CoChat
Sign Up
Cited by 4Liver Disease Diagnosis and TreatmentLiver Disease and TransplantationArtificial Intelligence in Healthcare

Fibro predict a machine learning risk score for advanced liver fibrosis in the general population using Israeli electronic health records

Open access
Kalka, Iris N., Hazzan, Rawi, Yacovzada, Nancy-Sarah, Igbaria, Saleh, Segal, Eran, Weinberger, Adina, Neeman, Ziv·2025·Scientific Reports
DOI 10.1038/s41598-025-17534-9PMC PMC12399762
Open this paper in CoChat — free accountOpen PDF

Abstract

Liver diseases, notably cirrhosis, pose a substantial global health challenge, resulting in millions of annual deaths. Existing diagnostic methods primarily target high-risk groups, leaving a significant portion of patients undiagnosed. This study aims to develop and validate an a machine-learning prediction model, Fibro-Predict, for the early detection of advanced liver fibrosis in the general population using nationwide electronic health records (EHRs). We constructed a diagnostic framework for liver cirrhosis by analyzing retrospective EHR data from 2,255,580 observations in Israel's largest healthcare provider. The Fibro-Predict model was trained using gradient boosted trees (XGBoost) predicting five-year disease diagnosis trajectories based on routine blood tests and validated both temporally and externally. We conducted a retrospective temporal validation of the model and an external prospective validation in clinical settings, employing transient elastography. The temporal validation of Fibro-Predict demonstrated a promising five-year AUC of 0.81 (95% CI 0.80-0.82) in the training set and 0.79 (95% CI 0.78-0.80) in the validation set. In a clinical context, our framework exhibited an impressive True Positive Rate (TPR) of 36.8% (28/76) when comparing predicted risk to observed outcomes, surpassing the widely used FIB-4, which had a TPR of only 3.7% (1/27). Fibro-Predict, relying solely on routine blood tests and standard demographics, emerges as a valuable tool for cost-effective patient prioritization in advanced fibrosis screening within the general population. By leveraging nationwide EHR data, this approach allows healthcare systems to flag potentially undiagnosed patients earlier and more broadly, streamlining clinical follow-up and expediting diagnosis. This approach holds the potential to significantly improve the early detection of advanced liver fibrosis and subsequently reduce its associated morbidity and mortality.Trial registration: ClinicalTrials.gov Identifier: NCT05218538.

AI snapshot

Written from the paper’s full text.

TL;DR

Fibro-Predict is an XGBoost survival model using routine blood tests and demographics from Israeli electronic health records to predict five-year risk of advanced liver fibrosis/cirrhosis in the general population. It outperformed FIB-4 in temporal validation (AUC 0.79 vs. 0.71) and in a prospective clinical cohort using transient elastography.

Key findings

  • In a retrospective cohort of 2,255,580 observations from Clalit (2004–2020), 57.5% were female, mean age was 56.9 (SD 9.9), and 11,337 individuals were diagnosed with liver cirrhosis during five-year follow-ups (0.5% of observations).
  • Temporal validation showed Fibro-Predict five-year AUC of 0.81 (95% CI 0.80–0.82) in the training set and 0.79 (95% CI 0.78–0.80) in the validation set; FIB-4 achieved AUC 0.71 in both train and validation sets on observations with available scores.
  • In the prospective external cohort, 90,136 individuals were eligible; 579 were invited (435 Fibro-Predict, 144 FIB-4), and 103 attended (76 Fibro-Predict, 27 FIB-4). Mean liver stiffness was 13.1 kPa for the Fibro-Predict group vs. 6.5 kPa for the FIB-4 group (two-sided Mann-Whitney P < 10⁻²).
  • Advanced fibrosis (kPa > 12) was reported in the abstract as a TPR of 36.8% (28/76) for Fibro-Predict vs. 3.7% (1/27) for FIB-4; the Results text reports 21/76 (27.6%) Fibro-Predict participants confirmed with advanced fibrosis vs. 1/27 (3.7%) FIB-4 participants. SHAP analysis identified AST as the top contributor.

