CoChat
Sign Up
Cited by 5Artificial Intelligence in HealthcareDiabetes, Cardiovascular Risks, and LipoproteinsBody Composition Measurement Techniques

Machine learning with the body roundness index and associated indicators: a new approach to predicting metabolic syndrome

Open access
He, Yaxuan, Chen, Zekai, Tang, Zhaohui, Qin, Yuexiang, Wang, Fang·2025·BMC Public Health
DOI 10.1186/s12889-025-23858-yPMC PMC12315366
Open this paper in CoChat — free accountOpen PDF

Abstract

BACKGROUND: Metabolic syndrome (MetS) is strongly associated with increased cardiovascular morbidity and mortality. Traditional invasive diagnostic methods are costly, inconvenient, and unsuitable for large-scale screening. Developing a non-invasive, accurate prediction model is clinically significant for early MetS detection and prevention. OBJECTIVE: To develop a non-invasive prediction model for MetS by integrating body roundness index (BRI), gender, age, height, and waist circumference with machine learning algorithms, and to validate its generalizability across different ethnic groups. METHODS: We trained and validated machine learning models using a retrospective health examination dataset from Central South University (D1, n = 268,942) and externally validated them on a European cohort (D2, n = 60,799). Five non-invasive features-BRI, waist circumference, height, age, and gender-were used as predictors. MetS was diagnosed based on the IDF criteria. Ten machine learning algorithms were evaluated using 10-fold cross-validation. Model performance was assessed via accuracy, specificity, sensitivity, F1-score, precision, ROC AUC and Brier score. Additional analyses included threshold sensitivity testing (0.1-0.9) and ablation of central adiposity features to evaluate robustness and generalizability. RESULTS: In the D1, the prevalence of MetS was 18.26%. In the D2, the prevalence was 9.07%. Among all predictors, BRI demonstrated the strongest correlation with MetS (r = 0.585 in D1; r = 0.426 in D2; both P < 0.01). Compared with a rule-based baseline (F1 score = 0.50), all machine learning models achieved superior performance. At the default 0.5 threshold, XGBoost attained an AUC of 0.94, sensitivity of 0.96, and F1 score of 0.53 in D2. Further threshold adjustment improved XGBoost's precision from 0.36 to 0.43, while maintaining high sensitivity (0.87). Model performance declined markedly when central adiposity features (BRI and WC) were removed, highlighting the critical predictive value of BRI. These findings support the utility of BRI-integrated ML approaches for scalable, non-invasive MetS screening. CONCLUSIONS: The combination of BRI and machine learning provides a non-invasive and effective method for predicting MetS and offers a promising strategy for its early prevention. Machine learning models demonstrate significant advantages over baseline rule-based models in terms of sensitivity and predictive performance.

AI snapshot

Written from the paper’s full text.

TL;DR

This study developed and externally validated machine-learning models using non-invasive anthropometric and demographic features—especially the body roundness index (BRI)—to predict metabolic syndrome (MetS). BRI showed the strongest correlation with MetS, and ML models outperformed a rule-based IDF baseline, though external validation in a European cohort revealed reduced precision and population-transfer challenges.

Key findings

  • MetS prevalence was 18.26% in the Chinese health-examination cohort (D1, n = 268,942) and 9.07% in the European external cohort (D2, n = 60,799). BRI had the strongest correlation with MetS in both datasets (D1: r = 0.585, P < 0.01; D2: r = 0.426, P < 0.01), and in D1 logistic regression BRI had a higher odds ratio than BMI (BRI OR 9.414, 95% CI 9.229–9.603; BMI OR 6.268, 95% CI 6.268–6.483).
  • ML models outperformed the rule-based baseline (baseline F1 = 0.50). In D1 using five non-invasive features, XGBoost achieved sensitivity 1.00, F1 0.75, and ROC AUC 0.95; SVM had sensitivity 0.78, specificity 0.89, and Brier score 0.15.
  • In external D2 validation at the 0.5 threshold, XGBoost achieved AUC 0.94, sensitivity 0.96, F1 0.53, and precision 0.36. Raising the threshold to 0.8 improved precision to 0.43 while maintaining sensitivity 0.873.
  • Ablation of central adiposity features (BRI and WC) markedly reduced performance: LightGBM and GBDT had sensitivity 0.00 despite specificity 1.00, while XGBoost retained sensitivity 0.76 but with precision 0.28 and accuracy 0.60. In false-positive/true-negative analysis, the D2 external set contained 45,990 TN and 9,294 FP; FP individuals had higher WC, SBP, DBP, TG, FBG, BRI, and BMI, with no significant HDL-C difference.

Methods

  • Retrospective health-examination dataset from Central South University (D1, n = 268,942; 2017–2022) for training/validation, and an external European working-population cohort from the Balearic Islands, Spain (D2, n = 60,799; 2012–2016) for external validation. MetS was diagnosed using IDF criteria.
  • Five non-invasive predictors were used: BRI, waist circumference, height, age, and gender. Ten ML algorithms were evaluated (LR, DT, RF, XGBoost, SVM, GBDT, LightGBM, KNN, NB, ANN) with 10-fold cross-validation, grid-search hyperparameter tuning, StandardScaler standardization, and class-imbalance handling via scale_pos_weight.
  • Performance was assessed by accuracy, specificity, sensitivity, F1-score, precision, ROC AUC, and Brier score at a 0.5 threshold, with additional threshold sensitivity testing from 0.1 to 0.9 and an ablation analysis removing central adiposity features.

Limitations

  • The paper states: algorithmic complexity and limited interpretability may impede clinical integration; external validation used a demographically distinct European cohort and may not capture global heterogeneity; feature selection omitted behavioral, biochemical, and genetic factors; cross-sectional data prevented assessment of temporal changes; retrospective design precluded causal inference; lack of shared code/datasets in prior ML studies precluded direct benchmarking; decision curve analysis (DCA) was not included.
  • Evident from the text: internal D1 performance was near-perfect for several models (e.g., RF, XGBoost, LightGBM with accuracy 1.00, specificity 1.00, F1 ≥ 0.98 in Table 4), while external D2 performance was substantially lower, suggesting possible overfitting or dataset-specific effects. The paper also notes that D2’s lower MetS prevalence likely contributed to increased false positives and reduced predictive balance under the default threshold.

