The DOI matches its Crossref record; 35 of the 35 references it cites were found in a registry or an index; 11 of the 25 sources its claims cite were checked for support.
Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency
Open accessAbstract
Background: Artificial Intelligence (AI)-driven Decision Support Systems (DSSs) promise improvements in diagnostic accuracy and clinical workflow efficiency, but their adoption is hindered by inadequate confidence calibration, limited transparency, and poor alignment with real-world decision processes, which limit clinician trust and lead to high override rates. Methods: We developed and validated a dynamic scoring framework to enhance trust in AI-generated diagnoses by integrating AI confidence scores, semantic similarity measures, and transparency weighting into the override decision process using 6689 cardiovascular cases from the MIMIC-III dataset. Override thresholds were calibrated and validated across varying transparency and confidence levels, with override rate as the primary acceptance measure. Results: The implementation of this framework reduced the override rate to 33.29%, with high-confidence predictions (90–99%) overridden at a rate of only 1.7%, and low-confidence predictions (70–79%) at a rate of 99.3%. Minimal transparency diagnoses had a 73.9% override rate compared to 49.3% for moderate transparency. Statistical analyses confirmed significant associations between confidence, transparency, and override rates (p < 0.001). Conclusions: These findings suggest that enhanced transparency and confidence calibration can substantially reduce override rates and promote clinician acceptance of AI diagnostics. Future work should focus on clinical validation to optimize patient safety, diagnostic accuracy, and efficiency.
Overview
CoChat Summary
TL;DR
The paper proposes a dynamic scoring framework combining LLM confidence, semantic similarity, transparency, and diagnosis length to calibrate when clinicians should accept or override AI-generated cardiovascular diagnoses. In 6,689 MIMIC-III cases, it reduced override rates from 87.64% to 33.29%, with high-confidence and higher-transparency outputs overridden far less often.
Key findings
- Overall override rate was 33.29% (2,227/6,689), a reduction from an initial 87.64% before systematic parameter adjustments.
- Confidence was strongly graded: 70–79% confidence had a 99.3% override rate (147/148), 80–89% had 48.1% (2,040/4,243), and 90–99% had 1.7% (390/2,298).
- Transparency was inversely related to overrides: minimal transparency 73.9% (99/157), moderate transparency 49.3% (1,631/3,307), and high confidence 14.9% (495/3,325); χ² = 991.01, p < 0.0001, Cramér’s V = 0.40.
- Interaction analysis showed confidence was dominant; at moderate confidence, overrides fell with transparency from 55.4% (minimal) to 49.9% (moderate) to 42.3% (high explainability with high confidence), and high explainability plus 90–99% confidence produced the lowest override rates.
Methods
- Retrospective cohort from MIMIC-III: 6,689 adult cardiovascular disease cases (≥18 years) with at least one discharge summary, identified by ICD-9 codes for ischemic heart disease, heart failure, arrhythmias, cardiomyopathies, valvular heart disease, and hypertensive heart disease.
- AI diagnoses were generated by off-the-shelf Qwen 2.5 (temperature 0.1, top-p 0.9, max 256 tokens) and compared with true diagnoses using exact string matching and Universal Sentence Encoder semantic similarity; confidence was model-reported from 70–99%, transparency was self-rated as low/moderate/high, and length scored 1.0 for 30–100 characters or 0.7 otherwise.
- The composite score was Sfinal = wc×Sconf + ws×Ssim + wl×Slen, with transparency-specific weights and thresholds; override occurred if Sfinal fell below threshold. Analysis used override rates, chi-square tests, Tukey HSD pairwise comparisons, and Cramér’s V.
Limitations
- Stated by the paper: transparency relied on AI self-ratings rather than human expert evaluations; the retrospective MIMIC-III design limits generalizability; the scoring system’s complexity may challenge real-time implementation and increase latency; cardiovascular-specific MIMIC-III data may introduce domain bias; and the work was a proof-of-concept without real-time adaptive calibration.
- Evident from the text: there was no prospective clinical validation, no external multi-institutional validation, no human transparency ratings, and no measurement of response timing or latency.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures


Verification
Claims the paper makes
6 Valid · 3 Not valid · 10 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
“Its modular and adaptable design supports broad generalizability across diverse clinical domains beyond cardiovascular diagnoses, such as oncology, neurology, and infectious diseases [37,38,39].”
§ Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency > 5. Strengths and Limitations
What the sources say
- Could not verifycites [37] Kalani, 2024, Revolutionizing Neurology: The Role of Artificial Intelligence in Advancing Diagnosis and Treatment, Cureus sourceExperimental — accuracy not yet measured
The Introduction and Review discuss AI applications in neurology and the importance of models generalizing to diverse populations. Neither those passages nor searches for a modular framework and cross-domain use address whether the citing paper’s framework can generalize beyond cardiovascular diagnosis to oncology, neurology, and infectious diseases. (LLM, conf=0.75)
- Partly supportedcites [38] McKinney, 2020, International evaluation of an AI system for breast cancer screening, Nature sourceExperimental — accuracy not yet measured
The paper provides an oncology example: it evaluates an AI system for breast cancer screening, and ‘Generalisation across populations’ (p. 4) reports testing across UK and US settings. It does not establish that the citing document’s framework has a modular, adaptable design or generalizes across clinical domains such as neurology and infectious diseases. (LLM, conf=0.85)
- Partly supportedcites [39] Sarantopoulos, A., Mastori Kourmpani, C., Yokarasa, A.L., Makamanzi, C., Antoniou, P., Spernovasilis, N., and Tsioutis, C. (2024). Artificial Intelligence in Infectious Disease Clinical Practice: An Overview of Gaps, Opportunities, and Limitations. Trop. Med. Infect. Dis., 9. sourceExperimental — accuracy not yet measured
Section 3, “AI Applications in Clinical Medicine,” describes AI tools used in oncology and cardiovascular and neurologic imaging, and suggests that some might later be applied to infectious diseases. It does not discuss the citing paper’s framework, its modular design, or demonstrated generalizability; it says the cited Aidoc algorithms had not yet been applied to infectious disease. (LLM, conf=0.84)
plainstates a cause- 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.
- Could not verify
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“A recent study clearly demonstrated this issue, revealing that AI models missed approximately 66% of life-threatening injuries in hospitalized patients, underscoring the urgent need for robust calibration procedures [5].”
§ Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency > 1. Introduction
What the sources say
- Partly supportedcites [5] Pias, 2025, Low responsiveness of machine learning models to critical or deteriorating health conditions, Commun. Med. sourceExperimental — accuracy not yet measured
The Abstract (Results; page unavailable) reports that models tested for in-hospital mortality prediction failed to recognize 66% of injuries. But those results came from synthesized cases, not observed injuries in hospitalized patients, and the passage does not say the 66% figure applies specifically to life-threatening injuries. (LLM, conf=0.88)
boosted- 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
“AI systems initially intended to lower healthcare costs often end up requiring substantial human oversight, illustrating the complexities involved in integrating AI effectively into clinical workflows without sacrificing clinician autonomy or patient safety [9,10].”
§ Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency > 1. Introduction
What the sources say
- Not matchedcites [9] Tahir, D. (2025, May 18). Health Care AI, Intended to Save Money, Turns Out to Require a Lot of Expensive Humans. Available online: https://www.hcinnovationgroup.com/analytics-ai/news/55263070/experts-healthcare-ai-will-require-expensive-humans.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Partly supportedcites [10] Scott, 2024, Achieving large-scale clinician adoption of AI-enabled decision support, BMJ Health Care Inform. sourceExperimental — accuracy not yet measured
In “The tool must not incur excessive opportunity cost or environmental impacts” (page unavailable), the paper describes costs for clinician involvement and ongoing maintenance. Its governance section also calls for ongoing monitoring and auditing. It does not say that AI systems are initially intended to lower healthcare costs or that they often end up requiring substantial human oversight. (LLM, conf=0.82)
plainstates a cause- 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.
- Not matched
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
“however, no Minimal vs. Moderate pairing at the same confidence range was significant.”
§ Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency > 3. Results > 3.4. Interaction Between Transparency and Confidence
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.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The proposed framework significantly reduced clinician override rates from an initial 87.64% to 33.29%, highlighting substantial improvements in the alignment between AI-generated diagnoses and clinician assessments.”
§ Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency > 4. Discussion
boostedstates a cause- 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: unclear — It is unclear why the paper says it holds.
