CoChat
Sign Up
Cited by 0Artificial Intelligence in HealthcareMachine Learning in HealthcareDiabetes, Cardiovascular Risks, and Lipoproteins

AI-driven cardiovascular risk prediction in patients with diabetes: bridging algorithmic innovation to equitable clinical application

Open access
Li, Hongxuan, Xu, Zheyi, Cen, Yanhui, Liu, Xin·2026·Frontiers in Medicine
DOI 10.3389/fmed.2026.1831220PMC PMC13268988
Open this paper in CoChat — free accountOpen PDF

Abstract

Machine learning models hold promise to revolutionize cardiovascular disease (CVD) prediction in patients with type 2 diabetes, with algorithms such as neural networks demonstrating superior discriminative performance in internal validations. However, a systematic review has revealed that existing models generally carry a high risk of bias and exhibit poor adherence to transparent reporting standards, severely hindering their clinical translation and real-world application. Furthermore, current models are predominantly developed using populations from Europe and North America, resulting in a critical lack of representativeness for Asian populations, where the burden of cardiovascular disease is particularly heavy. This article argues that the field is undergoing a pivotal transition—from an exclusive focus on algorithmic performance to ensuring clinical equity and fairness. Future advancements should prioritize external validation, calibration-aware assessment, subgroup-specific performance reporting, and cautious integration of biologically plausible biomarkers rather than relying on discrimination alone. Only through this approach can machine learning-driven predictive tools truly bridge the gap between innovation and equitable clinical implementation, ultimately alleviating the global burden of diabetes-related cardiovascular complications.

AI snapshot

Written from the paper’s full text.

TL;DR

This perspective argues that machine learning for cardiovascular risk prediction in type 2 diabetes must shift from a narrow focus on discrimination toward transparency, external validation, calibration, and subgroup fairness, especially given the underrepresentation of Asian populations. It synthesizes evidence from systematic reviews and cohort studies to propose a translational framework bridging algorithmic performance to equitable clinical application.

Key findings

  • Machine learning models often outperform conventional scores in derivation cohorts, with reported AUCs around 0.75–0.91 across reviews and cohort studies, and a pooled AUC of 0.753 in one meta-analysis of AI-based cardiovascular prediction models.
  • Systematic evaluations reveal high risk of bias (participant selection, outcome measurement, and analysis domains) and poor transparency: mean adherence to the TRIPOD checklist was only 53.75% across reviewed studies.
  • Current models draw predominantly on European and North American populations, leaving Asian populations underrepresented; the authors stress that "Asian populations" are also heterogeneous, and evidence from two multi-ethnic Singapore cohorts shows established models remain useful only after local recalibration, with calibration differing across models and subgroups.
  • Oxidative stress and lipid peroxidation markers are biologically plausible candidate predictors linked to endothelial dysfunction and vascular injury in diabetes, but evidence is heterogeneous in marker selection, assay platform, and standardization, and direct comparative evidence against standard clinic-based models remains limited.

Methods

  • Design: a narrative perspective/review article (not a primary study) synthesizing published systematic reviews, meta-analyses, and cohort studies on machine learning-based cardiovascular risk prediction in type 2 diabetes.
  • Data: no original dataset; the paper draws on cited literature, including the IDF Diabetes Atlas (approximately 589 million adults aged 20–79 with diabetes worldwide in 2024; at least USD 1 trillion in diabetes-related health expenditure in 2024), WHO figures (approximately 80% of global CVD deaths in low- and middle-income countries), and studies from Malaysia, China, and Singapore.
  • Analysis: thematic synthesis organized around model performance, external validation, population representativeness, reporting quality, and biomarker integration (summarized in Table 1), with reference to TRIPOD+AI and PROBAST+AI frameworks.

Limitations

  • The paper states no formal limitations section; as a perspective article it presents no original data, so all quantitative claims are inherited from the cited reviews and cohorts.
  • Evident from the text: the biomarker argument rests on biological plausibility rather than demonstrated incremental predictive value, and the authors themselves note that comparative predictive evidence for lipid peroxidation markers remains limited and heterogeneous in assay and standardization.
  • Evident from the text: the proposed framework (external validation, local recalibration, subgroup-aware evaluation, transparent reporting) is a set of recommendations rather than an empirically tested intervention, and the cited regional studies (Malaysia, China, Singapore) are described as still requiring broader external validation.