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

Integrity signals

Registry

Crossref · 10.1186/s12889-025-23858-y

Cites

51 of 51 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

40 of 51 cited sources checked · 17 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Machine learning has demonstrated effectiveness in the early detection and cost-effective screening of metabolic syndrome.

    “Notably, ML has demonstrated its effectiveness in the early detection and cost-effective screening of MetS [11].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [11] Xu W, Zhang Z, Hu K, Fang P, Li R, Kong D, Xuan M, Yue Y, She D, Xue Y. Identifying metabolic syndrome easily and cost effectively using non-invasive methods with machine learning models. Diabetes Metabolic Syndrome Obes. 2023;16:2141–51. source

      The Discussion reports strong detection performance for an ML model and describes its use for early screening. The Abstract Conclusion calls it a method for early, low-cost identification, but says it has the *potential* to be a cost-effective screening tool; it does not establish cost-effective screening as a demonstrated outcome. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
  • Interpretationown data

    The discrepancy in external validation underscores the potential influence of region- and population-specific factors on model transferability.

    “This discrepancy underscores the potential influence of region- and population-specific factors on model transferability, such as differences in dietary patterns, genetic susceptibility, and variations in metabolic phenotypes [38–41].”
    Quote in the paperSubstantial claim2 of 4 sources checked
    • cites [38] Han Y, Lv J, Yu C, Guo Y, Bian Z, Hu Y, Yang L, Chen Y, Du H, Zhao F, et al. Development and external validation of a breast cancer absolute risk prediction model in Chinese population. Breast Cancer Res. 2021;23(1): 62. source
    • cites [39] Asato CBH, Nelson-Hurwitz DC, Lee T, Grandinetti A. Comparative analysis of metabolic syndrome diagnostic criteria and its effects on prevalence in a multiethnic population. Metab Syndr Relat Disord. 2021;19(6):347–51. source

      The Discussion (page unavailable) says that differences between the China and Kohala findings may reflect the populations studied, and the Results report that diagnostic criteria performed differently by ethnicity. This supports the general relevance of population-specific factors, but the paper does not assess external validation or predictive-model transferability, or establish dietary patterns or genetic susceptibility as explanations. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [40] Titmuss A, Davis EA, Brown A, Maple-Brown LJ. Emerging diabetes and metabolic conditions among Aboriginal and Torres Strait Islander young people. Med J Aust. 2019;210(3):111-e113111. source
    • cites [41] Unwin N, Bhopal R, Hayes L, White M, Patel S, Ragoobirsingh D, Alberti G. A comparison of the new International Diabetes Federation definition of metabolic syndrome to WHO and NCEP definitions in Chinese, European and South Asian origin adults. Ethn Dis. 2007;17(3):522–8.
  • Study design and conductcitation

    Participants were classified as having metabolic syndrome or not based on the IDF criteria.

    “They were classified as having MetS or not based on the IDF criteria [18] (Fig. 2), with 49,107 identified as having MetS and 219,835 without.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [18] Grundy SM, Cleeman JI, Daniels SR, Donato KA, Eckel RH, Franklin BA, Gordon DJ, Krauss RM, Savage PJ, Smith SC Jr. Diagnosis and management of the metabolic syndrome: an American heart association/national heart, lung, and blood Institute scientific statement. Circulation. 2005;112(17):2735–52. source
  • Interpretationcitation

    Given existing evidence that even non-MetS obese individuals face elevated cardiovascular risk, these false positives may represent a clinically meaningful subgroup who could benefit from earlier lifestyle or medical interventions.

    “Given existing evidence that even non-MetS obese individuals face elevated cardiovascular risk [43–45], these so-called “false positives” may represent a clinically meaningful subgroup who could benefit from earlier lifestyle or medical interventions.”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites [43] Manoharan MP, Raja R, Jamil A, Csendes D, Gutlapalli SD, Prakash K, Swarnakari KM, Bai M, Desai DM, Desai A, et al. Obesity and coronary artery disease: an updated systematic review 2022. Cureus. 2022;14(9):e29480. source
    • cites [44] Sánchez-Iñigo L, Navarro-González D, Martinez-Urbistondo D, Pastrana JC, Fernandez-Montero A, Martinez JA. Repercussions of absolute and time-rated BMI yo-yo fluctuations on cardiovascular stress-related morbidities within the vascular-metabolic CUN cohort. Front Endocrinol (Lausanne). 2022;13: 1087554. source

      The Introduction (page unavailable) calls obesity a well-established cardiovascular risk factor. Neither that passage nor the Discussion passages distinguish obese people without metabolic syndrome, so the source does not establish the claim for the non-MetS subgroup. The suggestion about intervening for “false positives” is the citing authors’ own inference, not the point attributed to this source. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [45] Xu R, Gao X, Wan Y, Fan Z. Association of metabolically healthy overweight phenotype with abnormalities of glucose levels and blood pressure among Chinese adults. JAMA Netw Open. 2019;2(10):e1914025. source
  • Interpretationown data

    Central obesity measures-particularly BRI-are indispensable for accurate risk classification.

    “illustrating the indispensable value of central obesity measures—particularly BRI—in accurate risk classification [35–37].”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites [35] Fahed G, Aoun L, Bou Zerdan M, Allam S, Bou Zerdan M, Bouferraa Y, Assi HI. Metabolic syndrome: updates on pathophysiology and management in 2021. Int J Mol Sci. 2022. https://doi.org/10.3390/ijms23020786. source

      The Introduction (§1.1, page unavailable) identifies central obesity as a metabolic-syndrome risk factor and says waist circumference and waist-to-height ratio help assess risk. It does not identify BRI as particularly valuable, and it cautions that waist circumference alone is inconclusive, so the claim of indispensable value for accurate classification overstates this source. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites [36] Neeland IJ, Lim S, Tchernof A, Gastaldelli A, Rangaswami J, Ndumele CE, Powell-Wiley TM, Després JP. Metabolic syndrome. Nat Rev Dis Primers. 2024;10(1):77. source
    • cites [37] He M, Wang J, Liang Q, Li M, Guo H, Wang Y, Deji C, Sui J, Wang YW, Liu Y, et al. Time-restricted eating with or without low-carbohydrate diet reduces visceral fat and improves metabolic syndrome: A randomized trial. Cell Rep Med. 2022;3(10):100777. source

      The Results section (page unavailable) calls abdominal fat a pivotal metabolic risk factor and says waist-to-hip ratio is more closely correlated with metabolic syndrome than BMI. That supports the relevance of central-obesity measures, but the paper does not establish that they are indispensable for accurate risk classification or identify BRI as particularly valuable. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
  • Classified as probably not a substantial claim. 29
  • May combine two claims. 42
  • The quote only partly states this claim. 1

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

    Premisecitation

    The presence of metabolic syndrome increases the risk of developing cardiovascular diseases twofold.