40 more claims
Create a free account to see all 76 claims and to verify the 1 reference still open.
- Probably not a substantial claim. 14
- It may state two separate claims as one. 22
- The quote only partly states this claim. 3
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 major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
“The override rate ultimately decreased to 1.7% (390 out of 2298 cases; 95% CI: 1.4–2.4%)”
§ Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency > 3. Results > 3.3. Override Rates by AI Confidence Ranges
Reported 1.7 · recomputed 16.971279
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
“For diagnoses with Minimal Transparency, the override rate was 73.9% (99 out of 157 cases; 95% CI: 66.5–80.1%)”
§ Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency > 3. Results > 3.2. Override Rates by Transparency Level
Reported 73.9 · recomputed 63.057325
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
§ Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency > 3. Results > 3.4. Interaction Between Transparency and Confidence
“the High Explainability + High Confidence with a 90–99% Confidence Range group showed the lowest override rate”
§ Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency > 3. Results > 3.4. Interaction Between Transparency and Confidence
“Meanwhile, the Transparency Level did not show a high effect as shown by an override rate of 0% with minimal explainability, 7.7% for moderate, and 1.4% for high explainability.”
Citations
Cites
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 · cited 1×
“…have successfully employed AI to diagnose pneumonia and COVID-19 using lung ultrasound imagery [11,12], detect diabetic retinopathy from retinal fundus images [13], and correctly classify benign versus malignant skin lesions [14], highlighting significant potential in diagnostic applications.”
High-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×
“Our findings corroborate recent literature underscoring the risk of false-positive diagnoses and unnecessary workload increases when AI confidence is inaccurately assessed or poorly communicated [26].”
MIMIC-III, a freely accessible critical care database
Johnson, A. E. W., Pollard, T. J., Shen, L., Lehman, L.-W. H., Feng, M., Ghassemi, M., Moody, B., Szolovits, P., Anthony Celi, L., & Mark, R. G. (2016) · Scientific Data · doi:10.1038/sdata.2016.35 · Cited by 8,141 · cited 1×
“It includes a wide range of patient data, including demographics, laboratory tests, imaging reports, clinical notes, discharge notes, and ICD-9 codes [20].”
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 · cited 1×
“…have successfully employed AI to diagnose pneumonia and COVID-19 using lung ultrasound imagery [11,12], detect diabetic retinopathy from retinal fundus images [13], and correctly classify benign versus malignant skin lesions [14], highlighting significant potential in diagnostic applications.”
Rajkomar, A., Dean, J., & Kohane, I. (2019) · New England Journal of Medicine · doi:10.1056/nejmra1814259 · Cited by 4,579 · cited 2×
“These results align well with previous research by Ghassemi et al. and Rajkomar et al., which similarly found that increased AI transparency enhances clinician trust and promotes more effective integration of AI systems into clinical practice [27,28].”
International evaluation of an AI system for breast cancer screening
McKinney, S. M., Sieniek, M., Godbole, V., Godwin, J., Antropova, N., Ashrafian, H., Back, T., Chesus, M., Corrado, G. S., Darzi, A., et al. (2020) · Nature · doi:10.1038/s41586-019-1799-6 · Cited by 3,287
Key challenges for delivering clinical impact with artificial intelligence
Kelly, C. J., Karthikesalingam, A., Suleyman, M., Corrado, G., & King, D. (2019) · BMC Medicine · doi:10.1186/s12916-019-1426-2 · Cited by 3,136 · cited 1×
“Key barriers to its adoption include insufficient transparency, inconsistent confidence calibration, and challenges in defining clear override mechanisms that effectively balance AI suggestions with clinician expertise [1,3].”
Explainability for artificial intelligence in healthcare: a multidisciplinary perspective
the Precise4Q consortium, Amann, J., Blasimme, A., Vayena, E., Frey, D., & Madai, V. I. (2020) · BMC Medical Informatics and Decision Making · doi:10.1186/s12911-020-01332-6 · Cited by 2,203 · cited 1×
“…responses, this study addresses one of the most significant impediments to the effective implementation of AI-decision support tools: low clinician trust due to uncertainty about their accuracy and reliability, while considering confidence calibrations to mitigate clinician overreliance [22,23].”