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

Integrity signals

Registry

Crossref · 10.3389/fmed.2026.1831220

Cites

22 of 22 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

68 of 75 cited sources checked · 33 supported · 1 flagged · accuracy not yet measured

  • Premisecitation

    A recent meta-analysis reported a pooled AUC of 0.753 across AI-based cardiovascular prediction models.

    “Neural networks and gradient boosting approaches have shown promising performance in several cohorts, and a recent meta-analysis reported a pooled AUC of 0.753 across AI-based cardiovascular prediction models (8).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [8] Liu, 2025, Machine learning based prediction models for cardiovascular disease risk using electronic health records data: systematic review and meta-analysis, Eur Heart J Digit Health source

      The Results > Meta-analysis section (Table 5; page not provided) reports a pooled AUC of 0.815 for the 32 machine-learning models, not 0.753. It reports 0.765 for conventional models. (LLM, conf=0.90)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured
  • Premisecitation

    Neural networks and gradient boosting machines exhibit superior discriminative ability over traditional methods in predicting cardiovascular complications in diabetes.

    “Numerous studies have demonstrated that algorithms such as neural networks and gradient boosting machines exhibit superior discriminative ability over traditional methods in predicting cardiovascular complications in diabetes (6–8).”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites [6] Kee O, 2023, Cardiovascular complications in a diabetes prediction model using machine learning: a systematic review, Cardiovasc Diabetol source
    • cites [7] Cai, 2024, Artificial intelligence in the risk prediction models of cardiovascular disease and development of an independent validation screening tool: a systematic review, BMC Med. source

      The Background (page not available) says AI-driven cardiovascular risk prediction has shown better performance than traditional models, and the Results identify studies involving type 2 diabetes and neural networks. But our copy does not establish superior discriminative ability for both neural networks and gradient boosting machines specifically when predicting cardiovascular complications in diabetes. (LLM, conf=0.82)

      Experimental — accuracy not yet measured
    • cites [8] Liu, 2025, Machine learning based prediction models for cardiovascular disease risk using electronic health records data: systematic review and meta-analysis, Eur Heart J Digit Health source

      The Results (Characteristics of included studies; page not specified) report that deep-learning models, which include neural networks, and gradient-boosting models were often the best-performing models; the meta-analysis also finds better discrimination for ML models overall than conventional baselines. But the Methods (Eligibility criteria; page not specified) explicitly exclude studies focused on patients with specific high-risk conditions such as diabetes. The review therefore does not establish this advantage for predicting cardiovascular complications in diabetes. (LLM, conf=0.92)

      Experimental — accuracy not yet measured
  • Premisecitation

    Machine learning models can improve discrimination or reclassification in derivation datasets, especially when longitudinal or high-dimensional EHR features are available.

    “Rather than focusing on isolated “best-case” results, the more consistent message across the literature is that machine learning models can improve discrimination or reclassification in derivation datasets, especially when longitudinal or high-dimensional EHR features are available (8, 11–13).”
    Quote in the paperSubstantial claim3 of 4 sources checked
    • cites [8] Liu, 2025, Machine learning based prediction models for cardiovascular disease risk using electronic health records data: systematic review and meta-analysis, Eur Heart J Digit Health source

      The Meta-analysis reports higher pooled discrimination for random forest and deep learning models than for conventional risk scores, and the Characteristics of included studies says most studies used internal training/test splits or cross-validation. That section notes that four studies reported net reclassification improvement, but does not establish that reclassification improved; the paper also does not show that gains were especially pronounced with longitudinal or high-dimensional EHR features. Page numbers are not available in our copy. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [11] Wang, 2022, Artificial intelligence models for predicting cardiovascular diseases in people with type 2 diabetes: a systematic review, Intell Based Med source

      The Abstract (p. 1) reports that AI models outperformed conventional models in some scenarios and gives AUROC values for derivation cohorts. Table 2 (p. 5) lists one study with thousands of hospital diagnoses, procedures and prescriptions. The review does not establish that improvements are especially associated with high-dimensional or longitudinal EHR features, or report improved reclassification. (LLM, conf=0.87)

      Experimental — accuracy not yet measured
    • cites [12] Sang, 2024, Prediction model for cardiovascular disease in patients with diabetes using machine learning derived and validated in two independent Korean cohorts, Sci Rep source
    • cites [13] Nicolucci, 2022, Prediction of complications of type 2 Diabetes: a machine learning approach, Diabetes Res Clin Pract. source
  • Premisecitation

    Oxidative stress and lipid peroxidation are biologically linked to endothelial dysfunction and vascular injury in diabetes.