Methods

  • Retrospective EHR study using Clalit data from >5.5 million members, spanning January 1, 2004–December 31, 2020. Data were split with a rolling-origin-update design into three non-overlapping five-year periods: 2005–2010, 2010–2015, and 2015–2020. Eligibility required age 40–75, valid hemoglobin, platelet, and white blood cell counts within one year before the index date, and exclusions for prior cirrhosis or predefined diagnoses.
  • Model inputs were sex, age, and latest lab results: HB, PLT, WBC, AST, ALT, albumin, bilirubin, PT-INR, vitamin B12, glucose, HbA1c, cholesterol, HDL, LDL, triglycerides, total protein, and days from latest blood test. An XGBoost survival model (`objective="survival:cox"`) was trained on the first two periods and validated on the latest period; no imputation was applied, and missingness was handled internally. Performance was evaluated with annual time-dependent C/D ROC AUCs and compared with FIB-4.
  • Prospective external validation was performed in the Afula subdistrict: 90,136 eligible individuals had Fibro-Predict and FIB-4 scores computed; the highest-risk individuals were selected in a 3:1 ratio (Fibro-Predict:FIB-4) with no overlap, double-blinded. Of 579 invited, 103 attended and underwent transient elastography, height/weight measurement, and AUDIT questionnaire; advanced fibrosis was defined as kPa > 12. Recruitment ran April 19–August 24, 2022.

Limitations

  • The paper states: no formal power calculation was performed; no subgroup-specific data quality assessments were done; no adjustment was made for heterogeneity across geographic regions or healthcare sites; training used a single national provider as a pooled cohort; no class imbalance correction was applied; death before cirrhosis was treated as right-censoring rather than a competing risk; treatment data were not used as model input or explicitly tracked; and proportional hazards were not formally assessed.
  • The paper also notes: only 103 of 579 invited individuals attended the prospective cohort; 33 died before recruitment; diagnosis data were unavailable for the prospective cohort; and FIB-4 availability correlated with likelihood of cirrhosis diagnosis, with causal direction requiring further research.
  • The text makes evident that external validation was limited to the Afula subdistrict, the prospective sample was small and based on those who accepted invitation, and the abstract’s TPR (28/76) differs from the Results text’s confirmed advanced fibrosis count (21/76), which may complicate interpretation.

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

Integrity signals

Registry

Crossref · 10.1038/s41598-025-17534-9

Cites

24 of 24 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 2 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Key demographic and predictor variable distributions were broadly consistent between the retrospective development cohort and the prospective validation cohort.

    “Key demographic and predictor variable distributions were broadly consistent between the retrospective development cohort and the prospective validation cohort, supporting the comparability of these populations for model evaluation”
    Quote in the paperSubstantial claim
  • Findingown data

    The only participant from the FIB-4 group found to have advanced liver fibrosis had an AUDIT score above 15, indicating moderate-severe alcohol use disorder.

    “the only participant from the FIB-4 group, found to have advanced liver fibrosis, was the only participant with an Alcohol Use Disorders Identification Test (AUDIT)23 score above 15, indicating a moderate-severe alcohol use disorder”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [23] Saunders, J. B., Aasland, O. G., Babor, T. F., de la Fuente, J. R. & Grant, M. Development of the alcohol use disorders identification test (AUDIT): WHO collaborative project on early detection of persons with harmful alcohol Consumption–II. Addiction 88, 791–804 (1993). source
  • Findingnone given

    Fibro-Predict relies solely on routine blood tests and standard demographics.

    “Fibro-Predict, relying solely on routine blood tests and standard demographics, emerges as a valuable tool”
    Quote in the paperSubstantial claim
  • Findingown data

    FIB-4 achieved an AUC of 0.71 for both train and validation sets on observations with available scores.