    “The presence of MetS increases the risk of developing cardiovascular diseases (CVDs) twofold, and the risk of all-cause mortality by a factor of 1.5 [2, 3].”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [2] Gallardo-Alfaro L, Bibiloni MDM, Mascaró CM, Montemayor S, Ruiz-Canela M, Salas-Salvadó J, Corella D, Fitó M, Romaguera D, Vioque J. Leisure-time physical activity, sedentary behaviour and diet quality are associated with metabolic syndrome severity: the PREDIMED-plus study. Nutrients. 2020;12(4): 1013. source

      The Introduction reports a twofold higher risk of coronary heart disease and cerebrovascular disease in people with metabolic syndrome. Those are cardiovascular diseases, but the passage does not establish a twofold risk for cardiovascular diseases as a whole. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [3] Wilkinson MJ, Manoogian EN, Zadourian A, Lo H, Fakhouri S, Shoghi A, Wang X, Fleischer JG, Navlakha S, Panda S. Ten-hour time-restricted eating reduces weight, blood pressure, and atherogenic lipids in patients with metabolic syndrome. Cell Metabol. 2020;31(1):92–104. e105. source

  • Classified as probably not a substantial claim.

    Premisecitation

    Waist circumference indicates central obesity, a key part of metabolic syndrome.

    “Gender affects hormonal and metabolic functions, age is related to metabolic changes, height is associated with body proportions and health, and waist circumference (WC) indicates central obesity, a key part of MetS [12, 13].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [12] Rahman QAA, Razzak MA, Wahab MA. Metabolic syndrome: An emerging threat. J Armed Forces Med Coll Bangladesh. 2018;14(2):203–8. source
    • cites [13] Bagheri P, Khalili D, Seif M, Rezaianzadeh A. Dynamic behavior of metabolic syndrome progression: a comprehensive systematic review on recent discoveries. BMC Endocr Disord. 2021;21:1–14. source

      The Background identifies central obesity as a component of metabolic syndrome. Table 6 in the Results lists waist circumference as a predictor of MetS development, but the passages do not explicitly say that waist circumference indicates central obesity. (LLM, conf=0.78)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Machine learning algorithms analyze extensive clinical data and identify underlying patterns and trends, offering robust support for disease prediction and risk assessment.

    “ML algorithms, which analyze extensive clinical data and identify underlying patterns and trends, offer robust support for disease prediction and risk assessment [8, 9].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [8] Noorbakhsh-Sabet N, Zand R, Zhang Y, Abedi V. Artificial intelligence transforms the future of health care. Am J Med. 2019;132(7):795–801. source
    • cites [9] Handelman GS, Kok HK, Chandra RV, Razavi AH, Lee MJ, Asadi H. eD octor: machine learning and the future of medicine. J Intern Med. 2018;284(6):603–19. source

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

    Premisecitation

    Height is associated with body proportions and health.

    “Gender affects hormonal and metabolic functions, age is related to metabolic changes, height is associated with body proportions and health, and waist circumference (WC) indicates central obesity, a key part of MetS [12, 13].”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [12] Rahman QAA, Razzak MA, Wahab MA. Metabolic syndrome: An emerging threat. J Armed Forces Med Coll Bangladesh. 2018;14(2):203–8. source

      The Introduction (p. 1) discusses body mass index and waist-to-hip ratio, and Systemic Effects (p. 3) discusses fat distribution around the waist. Neither addresses an association between height, body proportions, and health. Searches for height, stature, and related body-proportion terms found no passage supporting that claim. (LLM, conf=0.73)

      Experimental — accuracy not yet measured
    • cites [13] Bagheri P, Khalili D, Seif M, Rezaianzadeh A. Dynamic behavior of metabolic syndrome progression: a comprehensive systematic review on recent discoveries. BMC Endocr Disord. 2021;21:1–14. source

      The Results and Discussion passages address age, sex, obesity, and waist circumference, but do not discuss height’s association with body proportions or health. Searches for “height body proportions health” and “height stature body proportions” found no passage addressing that claim. (LLM, conf=0.74)

      Experimental — accuracy not yet measured

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

    Premisecitation

    Gender affects hormonal and metabolic functions.

    “Gender affects hormonal and metabolic functions, age is related to metabolic changes, height is associated with body proportions and health, and waist circumference (WC) indicates central obesity, a key part of MetS [12, 13].”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [12] Rahman QAA, Razzak MA, Wahab MA. Metabolic syndrome: An emerging threat. J Armed Forces Med Coll Bangladesh. 2018;14(2):203–8. source

      The Introduction (p. 1) reports differences in metabolic-syndrome prevalence by sex, while Systemic Effects (p. 3) lists reproductive conditions associated with the syndrome. Neither passage says that gender affects hormonal and metabolic functions; searches for sex-related hormonal and metabolic differences found no passage making that claim. (LLM, conf=0.72)

      Experimental — accuracy not yet measured
    • cites [13] Bagheri P, Khalili D, Seif M, Rezaianzadeh A. Dynamic behavior of metabolic syndrome progression: a comprehensive systematic review on recent discoveries. BMC Endocr Disord. 2021;21:1–14. source

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

    Study design and conductcitation

    The study adhered to the TRIPOD guidelines to ensure transparency and rigor in model development and reporting.