    “Among candidate pathways, oxidative stress and lipid peroxidation deserve attention because they are biologically linked to endothelial dysfunction and vascular injury in diabetes; however, this does not yet justify routine inclusion of such biomarkers in all machine learning models (18–21).”
    Quote in the paperNot a substantial claim3 of 4 sources checked
    • cites [18] Ito, 2019, Measurement and clinical significance of lipid peroxidation as a biomarker of oxidative stress: oxidative stress in diabetes, atherosclerosis, and chronic inflammation, Antioxidants source
    • cites [19] Shabalala, 2022, Detrimental effects of lipid peroxidation in type 2 diabetes: exploring the neutralizing influence of antioxidants, Antioxidants source
    • cites [20] Sugur, 2024, Serum lipid peroxidation potential as a biomarker for risk-stratification of coronary artery disease in patients with type 2 diabetes mellitus, Diabetes Metab Syndr Clin Res Rev. source
    • cites [21] Tiwari, 2013, Markers of oxidative stress during diabetes mellitus, J Biomark. source
  • Interpretationcitation

    Inadequate representation of Asian and other non-Western groups directly compromises clinical applicability in high-burden regions.

    “A large proportion of studies rely on populations from Europe and North America, with inadequate representation of Asian and other non-Western groups, directly compromising clinical applicability in high-burden regions (6, 7).”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [6] Kee O, 2023, Cardiovascular complications in a diabetes prediction model using machine learning: a systematic review, Cardiovasc Diabetol source

      The Results and Discussion (page not available) say that most included studies used European/Caucasian populations, only two studied Chinese populations, and none studied Malay or Indian populations. The Discussion calls for prediction-model studies in Malaysia and other Asian countries, but does not establish that underrepresentation directly compromises clinical applicability there; its risk-of-bias assessment reports low applicability concern for the models relative to the review question. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [7] Cai, 2024, Artificial intelligence in the risk prediction models of cardiovascular disease and development of an independent validation screening tool: a systematic review, BMC Med. source

      The Discussion’s Principal findings say most studies were conducted in Europe and North America, with few in developing countries in Asia and South America and none in Africa. They also note the high CVD burden in low- and middle-income countries and recommend locally tailored models because ethnicity can affect prediction. The paper does not establish that this imbalance directly compromises clinical applicability in those regions; that causal wording is stronger than its conclusion. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
  • May combine two claims. 12
  • Classified as probably not a substantial claim. 7

  • May combine two claims.

    Premisecitation

    Established cardiovascular risk models can remain useful in Asian populations only after appropriate local recalibration, and calibration performance may still differ across models and subgroups.

    “Recent evidence from two population-based multi-ethnic cohorts in Singapore indicates that established cardiovascular risk models can remain useful in Asian populations only after appropriate local recalibration, and calibration performance may still differ across models and subgroups (4).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [4] Lim, 2026, Cardiovascular disease risk prediction in multi-ethnic Asian populations: evidence from two population-based cohorts in Singapore, Lancet Reg Health West Pac. source

  • May combine two claims.

    Premisecitation

    The transportability of traditional risk assessment tools across Asian populations is imperfect, and local recalibration is often necessary to maintain acceptable calibration and clinical usefulness.

    “Although these tools are clinically familiar and easy to apply, their transportability across Asian populations is imperfect, and recent evidence suggests that local recalibration is often necessary to maintain acceptable calibration and clinical usefulness (4).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [4] Lim, 2026, Cardiovascular disease risk prediction in multi-ethnic Asian populations: evidence from two population-based cohorts in Singapore, Lancet Reg Health West Pac. source

  • May combine two claims.

    Interpretationreasoning

    Assuming homogeneity across Asian populations or generalizing models developed from a single subgroup to the broader Asian population risks introducing systematic predictive bias and undermining equitable clinical utility.