The false hope of current approaches to explainable artificial intelligence in health care
Ghassemi, M., Oakden-Rayner, L., & Beam, A. L. (2021) · The Lancet Digital Health · doi:10.1016/s2589-7500(21)00208-9 · Cited by 1,724 · cited 1×
“These results align well with previous research by Ghassemi et al. and Rajkomar et al., which similarly found that increased AI transparency enhances clinician trust and promotes more effective integration of AI systems into clinical practice [27,28].”
Universal Sentence Encoder for English
Cer, D., Yang, Y., Kong, S.-Y., Hua, N., Limtiaco, N., St. John, R., Constant, N., Guajardo-Cespedes, M., Yuan, S., Tar, C., et al. (2018) · Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing: System Demonstrations · doi:10.18653/v1/d18-2029 · Cited by 1,277
Research works printed but not linked: 1
- Guo, C., Pleiss, G., Sun, Y., and Weinberger, K.Q. (2017, January 6–11). On calibration of modern neural networks. Proceedings of the International Conference on Machine Learning, Sydney, Australia.
Cited by
A Survey of the Application of Explainable Artificial Intelligence in Biomedical Informatics
Eshkiki, H., Tanhaei, F., Caraffini, F., & Mora, B. (2025) · Applied Sciences · doi:10.3390/app152412934 · Cited by 10
Chow, J. C. L., & Li, K. (2026) · Information · doi:10.3390/info17040361 · Cited by 13
Artificial Intelligence in Cardiovascular Medicine: Focus on Hypertension
Varzideh, F., Mone, P., Kansakar, U., Pande, S., Jankauskas, S. S., & Santulli, G. (2026) · Hypertension · doi:10.1161/hypertensionaha.126.26094 · Cited by 6
Borra, S., Dey, N., Fong, S., Sherratt, R. S., & Shi, F. (2026) · Cancers · doi:10.3390/cancers18091361 · Cited by 8
Backcasting the Trust Gap: A Strategic Road Map for Clinician Adoption of AI Diagnostics by 2040
Yu, Y. (2026) · Journal of Medical Internet Research · doi:10.2196/94234 · Cited by 10
“Second, the 2040 vision state and its risk-stratified trust thresholds (≥90% for autonomous tasks; 70% ‐ 85% for assistive support) reflect expert judgment rather than empirical consensus [13]; future work should establish this vision through a formal multistake-holder Delphi process.”
InfluentialSelf-citation
Metadata
- Authors
- Yu, Y., Gomez-Cabello, C. A., Haider, S. A., Genovese, A., Prabha, S., Trabilsy, M., Collaco, B. G., Wood, N. G., Bagaria, S., Tao, C., & Forte, A. J.
- Affiliations
- Zyter|TruCare, Rockville, MD 20852, USA
- Mayo Clinic in Florida (US)
- Division of Plastic Surgery, Mayo Clinic, Jacksonville, FL 32224, USA
- Mayo Clinic (US)
- Corresponding author
- Forte, Antonio J.
- ORCID
- Venue
- Diagnostics
- Publisher
- MDPI AG
- Published
- 2025-08-30
- Volume
- 15
- Issue
- 17
- Pages
- 2204
- ISSN
- 2075-4418 (ISSN-L)
- PMC
- PMC12428550
- OpenAlex
- W4413880252
- Cited by
- 32
- References
- 32
- Citations by year
- 2026: 30 · 2025: 2
- FWCI
- 7.70
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Diagnostics (publishedVersion, cc-by)
- PubMed (publishedVersion)
- PubMed Central (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
- Keywords
- AI diagnostics, clinician trust, confidence calibration, decision support system, semantic similarity, MIMIC-III
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 2026-10-10, 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).
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: Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency. https://app.cochat.ai/papers/10.3390/diagnostics15172204
- The paper
- Yu, Y., Gomez-Cabello, C. A., Haider, S. A., Genovese, A., Prabha, S., Trabilsy, M., Collaco, B. G., Wood, N. G., Bagaria, S., Tao, C., & Forte, A. J. (2025). Enhancing Clinician Trust in AI Diagnostics: A Dynamic Framework for Confidence Calibration and Transparency. Diagnostics. https://doi.org/10.3390/diagnostics15172204
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.