Skip to main content
Tessa NewsLink
Paste a health news link, or browse

Source study found

Story checked

AI tool provides early warning for organ failure in acute pancreatitis (opens in a new tab)

news-medical.net · 2026-09-24

Short answerEvidenceSource

Short answer

Mixed

Mixed.

One claim goes further than the study. One other point was not covered by the paper.

  • 4 supported
  • 1 overstated
  • 1 not covered

Checked against the study summary. The full text wasn't available, so some details couldn't be settled either way.

Share this check

Follow the evidence trail
1
2

NewsLink checks it

Mixed

One claim overstates the study. Four of six check out. One claim the study doesn't address.

  • 4 supported
  • 1 overstated
  • 1 not covered
Open claim evidence
3

The source study

Prediction of organ failure in acute pancreatitis via CT: A multicenter deep learning model with early clinical utility

Journal of Pancreatology · 2026
Open source record (opens in a new tab)
Then inspect each claim

Evidence layer

Claim by claim

Each claim gets a verdict. Expand it to see the evidence directly below.

6 claims in this story

Showing all 6 claimsChoose a verdict to focus the list.

Claim 1 of 6OverstatedThe authors say independent multicenter validation confirmed the tool's generalizability and that automated imaging-based risk stratification could support earlier intervention and better resource allocation in emergency and critical care settings.View evidenceHide evidence

Why this verdict

The profile says multicenter validation is presented as evidence of generalizability and that the tool could support risk-stratified management. But saying validation 'confirmed' generalizability is stronger than the abstract supports, because the study was retrospective and limited to two tertiary hospitals. Claims about earlier intervention and resource allocation are prospective clinical-utility implications, not demonstrated clinical impact in the abstract evidence.

1

Study evidence

ORACLE discrimination for predicting organ failure: AUC 0.85 (training), 0.89 (validation), 0.81 (test).AUC 0.85 (train); 0.89 (val); 0.81 (test)

“This multicenter study included 2746 AP patients from two tertiary hospitals (2011–2024). Patients were split into training (n=1820), validation (n=456), and test cohorts (n=470).”
Claim 2 of 6Not coveredResearchers at Changhai Hospital and Shanghai 411 Hospital developed ORACLE, which combines deep learning radiomics from multiphase CT scans with clinical variables to predict organ failure automatically.View evidenceHide evidence

Why this verdict

The technical description is supported: ORACLE integrates multiphase CT-derived deep learning radiomics with clinical variables, with automated CT segmentation as an enabling step. However, the abstract profile does not name Changhai Hospital or Shanghai 411 Hospital as the developing institutions, so that attribution is not verifiable at the supplied depth.

1

Study evidence

ORACLE discrimination for predicting organ failure: AUC 0.85 (training), 0.89 (validation), 0.81 (test).AUC 0.85 (train); 0.89 (val); 0.81 (test)

“This multicenter study included 2746 AP patients from two tertiary hospitals (2011–2024). Patients were split into training (n=1820), validation (n=456), and test cohorts (n=470).”
2

Study evidence

An nnMamba-based segmentation model was used to delineate pancreatic and peripancreatic regions on multiphase CT and served as the automated ROI-definition step for downstream radiomics-driven organ-failure prediction.

“An nnMamba-based segmentation model delineated pancreatic/peripancreatic regions on CT.”
Claim 3 of 6SupportedA new study published online in the Journal of Pancreatology reports an AI tool for predicting organ failure in acute pancreatitis.View evidenceHide evidence

Why this verdict

The abstract-level profile supports that the paper reports an AI/CT-based ORACLE tool for predicting organ failure in acute pancreatitis. Publication timing and venue details are not substantively profiled, but the central scientific claim is supported.

1

Study evidence

ORACLE discrimination for predicting organ failure: AUC 0.85 (training), 0.89 (validation), 0.81 (test).AUC 0.85 (train); 0.89 (val); 0.81 (test)

“This multicenter study included 2746 AP patients from two tertiary hospitals (2011–2024). Patients were split into training (n=1820), validation (n=456), and test cohorts (n=470).”
Claim 4 of 6SupportedIn a multicenter study of 2,746 acute pancreatitis patients, ORACLE achieved AUCs of 0.85, 0.89, and 0.81 across training, validation, and independent external test cohorts, outperforming the Modified CT Severity Index and clinical models.View evidenceHide evidence

As statedAUCs of 0.85, 0.89, and 0.81

Why this verdict

The profile reports n=2746 acute pancreatitis patients, training/validation/test cohorts, ORACLE AUCs of 0.85, 0.89, and 0.81, and better AUC performance than M-CTSI and clinical models.