    “Assuming homogeneity across Asian populations or generalizing models developed from a single subgroup to the broader Asian population risks introducing systematic predictive bias and undermining equitable clinical utility.”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [16] Rountree, 2025, Reporting of fairness metrics in clinical risk prediction models used for precision health: scoping review, Online J Public Health Inform source

      The Discussion (page not provided) says model-development data were often dominated by one racial or ethnic group, leaving insufficient data to assess fairness separately across groups. The Introduction also warns that bias can affect prediction accuracy and equity. The paper does not specifically address differences among Asian populations or generalizing a model from one Asian subgroup to all Asian populations. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
    • cites [17] Mihan, 2024, Artificial intelligence bias in the prediction and detection of cardiovascular disease, NPJ Cardiovasc Health. source

      The “Sources of AI bias” section (page unavailable) says that a training dataset that does not represent its target population can introduce bias, and that risk models may perform unequally across subpopulations. This supports the general warning about applying a model beyond the group in which it was developed, but the paper does not address homogeneity among Asian populations or generalization from one Asian subgroup to others. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Interpretationreasoning

    Treating 'Asian populations' as a homogeneous entity-or extrapolating models developed in one Asian subgroup to all Asian individuals-can introduce systematic prediction bias.

    “Treating “Asian populations” as a homogeneous entity—or extrapolating models developed in one Asian subgroup to all Asian individuals—can introduce systematic prediction bias.”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [16] Rountree, 2025, Reporting of fairness metrics in clinical risk prediction models used for precision health: scoping review, Online J Public Health Inform source

      The Discussion (page unavailable) says that aggregated model results can mask inequalities and that limited data make it difficult to assess models separately across racial or ethnic groups. This supports the general concern about grouping populations, but the paper does not address treating Asian populations as homogeneous or extrapolating a model from one Asian subgroup to all Asian individuals. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
    • cites [17] Mihan, 2024, Artificial intelligence bias in the prediction and detection of cardiovascular disease, NPJ Cardiovasc Health. source

      The “Sources of AI bias” section (page unavailable) says models can perform unequally across subpopulations when training data do not represent the target population, supporting the general risk of prediction bias from extrapolation. It does not discuss Asian populations, differences among Asian subgroups, or models developed in one Asian subgroup and applied to others. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    The present evidence base for lipid peroxidation markers remains heterogeneous in marker selection, assay platform, and standardization, and direct comparative evidence against standard clinic-based models is still limited.

    “This supports the idea that lipid peroxidation markers may serve as mechanistically plausible candidate predictors; however, the present evidence base remains heterogeneous in marker selection, assay platform, and standardization, and direct comparative evidence against standard clinic-based models is still limited (20–22).”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites [20] Sugur, 2024, Serum lipid peroxidation potential as a biomarker for risk-stratification of coronary artery disease in patients with type 2 diabetes mellitus, Diabetes Metab Syndr Clin Res Rev. source
    • cites [21] Tiwari, 2013, Markers of oxidative stress during diabetes mellitus, J Biomark. source

      Section 4.1 (page unavailable) identifies several lipid-peroxidation products and discusses MDA and TBARS, but does not assess heterogeneity in marker selection, assay platforms, or standardization. Searches for those methodological issues and for comparisons with standard clinic-based prediction models found no passage addressing the claimed evidence gaps. (LLM, conf=0.72)

      Experimental — accuracy not yet measured
    • cites [22] Nuñez-Selles, 2025, Linking oxidative stress biomarkers to disease progression and antioxidant therapy in hypertension and diabetes mellitus, Front Mol Biosci source

      The Introduction (page unavailable) says biomarker variability and a lack of standardized assays limit clinical adoption. Section 3 (page unavailable) identifies different lipid-peroxidation markers and assay platforms, and Section 7 notes biases from independently applied methods. The paper does not establish that direct comparisons with standard clinic-based prediction models are limited. (LLM, conf=0.85)

      Experimental — accuracy not yet measured

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

    Interpretationcitation

    Integrating mechanistic biomarkers such as those related to lipid peroxidation holds potential to improve biological relevance and possibly predictive performance.