    “outperforming FIB-4, which achieved an AUC of 0.71 for both train and validation sets on observations with available scores”
    Quote in the paperSubstantial claim
  • Findingown data

    TE examinations of the 103 participants showed a significant difference in mean liver stiffness: Fibro-Predict group kPa = 13.1 versus FIB-4 group kPa = 6.5 (two-sided Mann-Whitney P < 10−2).

    “TE examinations of the 103 participants showed a significant difference in the mean liver stiffness of the Fibro-Predict group, kPa = 13.1, versus the FIB-4 group, kPa = 6.5 (two-sided Mann-Whitney P < 10− 2)”
    Quote in the paperSubstantial claim
  • May combine two claims. 23
  • Classified as probably not a substantial claim. 23
  • The quote only partly states this claim. 1

  • May combine two claims.

    Findingown data

    Of the 103 participants, 21 (27.6%) from the Fibro-Predict group were confirmed as having advanced liver fibrosis compared to only one (3.7%) from the FIB-4 group.

    “Of the 103 participants, 21 (27.6%) from the Fibro-Predict group were confirmed as having advanced liver fibrosis compared to only one (3.7%) from the FIB-4 group.”
    Quote in the paperSubstantial claim

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

    Findingown data

    In the prospective external cohort, 90,136 individuals passed inclusion-exclusion criteria.

    “Of those, 90,136 individuals passed the inclusion-exclusion criteria.”
    Quote in the paperNot a substantial claim

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

    Findingreasoning

    There is a correlation between the availability of FIB-4 scores and the likelihood of being diagnosed with advanced liver fibrosis.

    “An interesting observation in our data is the correlation between the availability of FIB-4 scores and the likelihood of being diagnosed with advanced liver fibrosis.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    Of the 579 individuals, 33 passed away before recruitment (28/435 from the Fibro-Predict group and 5/144 from the FIB-4 group, corresponding to 6.4% and 3.5%, respectively).

    “Of the 579 individuals, 33 passed away before recruitment (28/435 from the Fibro-Predict group and 5/144 from the FIB-4 group, corresponding to 6.4% and 3.5%, respectively).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In the clinical context, Fibro-Predict exhibited a True Positive Rate (TPR) of 36.8% (28/76) when comparing predicted risk to observed outcomes, surpassing FIB-4, which had a TPR of 3.7% (1/27).

    “our framework exhibited an impressive True Positive Rate (TPR) of 36.8% (28/76) when comparing predicted risk to observed outcomes, surpassing the widely used FIB-4, which had a TPR of only 3.7% (1/27)”
    Quote in the paperSubstantial claim

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

    Findingown data

    A pooled sample of 435 individuals obtained the highest risk score according to Fibro-Predict, and 144 individuals had the highest FIB-4 scores, with no intersection between groups.

    “To maintain the ratio, a pooled sample of 435 individuals obtained the highest risk score according to Fibro-Predict, and 144 individuals had the highest FIB-4 scores. Notably, there was no intersection between individuals in the model and the FIB-4 group.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Both Fibro-Predict and FIB-4 exhibited high negative predictive value (NPV).

    “both scores exhibited high negative predictive value (NPV)”
    Quote in the paperNot a substantial claim

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

    Findingown data

    FIB-4 scores were computed for 86% of the training set and 90% of the validation set.

    “FIB-4 scores were computed for all observations with the requisite blood tests, consisting of 86% of the training set and 90% of the validation set.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    In the prospective cohort, 103 individuals accepted enrollment and underwent non-invasive physical examination (76 from Fibro-Predict group and 27 from FIB-4 group).

    “Of the remaining 546 individuals,103 accepted enrollment in the prospective study, arriving at the clinic and undergoing a non-invasive physical examination (76 from the Fibro-Predict group and 27 from the FIB-4 group)”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    The retrospective cohort comprised 2,255,580 observations from Clalit Health Services EHR data (2004-2020).