1

Study evidence

ORACLE discrimination for predicting organ failure: AUC 0.85 (training), 0.89 (validation), 0.81 (test).AUC 0.85 (train); 0.89 (val); 0.81 (test)

“This multicenter study included 2746 AP patients from two tertiary hospitals (2011–2024). Patients were split into training (n=1820), validation (n=456), and test cohorts (n=470).”
2

Study evidence

ORACLE achieved higher discriminative performance for predicting organ failure than M-CTSI and the reported clinical models on the same cohorts.ORACLE AUCs: 0.85 (training), 0.89 (validation), 0.81 (test) vs M-CTSI AUC 0.68–0.74 and clinical models AUC 0.67–0.71; DeLong’s P < 0.001 (abstract).

“...outperforming Modified CT Severity Index (M-CTSI) (AUC 0.68–0.74) and clinical models (AUC 0.67–0.71; DeLong’s P <0.001).”
Claim 5 of 6SupportedThe model provided a median early warning of 3.5 hours before organ failure became clinically apparent, and 55% of cases were predicted at least 3 hours in advance.View evidenceHide evidence

As stated3.5 hours; 55% of cases

Why this verdict

The abstract profile reports a median early warning time of 3.5 hours before clinical organ-failure onset and that 55% of predicted cases were identified at least 3 hours in advance. The abstract does not detail timing definitions, but the stated metrics match the profile.

1

Study evidence

ORACLE discrimination for predicting organ failure: AUC 0.85 (training), 0.89 (validation), 0.81 (test).AUC 0.85 (train); 0.89 (val); 0.81 (test)

“This multicenter study included 2746 AP patients from two tertiary hospitals (2011–2024). Patients were split into training (n=1820), validation (n=456), and test cohorts (n=470).”
Claim 6 of 6SupportedAmong high-risk patients flagged by the model, the incidence of organ failure reached 92.1%, and the overall negative predictive value was 97.2%.View evidenceHide evidence

As stated92.1%; 97.2%

Why this verdict

The profile reports that the high-risk subgroup had 92.1% organ-failure incidence and that the overall negative predictive value was 97.2%.

1

Study evidence

ORACLE discrimination for predicting organ failure: AUC 0.85 (training), 0.89 (validation), 0.81 (test).AUC 0.85 (train); 0.89 (val); 0.81 (test)

“This multicenter study included 2746 AP patients from two tertiary hospitals (2011–2024). Patients were split into training (n=1820), validation (n=456), and test cohorts (n=470).”
Then look for missing context

Context layer

What the story left out

Important study details the story did not include.

  • The study is a multicenter retrospective cohort of 2,746 patients from two tertiary hospitals, split into training, validation, and test cohorts.

    The story reflects the multicenter sample size and cohort split, but it does not mention the retrospective design or the limitation that the abstract-level evidence comes from only two tertiary hospitals.

    From Retrospective multicenter cohort prediction-model development and validation

  • Primary outcome was organ failure defined as Modified Marshall Score ≥2.

    The story discusses organ failure prediction but does not specify the operational outcome definition used in the paper.

    From Retrospective multicenter cohort prediction-model development and validation

  • NPV and high-risk subgroup interpretation require context: NPV depends on the 8.7% organ-failure prevalence, and the high-risk subgroup comprised only 1.4% of the cohort.

    The story reports the favorable metrics but does not mention the prevalence dependence of NPV or the small size of the high-risk stratum.

    From Retrospective multicenter cohort prediction-model development and validation

  • The abstract does not report calibration, confidence intervals, sensitivity/specificity thresholds, decision-curve analysis, or prospective clinical-impact evaluation.

    These are material limitations for interpreting a prediction model’s readiness for clinical triage, and the story’s caveats do not acknowledge them.

    From Retrospective multicenter cohort prediction-model development and validation; Observational prediction-model comparative

  • An nnMamba-based automated segmentation model delineated pancreatic/peripancreatic CT regions, but the abstract provides no segmentation performance or validation metrics.

    The story broadly says prediction was automated from CT, but it does not describe the segmentation component or the lack of reported segmentation-performance details.

    From Deep learning segmentation (nnMamba)

3 things the story did carry across
  • ORACLE is a CT-based deep learning radiomics plus clinical-variable model for early organ-failure prediction in acute pancreatitis.
  • ORACLE achieved AUCs of 0.85, 0.89, and 0.81 and outperformed M-CTSI and clinical models by AUC.
  • The model’s reported clinical-utility metrics include 97.2% overall NPV, 92.1% organ-failure incidence in a small high-risk subgroup, and 3.5-hour median early warning time.
Then read the study layer

Study layer

Study at a glance

Scan the study first. Expand only the parts you want to inspect.