    “Moreover, integrating mechanistic biomarkers—such as those related to lipid peroxidation—holds potential to improve biological relevance and possibly predictive performance (18–22).”
    Quote in the paperNot a substantial claim4 of 5 sources checked
    • cites [18] Ito, 2019, Measurement and clinical significance of lipid peroxidation as a biomarker of oxidative stress: oxidative stress in diabetes, atherosclerosis, and chronic inflammation, Antioxidants source

      The Introduction describes lipid-peroxidation measures as biologically grounded indicators of oxidative stress, and the Conclusions call lipid peroxidation a useful marker of disease risk and progression. The paper does not establish that integrating these biomarkers with other measures improves predictive performance; that part remains unsupported. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [19] Shabalala, 2022, Detrimental effects of lipid peroxidation in type 2 diabetes: exploring the neutralizing influence of antioxidants, Antioxidants source

      Section 3 identifies lipid-peroxidation products as biomarkers, and Section 6 says measuring them in combination helps detect and confirm the severity of oxidative stress. This supports their biological relevance, but the paper does not address whether integrating them improves predictive performance. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
    • cites [20] Sugur, 2024, Serum lipid peroxidation potential as a biomarker for risk-stratification of coronary artery disease in patients with type 2 diabetes mellitus, Diabetes Metab Syndr Clin Res Rev. source
    • cites [21] Tiwari, 2013, Markers of oxidative stress during diabetes mellitus, J Biomark. source

      Section 4.1, “Lipid Peroxidation” (page unavailable), identifies malondialdehyde as a biomarker of lipid damage and oxidative stress, supporting the biological relevance of the example. The paper does not assess integrating such biomarkers or suggest that doing so could improve predictive performance. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [22] Nuñez-Selles, 2025, Linking oxidative stress biomarkers to disease progression and antioxidant therapy in hypertension and diabetes mellitus, Front Mol Biosci source

  • May combine two claims.

    Premisecitation

    Oxidative stress plays a central role in the shared pathophysiology of diabetes and cardiovascular disease, with lipid peroxidation representing one of its major downstream consequences.

    “Oxidative stress plays a central role in the shared pathophysiology of diabetes and cardiovascular disease, with lipid peroxidation representing one of its major downstream consequences.”
    Quote in the paperNot a substantial claim4 of 5 sources checked
    • cites [18] Ito, 2019, Measurement and clinical significance of lipid peroxidation as a biomarker of oxidative stress: oxidative stress in diabetes, atherosclerosis, and chronic inflammation, Antioxidants source
    • cites [19] Shabalala, 2022, Detrimental effects of lipid peroxidation in type 2 diabetes: exploring the neutralizing influence of antioxidants, Antioxidants source
    • cites [20] Sugur, 2024, Serum lipid peroxidation potential as a biomarker for risk-stratification of coronary artery disease in patients with type 2 diabetes mellitus, Diabetes Metab Syndr Clin Res Rev. source
    • cites [21] Tiwari, 2013, Markers of oxidative stress during diabetes mellitus, J Biomark. source
    • cites [22] Nuñez-Selles, 2025, Linking oxidative stress biomarkers to disease progression and antioxidant therapy in hypertension and diabetes mellitus, Front Mol Biosci source

  • Classified as probably not a substantial claim.

    Premisecitation

    Bias can arise throughout problem formulation, dataset construction, model development, validation, and implementation in cardiovascular AI.

    “This is especially important because a recent scoping review found that fairness metrics are rarely reported in clinical risk prediction studies, while a recent cardiovascular AI perspective emphasized that bias can arise throughout problem formulation, dataset construction, model development, validation, and implementation (16, 17).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [17] Mihan, 2024, Artificial intelligence bias in the prediction and detection of cardiovascular disease, NPJ Cardiovasc Health. source

  • May combine two claims.

    Premisecitation

    Recent studies from Malaysia and China suggest that regionally developed machine learning models can improve performance in local cohorts, but local validation, calibration assessment, and broader external testing remain essential before wider implementation.

    “At the same time, recent studies from Malaysia and China suggest that regionally developed machine learning models can improve performance in local cohorts, but these same studies also show why local validation, calibration assessment, and broader external testing remain essential before wider implementation (14, 15).”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [14] Kasim, 2025, Enhanced cardiovascular risk prediction in the Western Pacific: a machine learning approach tailored to the Malaysian population, PLoS ONE source

      Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [15] Huang, 2025, Predicting cardiovascular outcomes in Chinese patients with type 2 diabetes by combining risk factor trajectories and machine learning algorithm: a cohort study, Cardiovasc Diabetol. source

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

    Premisecitation

    The association between lipid peroxidation and diabetic cardiovascular complications has been extensively discussed in the literature.