    “A comprehensive dataset of 2,255,580 observations, each representing an eligible individual at the respective index date, was curated based on predefined inclusion and exclusion criteria”
    Quote in the paperSubstantial claim

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

    Findingreasoning

    This participant was never diagnosed with alcohol abuse and therefore was not excluded from the study.

    “This participant was never diagnosed with alcohol abuse and, therefore, was not excluded from the study.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    When sensitivity was fixed, Fibro-Predict consistently outperformed FIB-4 in specificity and PPV.

    “when sensitivity was fixed, Fibro-Predict consistently outperformed FIB-4 in specificity and PPV”
    Quote in the paperSubstantial claim

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

    Findingown data

    Among eligible observations, 57.5% were female and the average age was 56.9 (SD = 9.9).

    “Among eligible observations, 57.5% were female, and the average age was 56.9 (SD = 9.9)”
    Quote in the paperNot a substantial claim

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

    Findingown data

    FIB-4 and Fibro-Predict's scores were computed for the eligible 90,136 individuals, 2.5% of whom had missing data and could not receive a FIB-4 score.

    “FIB-4 and Fibro-Predict’s scores were computed for the eligible (Table S5) 90,136 individuals (2.5% of whom had missing data and could not receive a FIB-4 score).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Holding specificity constant, Fibro-Predict consistently achieved higher sensitivity and positive predictive value (PPV) than FIB-4 at all levels.

    “Holding specificity constant, Fibro-Predict consistently achieved higher sensitivity and positive predictive value (PPV) than FIB-4 at all levels”
    Quote in the paperSubstantial claim

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

    Findingown data

    In the prospective cohort, 115,211 Afula subdistrict's Clalit members were obtained, of whom 90,136 passed inclusion-exclusion.

    “Age, gender, and latest laboratory test results (as of December 2021) were obtained from 115,211 Afula subdistrict’s Clalit members (see Methods). Of those, 90,136 individuals passed the inclusion-exclusion criteria.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Fibro-Predict demonstrated an AUC of 0.81 for the train set and 0.79 for the validation set in temporal validation.

    “Fibro-Predict demonstrated an Area Under the C/D ROC (AUC) of 0.81 for the train set and AUC = 0.79 for the validation set”
    Quote in the paperSubstantial claim

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

    Findingown data

    Within the retrospective cohort, 11,337 individuals were diagnosed with liver cirrhosis during their respective five-year follow-ups, constituting 0.5% of total observations.

    “Within this cohort, 11,337 individuals were diagnosed with liver cirrhosis during their respective five-year follow-ups, constituting 0.5% of the total observations”
    Quote in the paperNot a substantial claim

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

    Study design and conductreasoning

    We did not examine or adjust for heterogeneity in model performance or parameter estimates across clusters such as geographic regions or healthcare sites.

    “We did not examine or adjust for heterogeneity in model performance or parameter estimates across clusters such as geographic regions or healthcare sites.”
    Quote in the paperNot a substantial claim

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

    Study design and conductreasoning

    The study was exempt from the requirement for patients' written informed consent because it is based on retrospective data.

    “This study is based on retrospective data; therefore, it was exempt from the requirement for patients’ written informed consent.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    The retrospective portion of the study constitutes one of the largest cohorts of its kind, spanning an extensive time.

    “The retrospective portion of our study constitutes one of the largest cohorts of its kind, spanning an extensive time.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductreasoning

    No class imbalance correction methods were applied; the use of a Cox survival model allowed incorporation of time-to-event and censoring information, mitigating the need for binary classification thresholding or label balancing.

    “Although the outcome of interest—liver cirrhosis diagnosis—is relatively rare in the dataset, no class imbalance correction methods (e.g., reweighting or resampling) were applied. The use of a Cox survival model allowed the algorithm to incorporate time-to-event and censoring information directly, mitigating the need for binary classification thresholding or label balancing.”
    Quote in the paperSubstantial claim

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

    Study design and conductreasoning

    The model was trained using time-fixed covariates derived from baseline measurements, and proportional hazards were not formally assessed; however, gradient boosted Cox models allow flexible modeling of non-linear effects without assuming strict proportional hazards.