Pieces of work

3

Evidence read

study summary

Lead result

human in vivo

1Lead resulthuman in vivoDevelop and validate a CT-based deep learning model (ORACLE) that integrates automated CT segmentation, deep learning radiomics, and clinical variables to predict early organ failure in acute pancreatitis, with multicenter generalizability and clinical utility (early warning time, NPV, high-risk stratification).Retrospective multicenter cohort prediction-model development and validationExpand

In plain English

Multicenter retrospective study (n=2746 from two tertiary hospitals, 2011–2024) developing and validating ORACLE, an automated CT-based deep learning radiomics plus clinical-variable model (SOF-DLR: 57 features) to predict early organ failure (Modified Marshall Score ≥2) in acute pancreatitis. Reported test-set discrimination (AUC 0.81), high overall NPV (97.2%), a small high-risk stratum with very high event rate, and a median early warning time of 3.5 hours before clinical organ failure.

Key findings

  • ORACLE discrimination for predicting organ failure: AUC 0.85 (training), 0.89 (validation), 0.81 (test).AUC 0.85 (train); 0.89 (val); 0.81 (test)
  • Overall negative predictive value for the entire cohort was 97.2%.NPV 97.2%
“This multicenter study included 2746 AP patients from two tertiary hospitals (2011–2024). Patients were split into training (n=1820), validation (n=456), and test cohorts (n=470).”
What this piece can’t prove
  • Abstract does not report calibration, confidence intervals for key metrics, decision-curve analysis, or prospective clinical impact evaluation.

3 further details could not be confirmed from the summary.

2secondary dataDevelop an nnMamba-based automated segmentation pipeline to delineate pancreatic/peripancreatic regions on multiphase CT as an enabling component for downstream radiomics risk prediction.Deep learning segmentation (nnMamba)Expand

In plain English

The abstract reports an nnMamba-based deep-learning segmentation model that automatically delineated pancreatic and peripancreatic regions on multiphase CT to provide regions of interest for downstream radiomics (SOF-DLR) used in organ-failure risk prediction. The abstract does not provide training, annotation, or performance details for the segmentation component.

Key findings

  • An nnMamba-based segmentation model was used to delineate pancreatic and peripancreatic regions on multiphase CT and served as the automated ROI-definition step for downstream radiomics-driven organ-failure prediction.
“An nnMamba-based segmentation model delineated pancreatic/peripancreatic regions on CT.”
What this piece can’t prove
  • The abstract lacks methodological detail specific to the segmentation component: training set size for the segmentation model, annotation procedures, model architecture beyond the name, hyperparameters, and evaluation metrics are not reported.
  • Unclear whether the segmentation model was validated independently of the downstream prediction model or whether segmentation errors were accounted for in prediction performance.

1 further detail could not be confirmed from the summary.

3human in vivoBenchmark ORACLE performance against existing CT severity scoring (M-CTSI) and clinical models to show superiority for organ failure prediction.Observational prediction-model comparative evaluationExpand

In plain English

The authors benchmarked the ORACLE deep-learning CT model against the Modified CT Severity Index (M-CTSI) and unspecified clinical models for predicting early organ failure in acute pancreatitis. On the same multicenter cohorts (n=2746; train 1820, val 456, test 470), ORACLE achieved higher AUCs (training 0.85, validation 0.89, test 0.81) than M-CTSI (AUC range 0.68–0.74) and clinical models (AUC range 0.67–0.71); DeLong’s test reported P < 0.001 for the comparison reported in the abstract.

Key findings

  • ORACLE achieved higher discriminative performance for predicting organ failure than M-CTSI and the reported clinical models on the same cohorts.ORACLE AUCs: 0.85 (training), 0.89 (validation), 0.81 (test) vs M-CTSI AUC 0.68–0.74 and clinical models AUC 0.67–0.71; DeLong’s P < 0.001 (abstract).
“...outperforming Modified CT Severity Index (M-CTSI) (AUC 0.68–0.74) and clinical models (AUC 0.67–0.71; DeLong’s P <0.001).”
What this piece can’t prove

4 further details could not be confirmed from the summary.

Finally, the search trail

Method layer

NewsLink found the paper. Tessa takes you deeper.

NewsLink checks the story. Tessa is where you inspect the paper, authors, evidence, and research context.

Open the paper in Tessa

Prediction of organ failure in acute pancreatitis via CT: A multicenter deep learning model with early clinical utility

Journal of Pancreatology · 2026

Source record (opens in a new tab)

Why this one

Near certain

NewsLink found the paper. Tessa is where you inspect it deeply.

Papers considered

The selected paper, plus nearby candidates.

Crossref, Europe PMC, PubMed · 36 candidate papers

Selected

Prediction of organ failure in acute pancreatitis via CT: A multicenter deep learning model with early clinical utility

Journal of Pancreatology · 2026 · Crossref

Source (opens in a new tab)
Candidate

Development of pancreatic cancer organoids from low cellularity Eus-guided sampling biopsy specimens

Pancreatology · 2026 · Crossref

Source (opens in a new tab)

And 30 more candidates considered.