    “We adhered to the TRIPOD guidelines to ensure transparency and rigor in model development and reporting [27, 28].”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [27] Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350: g7594. source

      The TRIPOD components and Discussion sections identify TRIPOD as a guideline for transparent reporting of prediction-model studies. They explicitly say it does not prescribe how models should be developed or serve as a quality-assessment tool, so the source does not support the implication that adherence ensures rigor in model development. The source also cannot establish whether the citing study adhered to it. (LLM, conf=0.89)

      Experimental — accuracy not yet measured
    • cites [28] Collins GS, Dhiman P, Andaur Navarro CL, Ma J, Hooft L, Reitsma JB, Logullo P, Beam AL, Peng L, Van Calster B, et al. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence. BMJ Open. 2021;11(7):e048008. source

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

    Premisecitation

    Age is related to metabolic changes.

    “Gender affects hormonal and metabolic functions, age is related to metabolic changes, height is associated with body proportions and health, and waist circumference (WC) indicates central obesity, a key part of MetS [12, 13].”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [12] Rahman QAA, Razzak MA, Wahab MA. Metabolic syndrome: An emerging threat. J Armed Forces Med Coll Bangladesh. 2018;14(2):203–8. source

      The Diagnostic Criteria section (p. 1) lists advancing age among factors considered in assessing metabolic syndrome, supporting an association between age and metabolic health. It does not describe metabolic changes that occur with age, so the claim is broader than the passage supports. (LLM, conf=0.76)

      Experimental — accuracy not yet measured
    • cites [13] Bagheri P, Khalili D, Seif M, Rezaianzadeh A. Dynamic behavior of metabolic syndrome progression: a comprehensive systematic review on recent discoveries. BMC Endocr Disord. 2021;21:1–14. source

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

    Premisecitation

    The presence of metabolic syndrome increases the risk of all-cause mortality by a factor of 1.5.

    “The presence of MetS increases the risk of developing cardiovascular diseases (CVDs) twofold, and the risk of all-cause mortality by a factor of 1.5 [2, 3].”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites [2] Gallardo-Alfaro L, Bibiloni MDM, Mascaró CM, Montemayor S, Ruiz-Canela M, Salas-Salvadó J, Corella D, Fitó M, Romaguera D, Vioque J. Leisure-time physical activity, sedentary behaviour and diet quality are associated with metabolic syndrome severity: the PREDIMED-plus study. Nutrients. 2020;12(4): 1013. source
    • cites [3] Wilkinson MJ, Manoogian EN, Zadourian A, Lo H, Fakhouri S, Shoghi A, Wang X, Fleischer JG, Navlakha S, Panda S. Ten-hour time-restricted eating reduces weight, blood pressure, and atherogenic lipids in patients with metabolic syndrome. Cell Metabol. 2020;31(1):92–104. e105. source

      The Introduction says metabolic syndrome doubles CVD risk and that treating it matters for reducing mortality from T2DM and CVD. It does not give an all-cause mortality risk or the claimed 1.5-fold figure. Searches for all-cause mortality and a 1.5-fold mortality risk found no supporting passage in our copy. (LLM, conf=0.72)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Traditional medical diagnostics rely heavily on blood sampling, which is invasive and poses discomfort and risks to patients.

    “Traditional medical diagnostics rely heavily on blood sampling to assess metabolite levels and enzyme activities, which, while accurate, is invasive and poses discomfort and risks to patients [5, 6].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [5] Hirose H, Takayama T, Hozawa S, Hibi T, Saito I. Prediction of metabolic syndrome using artificial neural network system based on clinical data including insulin resistance index and serum adiponectin. Comput Biol Med. 2011;41(11):1051–6. source
    • cites [6] Wang H, Wang Y, Li X, Deng X, Kong Y, Wang W, Zhou Y. Machine learning of plasma metabolome identifies biomarker panels for metabolic syndrome: findings from the China suboptimal health cohort. Cardiovasc Diabetol. 2022;21(1): 288. source

      The Materials and methods section (Measurements and sample collection; page unavailable) says blood was collected by venipuncture for plasma metabolomics, supporting the blood-sampling and invasive-procedure aspects. It does not say that traditional diagnostics rely heavily on blood sampling or that sampling causes patient discomfort or risks; it also does not address enzyme-activity testing. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

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

    Premisecitation

    The risk or prevalence of metabolic syndrome increases with age.

    “Data from the National Institute of Health (NIH) indicate that the risk of MetS also increases with age [4].”
    • Also stated in: § Feature selection and extraction · plain
      “Specifically, the prevalence of MetS increases with age [19, 20]”
    Quote in the paperNot a substantial claim3 of 3 sources checked
    • cites [4] Naghipour M, Joukar F, Nikbakht H-A, Hassanipour S, Asgharnezhad M, Arab-Zozani M, Mansour-Ghanaei F. High prevalence of metabolic syndrome and its related demographic factors in North of iran: results from the Persian Guilan cohort study. Int J Endocrinol. 2021;2021(1):8862456. source

      The Results report that metabolic syndrome prevalence increased with age in the Guilan cohort, and the Discussion also describes an age-related increase in an American survey. The cited paper does not identify the National Institutes of Health (NIH) as the source of those data, so the sentence’s NIH attribution is not supported. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [19] Rigamonti AE, Cicolini S, Tamini S, Caroli D, Cella SG, Sartorio A. The age-dependent increase of metabolic syndrome requires more extensive and aggressive non-pharmacological and pharmacological interventions: a cross-sectional study in an Italian cohort of obese women. Int J Endocrinol. 2021;2021: 2021:5576286. source

      The Results (Section 3; page unavailable) report that metabolic syndrome prevalence rose from 25.5% among women aged 18–30 to 71.0% among those over 70. This supports the age-related prevalence claim in the Italian cohort of obese women, but the paper does not present the finding as NIH data. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [20] Kim S, Choi SY, Lee H, Ju Kim J, Eun Park H. Sex and age differences in the impact of metabolic syndrome and its components including A body shape index on arterial stiffness in the general population. J Atheroscler Thromb. 2022;29(12):1774–90. source

      The Results section reports that metabolic syndrome prevalence increased significantly with age in women, but found no age-related difference in men. The cited paper therefore supports the claim only for women in its study population, not as an unqualified statement; the passage also does not identify the finding as NIH data. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Numerous studies have confirmed a relationship between gender and metabolic syndrome, with varying rates of metabolic syndrome occurrence between sexes.