    “The model was trained using time-fixed covariates derived from baseline (index date) measurements, and we did not formally assess proportional hazards; however, the use of gradient boosted Cox models allows for flexible modeling of non-linear effects without assuming strict proportional hazards.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    This correlation might be because liver function blood tests, including AST and ALT, are typically part of standard clinical practice when investigating liver-related concerns.

    “This might be because liver function blood tests, including AST and ALT, are typically part of the standard clinical practice when investigating liver-related concerns.”
    Quote in the paperNot a substantial claim

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

    Study design and conductreasoning

    No formal power calculation was performed; however, the large sample size of over 2 million observations provides robust statistical power for model training and validation.

    “No formal power calculation was performed; however, the large sample size of over 2 million observations provides robust statistical power for model training and validation.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Fibro-Predict holds the potential to significantly improve early detection of advanced liver fibrosis and subsequently reduce its associated morbidity and mortality.

    “This approach holds the potential to significantly improve the early detection of advanced liver fibrosis and subsequently reduce its associated morbidity and mortality.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    Computed risk scores were concealed during the clinical trial to ensure a double-blinded study.

    “Computed risk scores were concealed during the clinical trial to ensure a double-blinded study.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    The study was approved by Clalit Helsinki committee 0195-17-COM2.

    “The electronic study protocol was approved by the Clalit Helsinki committee 0195-17-COM2.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductnone given

    All training data were obtained from a single national healthcare provider (Clalit) using unified coding and data standards.

    “All training data were obtained from a single national healthcare provider (Clalit) using unified coding and data standards.”
    Quote in the paperSubstantial claim

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

    Study design and conductnone given

    The fibro-predict and FIB-4 scores were hidden from the study team to maintain a double-blinded study.

    “The fibro-predict and FIB-4 scores were hidden from the study team to maintain a double-blinded study.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    Emek Medical Center Helsinki Committee 0208-20-EMC approved the clinical cohort and the entire study protocol.

    “Emek Medical Center Helsinki Committee 0208-20-EMC approved the clinical cohort and the entire study protocol.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    These findings collectively indicate that Fibro-Predict excels in accurately identifying actual positive cases while maintaining specificity and positive predictive value (PPV), making it a preferred choice across various specificity and sensitivity requirements.

    “These findings collectively indicate that Fibro-Predict excels in accurately identifying actual positive cases while maintaining specificity and positive predictive value (PPV), making it a preferred choice across various specificity and sensitivity requirements.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    By leveraging nationwide EHR data, Fibro-Predict allows healthcare systems to flag potentially undiagnosed patients earlier and more broadly, streamlining clinical follow-up and expediting diagnosis.

    “By leveraging nationwide EHR data, this approach allows healthcare systems to flag potentially undiagnosed patients earlier and more broadly, streamlining clinical follow-up and expediting diagnosis.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Based on real-world EHRs, Fibro-Predict can be seamlessly implemented across the entire population without requiring proactive routine checkups, accommodating missing data, and with a low referral rate, adding minimal cost to the Healthcare System.

    “Based on real-world Electronic Health Records (EHRs), our work can be seamlessly implemented across the entire population without requiring proactive routine checkups, accommodating missing data, and with a low referral rate, adding minimal cost to the Healthcare System.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    Individuals with a higher risk of advanced liver fibrosis are more likely to have available FIB-4 scores.

    “This creates a two-way relationship: individuals with a higher risk of advanced liver fibrosis are more likely to have available FIB-4 scores.”
    Quote in the paperNot a substantial claim
35 of 35

  • The authors state this as an assumption.

    The validation set (last five years of data) might contain underlying information that affects model performance (e.g., changes due to COVID-19 or regulations).

    “This was based on the assumption that the validation set, which comprised the last five years of data, might contain underlying information that affects model performance (e.g., changes due to COVID-19 or regulations).”