    “The association between lipid peroxidation and diabetic cardiovascular complications has been extensively discussed in the literature (18, 19, 21, 22).”
    Quote in the paperNot a substantial claim4 of 4 sources checked
    • cites [18] Ito, 2019, Measurement and clinical significance of lipid peroxidation as a biomarker of oxidative stress: oxidative stress in diabetes, atherosclerosis, and chronic inflammation, Antioxidants source
    • cites [19] Shabalala, 2022, Detrimental effects of lipid peroxidation in type 2 diabetes: exploring the neutralizing influence of antioxidants, Antioxidants source
    • cites [21] Tiwari, 2013, Markers of oxidative stress during diabetes mellitus, J Biomark. source
    • cites [22] Nuñez-Selles, 2025, Linking oxidative stress biomarkers to disease progression and antioxidant therapy in hypertension and diabetes mellitus, Front Mol Biosci source

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

    Interpretationcitation

    Adherence to TRIPOD+AI and PROBAST+AI is necessary for evaluating whether apparently strong models are actually fit for clinical use.

    “At the same time, adherence to TRIPOD+AI and PROBAST+AI should be treated as a methodological expectation, not an optional add-on, because transparency and structured bias appraisal are necessary for evaluating whether apparently strong models are actually fit for clinical use (9, 10).”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [9] Collins G, 2024, TRIPOD+ AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods, BMJ source

      The Discussion recommends making TRIPOD+AI adherence an expectation, and the paper says transparent reporting is essential to judging whether models suit a clinical setting. The TRIPOD+AI statement refers readers to PROBAST and the forthcoming PROBAST+AI for quality and risk-of-bias assessment, but does not say that adherence to PROBAST+AI is itself a necessary expectation. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [10] Moons, 2025, PROBAST+ AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods, BMJ source

  • May combine two claims.

    Premisecitation

    Current models are predominantly developed using populations from Europe and North America, resulting in a critical lack of representativeness for Asian populations.

    “Furthermore, current models are predominantly developed using populations from Europe and North America, resulting in a critical lack of representativeness for Asian populations, where the burden of cardiovascular disease is particularly heavy.”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [6] Kee O, 2023, Cardiovascular complications in a diabetes prediction model using machine learning: a systematic review, Cardiovasc Diabetol source
    • cites [7] Cai, 2024, Artificial intelligence in the risk prediction models of cardiovascular disease and development of an independent validation screening tool: a systematic review, BMC Med. source

      The Discussion (Principal findings; page unavailable) says most studies were conducted in Europe and North America, with few in developing countries in Asia, and calls for locally tailored models because ethnic differences can affect prediction. It describes a growing CVD burden across low- and middle-income countries, but does not establish that the burden is particularly heavy in Asia specifically. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation

    Clinical prediction studies using machine learning should report not only discrimination, but also calibration, intended use, validation strategy, and risk-of-bias considerations in a structured way.

    “Recent guidance now makes clear that clinical prediction studies using machine learning should report not only discrimination, but also calibration, intended use, validation strategy, and risk-of-bias considerations in a structured way (9, 10).”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [9] Collins G, 2024, TRIPOD+ AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods, BMJ source
    • cites [10] Moons, 2025, PROBAST+ AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods, BMJ source

      The Model evaluation section identifies discrimination, calibration, clinical usefulness, and internal and external validation as aspects of performance evaluation. The PROBAST+AI and Steps for using PROBAST+AI sections describe a structured assessment of quality, risk of bias, and applicability, including intended use. The paper supports considering these issues in machine-learning prediction studies, but presents an appraisal tool rather than a requirement that studies report each item in a structured way. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Machine learning can autonomously discern non-linear risk patterns from high-dimensional data, iteratively train and validate across algorithms, and identify optimal performers through comparative evaluation.

    “Unlike traditional regression models reliant on fixed assumptions and linear frameworks, machine learning can autonomously discern non-linear risk patterns from high-dimensional data, iteratively train and validate across algorithms, and identify optimal performers through comparative evaluation.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [11] Wang, 2022, Artificial intelligence models for predicting cardiovascular diseases in people with type 2 diabetes: a systematic review, Intell Based Med source

      The Model development summary (p. 2) describes data-driven machine-learning models and hyperparameter tuning, while the Model performance summary (p. 3) compares algorithms using AUROC and identifies the best-performing models. The paper does not establish that these models autonomously discerned non-linear patterns from high-dimensional data or that they were iteratively trained and validated across algorithms; it reports that only three studies described data splits and tuning, and only one conducted external validation. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    The core strengths of machine learning in cardiovascular risk prediction lie in its capacity to handle complex, high-dimensional data and to adapt effectively to class imbalance, enabling the identification of non-linear risk patterns that traditional models often fail to capture.