    “numerous studies have confirmed the relationship between gender and MetS, with varying rates of MetS occurrence between sexes [21].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [21] Xie Q, Kuang M, Lu S, Huang X, Wang C, Zhang S, Sheng G, Zou Y. Association between MetS-IR and prediabetes risk and sex differences: a cohort study based on the Chinese population. Front Endocrinol (Lausanne). 2023;14: 1175988. source

      The Results (Baseline characteristics; page unavailable) give different prediabetes incidence rates for men and women, not rates of metabolic syndrome. The Introduction describes MetS-IR as an insulin-resistance index; it does not establish the claimed sex difference in metabolic syndrome occurrence. A search for sex-specific metabolic syndrome prevalence found no supporting passage in our copy. (LLM, conf=0.78)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Compared with traditional methods, machine learning excels in handling larger datasets and delivering accurate predictive outcomes more efficiently, with reduced bias.

    “Compared with traditional methods, ML excels in handling larger datasets and delivering accurate predictive outcomes more efficiently, with reduced bias [10].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [10] Greener JG, Kandathil SM, Moffat L, Jones DT. A guide to machine learning for biologists. Nat Rev Mol Cell Biol. 2022;23(1):40–55. source

      The Introduction (page not supplied) says machine learning is useful for datasets too large or complex for human analysis and can make analysis reproducible and time-efficient. It does not establish that ML generally produces more accurate predictions or reduces bias compared with traditional methods. The Traditional machine learning section also cautions that deep learning can take more time and computation than traditional ML methods. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Many studies have shown that BRI is associated with metabolic syndrome, is an independent risk factor, and has a high AUC value in predicting metabolic syndrome, even across different ethnic groups.

    “Many studies have shown its association with MetS, being an independent risk factor with a high AUC value in predicting MetS, even across different ethnic groups [15, 16].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [15] Stefanescu A, Revilla L, Lopez T, Sanchez SE, Williams MA, Gelaye B. Using a body shape index (ABSI) and body roundness index (BRI) to predict risk of metabolic syndrome in Peruvian adults. J Int Med Res. 2020;48(1): 0300060519848854. source
    • cites [16] Tian S, Zhang X, Xu Y, Dong H. Feasibility of body roundness index for identifying a clustering of cardiometabolic abnormalities compared to BMI, waist circumference and other anthropometric indices: the China health and nutrition survey, 2008 to 2009. Medicine (Baltimore). 2016;95(34): e4642. source

      The Results (page unavailable) report an association between BRI and metabolic syndrome and evaluate its predictive AUC in a Chinese sample. The paper does not establish prediction across different ethnic groups or BRI as an independent risk factor for future metabolic syndrome; its Discussion notes that prospective studies are needed and that only a few earlier studies had examined BRI’s predictive abilities. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    The body roundness index is linked to body fat and visceral adipose tissue.

    “Additionally, the body roundness index (BRI) [14], a novel index calculated by a specific formula (1), ranges from 1 to 16 and is linked to body fat and visceral adipose tissue.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [14] Thomas DM, Bredlau C, Bosy-Westphal A, Mueller M, Shen W, Gallagher D, Maeda Y, McDougall A, Peterson CM, Ravussin E. Relationships between body roundness with body fat and visceral adipose tissue emerging from a new geometrical model. Obesity. 2013;21(11):2264–71. source

  • May combine two claims.

    Interpretationown data

    This finding aligns with existing evidence and reinforces the potential of BRI as a robust predictor for identifying individuals at high risk of MetS.

    “This finding not only aligns with existing evidence but also reinforces the potential of BRI as a robust predictor for identifying individuals at high risk of MetS [29, 30].”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [29] Rico-Martín S, Calderón-García JF, Sánchez-Rey P, Franco-Antonio C, Martínez Alvarez M, Sánchez Muñoz-Torrero JF. Effectiveness of body roundness index in predicting metabolic syndrome: a systematic review and meta-analysis. Obes Rev. 2020;21(7): e13023. source
    • cites [30] Li Z, Fan C, Huang J, Chen Z, Yu X, Qian J. Non-linear relationship between the body roundness index and metabolic syndrome: data from National Health and nutrition examination survey (NHANES) 1999–2018. Br J Nutr. 2024;131(11):1852–9. source

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

    Premisecitation

    Similar cross-population differences have been documented in previous comparative studies, such as J. Wang's analysis of Chinese and Spanish adolescents.

    “Notably, similar cross-population differences have been documented in previous comparative studies, such as J. Wang’s analysis of Chinese and Spanish adolescents [42].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [42] Wang J, Perona JS, Schmidt-RioValle J, Chen Y, Jing J, González-Jiménez E. Metabolic syndrome and its associated early-life factors among Chinese and Spanish adolescents: a pilot study. Nutrients. 2019. https://doi.org/10.3390/nu11071568. source

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

    Premisecitation

    ML methods have increasingly been applied to MetS prediction based on clinical and biochemical data, with encouraging outcomes across various populations.

    “In recent years, ML methods have increasingly been applied to MetS prediction based on clinical and biochemical data, with encouraging outcomes across various populations [46–50].”
    Quote in the paperNot a substantial claim2 of 5 sources checked
    • cites [46] Datta S, Schraplau A, Da Cruz HF, Sachs JP, Mayer F, Böttinger E. A machine learning approach for non-invasive diagnosis of metabolic syndrome. In: 2019 IEEE 19th international conference on bioinformatics and bioengineering (BIBE). IEEE; 2019:933–40. source
    • cites [47] Masi D, Risi R, Biagi F, Vasquez Barahona D, Watanabe M, Zilich R, Gabrielli G, Santin P, Mariani S, Lubrano C, et al. Application of a machine learning technology in the definition of metabolically healthy and unhealthy status: a retrospective study of 2567 subjects suffering from obesity with or without metabolic syndrome. Nutrients. 2022. https://doi.org/10.3390/nu14020373. source