  • The authors state this as an assumption.

    Death prior to cirrhosis diagnosis can be appropriately treated as a right-censoring event rather than a competing risk, consistent with standard Cox model assumptions.

    “We note that death prior to cirrhosis diagnosis was treated as a right-censoring event rather than as a competing risk, consistent with standard Cox model assumptions; as such, competing risks were not explicitly modeled.”

  • The paper relies on this without support.

    ICD-9 diagnosis codes accurately identify liver cirrhosis.

    “Individual follow-up ended at an observed outcome event of liver cirrhosis diagnosis (Table S10)”

  • The paper relies on this without support.

    The participants who attended the clinic (103 of 579 invited) are representative of the high-risk individuals invited, so the observed TPR and liver stiffness differences are not due to selection bias.

    “Of the 579 individuals, 33 passed away before recruitment (28/435 from the Fibro-Predict group and 5/144 from the FIB-4 group, corresponding to 6.4% and 3.5%, respectively). Of the remaining 546 individuals,103 accepted enrollment”

  • The paper relies on this without support.

    The Afula subdistrict prospective cohort is representative of the general population for the purpose of evaluating Fibro-Predict.

    “In the prospective clinical cohort, Fibro-Predict surpassed the widely used FIB-4 score in diagnosing advanced liver fibrosis in the general population.”

  • The paper relies on this without support.

    The missing data mechanism is handled appropriately by XGBoost without imputation.

    “Therefore, no imputation was applied, missingness was handled internally by the model.”
6 of 6

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.

  • Prospective fibrosis detection count differs between abstract and results

    Major

    The abstract reports 28 of 76 Fibro-Predict participants with advanced fibrosis (36.8%), whereas the Results report 21 of 76 (27.6%). This changes the reported outcome of the prospective comparison.

  • FIB-4 scores were not available for all prospective individuals

    Minor

    The figure caption says FIB-4 risks were computed for all individuals, but the prospective evaluation says 2.5% could not receive a FIB-4 score.

2 of 2

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. Lundberg, S. M. & Lee, S. I. A unified approach to interpreting model predictions. Adv. Neural. Inf. Process. Syst. 2017-Decem, 4766–4775 (2017).

Cited by

Sorted by citations

Related

Authors
Kalka, I. N., Hazzan, R., Yacovzada, N.-S., Igbaria, S., Segal, E., Weinberger, A., & Neeman, Z.
Affiliations
  • Weizmann Institute of Science (IL)
  • Bar-Ilan University (IL)
  • Emek Medical Center (IL)
  • Technion – Israel Institute of Technology (IL)
Corresponding author
Neeman, Ziv
ORCID
Venue
Scientific Reports
Publisher
Springer Science and Business Media LLC
Published
2025-09-01
Volume
15
Issue
1
Pages
32035-32035
ISSN
2045-2322 (ISSN-L)
DOI
10.1038/s41598-025-17534-9
PMC
PMC12399762
OpenAlex
W4413866760
Cited by
4
References
24
Citations by year
2026: 4
FWCI
2.41
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • Europe PMC (PubMed Central) (submittedVersion, cc-by-nc-nd)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • PubMed Central (submittedVersion, cc-by-nc-nd)
Topics
  • Health Sciences › Medicine › Epidemiology › Liver Disease Diagnosis and Treatment
  • Health Sciences › Medicine › Hepatology › Liver Disease and Transplantation
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
Keywords
advanced liver fibrosis, liver cirrhosis, machine learning prediction models, early detection, general population screening, routine blood tests, electronic health records, transient elastography, FIB-4, XGBoost, risk score, undiagnosed patients
MeSH terms
Machine Learning, Boosting Machine Learning Algorithms, Adult, Aged, Female, Humans, Israel, Liver Cirrhosis, Male, Middle Aged, Retrospective Studies, Risk Factors, Risk Assessment, Elasticity Imaging Techniques, Electronic Health Records

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 2026-10-27.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-27.

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