    “The core strengths of machine learning in cardiovascular risk prediction lie in its capacity to handle complex, high-dimensional data and to adapt effectively to class imbalance, enabling the identification of non-linear risk patterns that traditional models often fail to capture (11).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [11] Wang, 2022, Artificial intelligence models for predicting cardiovascular diseases in people with type 2 diabetes: a systematic review, Intell Based Med source

      The Model development summary (p. 2) reports that three reviewed studies addressed class imbalance using class weights, sub-sampling or ensemble modelling. The Model performance summary (p. 3) reports that machine-learning models outperformed regression models for several outcomes, though the differences were small. The paper does not establish that handling high-dimensional data or detecting non-linear patterns explains that performance, and it does not show effective adaptation to imbalance across all the models. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

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

    Interpretationreasoning

    By incorporating emerging mechanistic biomarkers and addressing population-specific differences, machine learning-driven tools can deliver more personalized and equitable risk assessment.

    “By incorporating emerging mechanistic biomarkers and addressing population-specific differences, machine learning-driven tools can deliver more personalized and equitable risk assessment, thereby advancing the realization of precision medicine.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The predominance of models developed from European and North American populations results in inadequate representativeness for other groups, potentially exacerbating healthcare disparities.

    “The predominance of models developed from European and North American populations results in inadequate representativeness for other groups, potentially exacerbating healthcare disparities.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Machine learning models often exhibit excellent performance in controlled or internally validated settings, but data bias and limited external generalizability restrict their widespread adoption, particularly in high-risk regions such as Asia.

    “While these models often exhibit excellent performance in controlled or internally validated settings, issues of data bias and limited external generalizability restrict their widespread adoption, particularly in high-risk regions such as Asia.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    The field is undergoing a pivotal transition from an exclusive focus on algorithmic performance to ensuring clinical equity and fairness.

    “This article argues that the field is undergoing a pivotal transition—from an exclusive focus on algorithmic performance to ensuring clinical equity and fairness.”
    Quote in the paperNot a substantial claim
19 of 19

  • The paper relies on this without support.

    The addition of laboratory assays for lipid peroxidation biomarkers may increase the financial and logistical burden on healthcare systems in low- and middle-income countries, potentially limiting implementation.

    “Second, the addition of laboratory assays may increase the financial and logistical burden on healthcare systems in low- and middle-income countries, potentially limiting implementation.”

  • The paper relies on this without support.

    Measurement of lipid peroxidation biomarkers requires standardization to ensure comparability across studies.

    “First, measurement of these biomarkers requires standardization to ensure comparability across studies.”
2 of 2

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.

  • IDF Atlas edition and year conflict with cited reference

    Minor

    The text attributes the 2024 diabetes estimate to the 11th edition of the IDF Diabetes Atlas, but reference 1 identifies the cited Atlas as the 7th edition, published in 2015.

1 of 1

Cites

Flagged first, then by citations

Research works printed but not linked: 2

  1. Atlas, 2015, IDF Diabetes Atlas., 3
  2. 2021, Cardiovascular Diseases (CVDs)

Authors
Li, H., Xu, Z., Cen, Y., & Liu, X.
Affiliations
Guangxi University of Chinese Medicine (CN)
Corresponding author
Liu, Xin
ORCID
Liu, Xin
Venue
Frontiers in Medicine
Publisher
Frontiers Media SA
Published
2026-06-02
Volume
13
Pages
1831220-1831220
ISSN
2296-858X (ISSN-L)
DOI
10.3389/fmed.2026.1831220
PMC
PMC13268988
OpenAlex
W7163219793
Semantic Scholar
b3f3d07061aa926626f64e44a6489e50ed937201
Cited by
0
References
22
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
  • Frontiers in Medicine (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • PubMed Central (submittedVersion, cc-by)
Topics
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
  • Health Sciences › Medicine › Endocrinology, Diabetes and Metabolism › Diabetes, Cardiovascular Risks, and Lipoproteins
Keywords
cardiovascular risk prediction, type 2 diabetes, machine learning, algorithmic fairness, external validation, model calibration, Asian populations, biomarkers, reporting guidelines, clinical translation

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 2026-10-13.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-13.

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