      The Methods (§2.3) and Results (§3.2) describe an ML model using clinical and biochemical variables to distinguish metabolically healthy from unhealthy obesity, with predictive precision of roughly 67–72%. This supports an example of encouraging ML results related to metabolic syndrome, but the study examines one cohort of patients with obesity, not outcomes across various populations or prospective prediction of MetS. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [48] Ivanović D, Kupusinac A, Stokić E, Doroslovački R, Ivetić D. Ann prediction of metabolic syndrome: a complex puzzle that will be completed. J Med Syst. 2016;40(12):264. source
    • cites [49] Zhang H, Chen D, Shao J, Zou P, Cui N, Tang L, Wang X, Wang D, Wu J, Ye Z. Machine learning-based prediction for 4-year risk of metabolic syndrome in adults: a retrospective cohort study. Risk Manag Healthc Policy. 2021;14:4361–8. source
    • cites [50] Shah HA, Liu J, Yang Z, Zhang X, Feng J. Deeprf: a deep learning method for predicting metabolic pathways in organisms based on annotated genomes. Comput Biol Med. 2022;147: 105756. source

  • May combine two claims.

    Findingown data

    At the default 0.5 threshold, XGBoost attained an AUC of 0.94, sensitivity of 0.96, and F1 score of 0.53 in D2.

    “At the default 0.5 threshold, XGBoost attained an AUC of 0.94, sensitivity of 0.96, and F1 score of 0.53 in D2.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Waist circumference had a significant correlation with metabolic syndrome and was included as a feature.

    “WC also had a significant correlation and was included.”
    Quote in the paperSubstantial claim

  • May combine two claims. The quote only partly states this claim.

    Findingown data

    Random Forest, XGBoost, and LightGBM achieved accuracy of 1.00, specificity of 1.00, and F1 scores exceeding 0.98 in D1.

    “In contrast, Random Forest, XGBoost, and LightGBM models achieved the highest performance, with accuracy values of 1.00, specificity of 1.00, and F1 scores exceeding 0.98, demonstrating their superior predictive power.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Further threshold adjustment improved XGBoost's precision from 0.36 to 0.43 while maintaining high sensitivity (0.87).

    “Further threshold adjustment improved XGBoost’s precision from 0.36 to 0.43, while maintaining high sensitivity (0.87).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Random Forest achieved overall accuracy of 0.87, specificity of 0.97, and sensitivity of 0.70 in D1.

    “RF achieved comparable overall accuracy (0.87) and excellent specificity (0.97) but sacrificed sensitivity (0.70)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The baseline model had an accuracy of 0.72 and an F1 score of 0.50.

    “The baseline model showed lower performance with an accuracy of 0.72 and an F1 score of 0.50.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    SVM yielded sensitivity of 0.78, specificity of 0.89, and the lowest Brier score of 0.15 in D1.

    “SVM yielded the most balanced screening profile, coupling sensitivity (0.78) and specificity (0.89) with the lowest Brier score (0.15), indicating wellcalibrated probability estimates.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    At the default 50% threshold, XGBoost detected every metabolic syndrome case in the internal D1 cohort (sensitivity = 1.00) with precision of 0.60.

    “At the default decision threshold of 50%, XGBoost detected every MetS case in the internal D1 cohort (sensitivity = 1.00) but did so with only moderate precision (0.60), reflecting a higher falsepositive burden (Table 5).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the D1 dataset, XGBoost achieved a specificity of 1.00, an F1-score of 0.99, and a precision of 1.00, while the baseline model showed a specificity of 0.33, an F1-score of 0.50, and a precision of 0.70.

    “In the D1 dataset, XGBoost achieved a specificity of 1.00, an F1-score of 0.99, and a precision of 1.00, while the baseline model showed a specificity of 0.33, an F1-score of 0.50, and a precision of 0.70 (see Table 4).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The ANN displayed an F1 score of 0.75 driven by high sensitivity (0.87) but offered no precision advantage over tree-based methods in D1.

    “The ANN displayed a favourable F1score (0.75) driven by high sensitivity (0.87) but offered no precision advantage over treebased methods.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Logistic Regression, KNN, and Naïve Bayes showed lower sensitivities (0.56-0.77), and Naïve Bayes had the poorest precision (0.60) in D1.

    “Simpler learners—Logistic Regression, KNN, and Naïve Bayes—showed lower sensitivities (0.56–0.77) and, in the case of Naïve Bayes, the poorest precision (0.60).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Among all models, XGBoost was the only classifier that maintained high sensitivity (≥0.87) while achieving a notable increase in precision and F1-score under threshold tuning, without compromising overall discrimination.

    “Among all models, XGBoost was the only classifier that maintained high sensitivityl (≥ 0.87) while achieving a notable increase in precision and F1-score under threshold tuning, without compromising overall discrimination.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    At the default 0.50 threshold, XGBoost achieved a high sensitivity of 0.96 but low precision (0.36) on the D2 external validation set.

    “While it initially achieved a high sensitivity of 0.96 but low precision (0.36) at the default 0.50 threshold (Table 7)”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Certain models such as LightGBM and GBDT retained high specificity but failed to correctly identify MetS cases.

    “Although certain models such as LightGBM and GBDT retained high specificity, they failed to correctly identify MetS cases”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The model retained high AUC and sensitivity, but precision declined relative to the internal validation cohort.

    “While the model retained high AUC and sensitivity, precision declined relative to the internal validation cohort.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The true-negative group was not significantly different from the false-positive group in HDL-C but presented better health indicators in other metrics.

    “In contrast, the TN group was not significantly different from the FP group in terms of HDL-C but presented better health indicators in terms of other metrics.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    GBDT achieved a precision of 0.695 at threshold 0.8, but sensitivity dropped sharply to 0.193.

    “GBDT achieved a precision of 0.695 at threshold 0.8, but sensitivity dropped sharply to 0.193”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    XGBoost was selected for further analysis based on its lowest false-negative rate and best overall performance following threshold adjustment.

    “We selected XGBoost for further in-depth analysis based on its lowest false-negative rate and its best overall performance following threshold adjustment.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    When central obesity-related features were excluded, XGBoost preserved moderate sensitivity (0.76) but exhibited reduced precision (0.28) and overall accuracy (0.60).

