., Chen, E., Banerjee, O., & Topol, E. J. (2022) · Nature Medicine · doi:10.1038/s41591-021-01614-0 · Cited by 3,212
CONSORT 2010 statement: Updated guidelines for reporting parallel group randomised trials
Schulz, K. F., Altman, D. G., & Moher, D. (2010) · Journal of Pharmacology and Pharmacotherapeutics · doi:10.4103/0976-500x.72352 · Cited by 2,165
Liu, X., Faes, L., Kale, A. U., Wagner, S. K., Fu, D. J., Bruynseels, A., Mahendiran, T., Moraes, G., Shamdas, M., Kern, C., et al. (2019) · The Lancet Digital Health · doi:10.1016/s2589-7500(19)30123-2 · Cited by 1,883
Nagendran, M., Chen, Y., Lovejoy, C. A., Gordon, A. C., Komorowski, M., Harvey, H., Topol, E. J., Ioannidis, J. P. A., Collins, G. S., & Maruthappu, M. (2020) · BMJ · doi:10.1136/bmj.m689 · Cited by 1,152 · cited 2×
“First, the evidencebase is heavily imbalanced, dominated by a vast number of early-stage, retrospective validation studies, while robust evidence from prospective, randomized controlled trials (RCTs) remains scarce 8 .”
Vasey, B., Nagendran, M., Campbell, B., Clifton, D. A., Collins, G. S., Denaxas, S., Denniston, A. K., Faes, L., Geerts, B., Ibrahim, M., et al. (2022) · Nature Medicine · doi:10.1038/s41591-022-01772-9 · Cited by 1,039
A governance model for the application of AI in health care
Reddy, S., Allan, S., Coghlan, S., & Cooper, P. (2020) · Journal of the American Medical Informatics Association · doi:10.1093/jamia/ocz192 · Cited by 979
Cruz Rivera, S., Liu, X., Chan, A.-W., Denniston, A. K., Calvert, M. J., The SPIRIT-AI and CONSORT-AI Working Group, SPIRIT-AI and CONSORT-AI Steering Group, Darzi, A., Holmes, C., Yau, C., et al. (2020) · Nature Medicine · doi:10.1038/s41591-020-1037-7 · Cited by 933
Research works printed but not linked: 1
- Longhurst, C. A., Singh, K., Chopra, A., Atreja, A. & Brownstein, J. S. A Call for Artificial Intelligence Implementation Science Centers to Evaluate Clinical Effectiveness. 1, https://doi.org/10.1056/Alp2400223 (2024). · doi:10.1056/alp2400223
Cited by
Yadav, K., & Gupta, T. (2026) · Journal of the Egyptian National Cancer Institute · doi:10.1186/s43046-026-00399-y
“In a 2026 scoping review of 4,667 primary studies, 88.2% were preclinical, and only 2.4% were randomized controlled trials; although neoplasms accounted for 32.5% of all AI studies, oncology represented only 10.6% of the RCTs [90].”
Influential
Metadata
- Authors
- Wang, Q., Yang, Y., Qiu, L., Xu, Y., Sun, R., Xue, P., & Qiao, Y.
- Affiliations
- Chinese Academy of Medical Sciences & Peking Union Medical College (CN)
- Xiamen University (CN)
- Corresponding author
- Xue, Peng
- Qiao, Youlin
- Venue
- npj Digital Medicine
- Published
- 2026-04-30
- Volume
- 9
- Issue
- 1
- ISSN
- 2398-6352 (ISSN-L)
- OpenAlex
- W7159600031
- Cited by
- 4
- References
- 39
- Citations by year
- 2026: 4
- FWCI
- 7.78
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by-nc-nd
- creativecommons.org/licenses/by-nc-nd/4.0 (tdm)
- creativecommons.org/licenses/by-nc-nd/4.0 (vor, 27 d)
- Locations
- npj Digital Medicine (publishedVersion, cc-by-nc-nd)
- PubMed (publishedVersion)
- PubMed Central (publishedVersion, cc-by-nc-nd)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- National Natural Science Foundation of China (82404286)
- Chinese Academy of Medical Sciences Innovation Fund for Medical Sciences (CIFMS 2021-I2M-1-004)
- Tencent Sustainable Social Value Inclusive Health Lab and through the ChongQing Tencent Sustainable Development Foundation (SD20240904145730)
- Topics
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Meta-analysis and systematic reviews
- Social Sciences › Library and Information Sciences › scientometrics and bibliometrics research
- Keywords
- artificial intelligence in healthcare, randomized controlled trials, research gaps, reporting guidelines, allocation concealment, blinding, methodological quality of trials, neoplasms, musculoskeletal disorders, gastrointestinal diseases, circulatory diseases
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 2026-10-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: A quantitative analysis of global AI medical studies: gaps in randomized controlled trials. https://app.cochat.ai/papers/10.1038/s41746-026-02698-z
- The paper
- Wang, Q., Yang, Y., Qiu, L., Xu, Y., Sun, R., Xue, P., & Qiao, Y. (2026). A quantitative analysis of global AI medical studies: gaps in randomized controlled trials. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02698-z
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.