    “While XGBoost preserved moderate sensitivity (0.76), it exhibited reduced precision (0.28) and overall accuracy (0.60).”
    Quote in the paperSubstantial claim

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

    Findingown data

    The D2 external validation dataset comprised 45,990 true-negative individuals and 9,294 false-positive cases.

    “The dataset comprised 45,990 TN individuals and 9,294 FP cases.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Multiple indicators including age, height, weight, BMI, BRI, WC, blood pressure, and levels of FBG, TG, and HDL-C differed significantly between metabolic syndrome and control groups in both D1 and D2.

    “There were significant differences (P < 0.01) in multiple indicators including age, height, weight, BMI, BRI, WC, blood pressure, and levels of FBG, TG, and HDL-C between MetS and control groups in both datasets, all showing some correlation with MetS.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    There is a critical need for noninvasive, cost-effective diagnostic methods.

    “Thus, there is a critical need for noninvasive, cost-effective diagnostic methods.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The study used 10-fold cross-validation to ensure model robustness and accuracy.

    “To ensure model robustness and accuracy, we used 10-fold cross-validation, where the data was split into 10 parts, each used for validation once, and the model was trained on the remaining 9 parts.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The combination of BRI and machine learning provides a non-invasive and effective method for predicting metabolic syndrome and offers a promising strategy for its early prevention.

    “The combination of BRI and machine learning provides a non-invasive and effective method for predicting MetS and offers a promising strategy for its early prevention.”
    Quote in the paperSubstantial claim

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

    Study design and conductnone given

    Grid search was employed to optimize hyperparameters for each machine learning model.

    “Importantly, grid search was employed to optimize hyperparameters for each machine learning model.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    Class imbalance was managed with the scale_pos_weight parameter in the classifiers.

    “The class imbalance was managed with the scale_pos_weight parameter in the classifiers.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    The model was trained using the D1 dataset and externally validated with the D2 dataset from the Balearic Islands to assess its generalizability across different demographics.

    “The model was trained using the D1 dataset and externally validated with the D2 dataset from the Balearic Islands to assess its generalizability across different demographics.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Previous research on BRI is limited to specific populations, lacking verification in diverse ethnic, geographical, and cultural contexts.

  • May combine two claims.

    Interpretationown data

    Five variables (gender, age, BRI, height, and WC) were chosen, with BRI being the most valuable due to its strongest correlation with metabolic syndrome.

    “Ultimately, five variables (gender, age, BRI, height, and WC) were chosen, with BRI being the most valuable due to its strongest correlation with MetS.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    This approach provides a non-invasive auxiliary strategy for disease risk prediction, which may serve as a complement to traditional diagnostic methods.

    “This approach provides a non-invasive auxiliary strategy for disease risk prediction, which may serve as a complement to traditional diagnostic methods.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    The results of this study highlight the potential of machine learning in improving disease risk stratification.

    “The results of this study highlight the potential of machine learning in improving disease risk stratification.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    BRI may be associated with pathophysiological mechanisms not fully reflected by traditional obesity markers, supporting its role as a candidate risk factor in MetS development.

    “Moreover, our results suggest that BRI may be associated with pathophysiological mechanisms not fully reflected by traditional obesity markers, supporting its role as a candidate risk factor in MetS development.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The shift in class distribution likely contributed to increased false positives and reduced predictive balance under the default threshold.

    “This shift in class distribution likely contributed to increased false positives and reduced predictive balance under the default threshold.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    These findings emphasize the need for localized calibration and underscore the relevance of regional context when applying ML-based prediction tools across diverse healthcare settings.

    “These findings emphasize the need for localized calibration and underscore the relevance of regional context when applying ML-based prediction tools across diverse healthcare settings.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The high sensitivity and precision observed with these models suggest their potential utility in supporting real-world screening programs, providing robust and reliable screening for MetS across diverse populations.

    “The high sensitivity and precision observed with these models suggest their potential utility in supporting real-world screening programs, providing robust and reliable screening for MetS across diverse populations.”
    Quote in the paperSubstantial claim

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

    Interpretationown data

    These findings underscore the practical value of adjusting decision thresholds to improve model adaptability across diverse populations with varying disease prevalence and ethnic backgrounds.

    “These findings underscore the practical value of adjusting decision thresholds to improve model adaptability across diverse populations with varying disease prevalence and ethnic backgrounds.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Central adiposity measures are indispensable for effective metabolic syndrome detection, and BRI is necessary in predictive modeling.

    “These results demonstrate the indispensable contribution of central adiposity measures to effective MetS detection and reinforce the necessity of BRI in predictive modeling.”
    Quote in the paperSubstantial claim

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

    Study design and conductnone given

    The D1 retrospective study was approved by the hospital's ethics committee, adhered to the Declaration of Helsinki, and participants provided informed consent.

    “This retrospective study, designated as D1, was approved by the hospital’s ethics committee and adhered to the Declaration of Helsinki. The participants provided their informed consent.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    XGBoost maximises case detection, while SVM and Random Forest deliver the most clinically useful trade-off between sensitivity and precision for metabolic syndrome screening at the 50% threshold.

    “Taken together, these results indicate that XGBoost maximises case detection, SVM and Random Forest deliver the most clinically useful tradeoff between sensitivity and precision for MetS screening at te 50% threshold.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    The D2 dataset was reassessed using IDF standards and European waist circumference thresholds for consistent diagnostic comparison.

    “The D2 dataset, initially diagnosed by the NCEP-ATP III criteria, was reassessed using the IDF standards for consistent diagnostic comparison. We applied European WC thresholds (≥ 94 cm for males and ≥ 80 cm for females).”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    The generalizability of the findings to other ethnic or regional populations remains to be further validated.

    “As such, the generalizability of our findings to other ethnic or regional populations remains to be further validated.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    This is the first study to incorporate BRI as a core feature in ML-based MetS prediction, addressing a notable gap in existing research.

    “Second, to our knowledge, this is the first study to incorporate BRI as a core feature in ML-based MetS prediction, addressing a notable gap in existing research and highlighting its clinical value as a non-invasive risk marker.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: There is a notable gap in existing research on incorporating BRI as a core feature in ML-based MetS prediction.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationnone given

    The algorithmic complexity and limited interpretability of the machine learning models may impede clinical integration, especially in settings that prioritize transparency and explainability.

    “While the machine learning models demonstrated strong predictive performance, their algorithmic complexity and limited interpretability may impede clinical integration, especially in settings that prioritize transparency and explainability.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The external validation was conducted using a European cohort demographically distinct from the training population, which may not capture the full heterogeneity of global metabolic patterns.

    “The external validation was conducted using a European cohort demographically distinct from the training population, which may not capture the full heterogeneity of global metabolic patterns.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    This work lays a solid foundation for the development of non-invasive screening tools that can be widely implemented across diverse healthcare settings to support early intervention and disease prevention.

    “This work lays a solid foundation for the development of non-invasive screening tools that can be widely implemented across diverse healthcare settings to support early intervention and disease prevention.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    The lack of shared code or datasets in most existing ML studies in this domain limits the ability to systematically evaluate model superiority across different studies and highlights the need for improved transparency and open science practices.

    “This limits the ability to systematically evaluate model superiority across different studies and highlights the need for improved transparency and open science practices in predictive modeling research.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    Due to the cross-sectional nature of the data, individual changes over time could not be assessed, limiting the capacity to reflect the temporal dynamics of disease progression.

    “Additionally, due to the cross-sectional nature of the data, individual changes over time could not be assessed, limiting our capacity to reflect the temporal dynamics of disease progression.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    We are confident in the potential of this approach to support early screening and individualized risk stratification, particularly within the Chinese population.

    “We are confident in the potential of this approach to support early screening and individualized risk stratification, particularly within the Chinese population.”
    Quote in the paperNot a substantial claim

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

    Study design and conductown data

    Due to the retrospective design, causal inferences cannot be established.

    “Another notable limitation is that, due to the retrospective design, causal inferences cannot be established.”
    Quote in the paperNot a substantial claim
65 of 65

  • The paper relies on this without support.

    In existing studies, the potential of non-invasive indicators to integrate with machine learning algorithms to predict metabolic syndrome has not been fully explored.

    “However, in existing studies, the potential of non-invasive indicators to integrate with ML algorithms to predict MetS has not been fully explored.”

  • The paper relies on this without support.

    Previous research on BRI is limited to specific populations, lacking verification in diverse ethnic, geographical, and cultural contexts.

    “However, previous research on BRI is limited to specific populations, lacking verification in diverse ethnic, geographical, and cultural contexts.”

  • The paper relies on this without support.

    There is a notable gap in existing research on incorporating BRI as a core feature in ML-based MetS prediction.

    “Second, to our knowledge, this is the first study to incorporate BRI as a core feature in ML-based MetS prediction, addressing a notable gap in existing research and highlighting its clinical value as a non-invasive risk marker.”
3 of 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.

  • SVM is incorrectly described as having the lowest Brier score

    Major

    The text assigns SVM a Brier score of 0.15 and calls it the lowest. Table 5 reports 0.10 for SVM and 0.03 for both Decision Tree and Random Forest.

  • Claimed sensitivity advantage conflicts with the baseline results

    Major

    The baseline has sensitivity 1.00 in Table 4, whereas the listed machine-learning models have sensitivities of 0.68-0.98. They therefore do not show the stated advantage over the baseline in sensitivity.

  • Statement that all analyses used t-tests conflicts with listed group tests

    Minor

    The statistical methods specify Mann-Whitney U and chi-square tests for group comparisons, then state that all analyses used two-tailed t-tests.

  • Claim of improvement in ROC AUC does not hold for every model

    Minor

    The discussion says all ML models improved ROC AUC over the baseline, but Table 4 gives Logistic Regression an AUC of 0.82 versus the baseline's 0.83.

  • Table 8 does not report the claimed body-weight comparison

    Minor

    The text says Table 8 shows differences in body weight and BMI. Its two reported measures are BRI and BMI; body weight is absent.

  • D1 BRI correlation has two reported values

    Minor

    The feature-selection text reports a BRI-MetS correlation of 0.582 for D1, while Table 1 reports 0.585.

6 of 6

Cites

Flagged first, then by citations

Cited by

Still being checked.

Related

Authors
He, Y., Chen, Z., Tang, Z., Qin, Y., & Wang, F.
Affiliations
  • Third Xiangya Hospital (CN)
  • Central South University (CN)
Corresponding author
  • Qin, Yuexiang
  • Wang, Fang
ORCID
Venue
BMC Public Health
Publisher
Springer Science and Business Media LLC
Published
2025-08-01
Volume
25
Issue
1
Pages
2608-2608
ISSN
1471-2458 (ISSN-L)
DOI
10.1186/s12889-025-23858-y
PMC
PMC12315366
OpenAlex
W4412842535
Cited by
5
References
51
Citations by year
2026: 4 · 2025: 1
FWCI
3.69
Type
journal article
Language
en
Open access
Licence
Locations
  • BMC Public Health (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • Europe PMC (PubMed Central) (submittedVersion, cc-by-nc-nd)
  • PubMed Central (submittedVersion, cc-by-nc-nd)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • 2025 Central South University Graduate Student Independent Exploration and Innovation Project (2025ZZTS0973)
  • National Natural Science Foundation of China (82303133)
  • Hunan Provincial Health Commission (B202307017799)
  • Hunan Provincial Natural Science Foundation (2021JJ40951, 2024JJ5520)
  • Changsha Municipal Natural Science Foundation (kq2403054)
Topics
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
  • Health Sciences › Medicine › Endocrinology, Diabetes and Metabolism › Diabetes, Cardiovascular Risks, and Lipoproteins
  • Health Sciences › Medicine › Physiology › Body Composition Measurement Techniques
Keywords
metabolic syndrome, body roundness index, non-invasive prediction model, metabolic syndrome screening, abdominal obesity, waist circumference, XGBoost, threshold sensitivity analysis, cardiovascular risk prevention, IDF criteria, body composition assessment, model interpretability, external validation, health examination data, model robustness
MeSH terms
Machine Learning, Adult, Aged, Algorithms, Female, Humans, Male, Middle Aged, Retrospective Studies, Metabolic Syndrome, Waist Circumference

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 2026-10-20.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-20.

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