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

Source study found

Story checked

AI tools help predict immunotherapy outcomes in lung cancer patients (opens in a new tab)

news-medical.net · 2026-09-16

Short answerEvidenceSource

Short answer

Not supported

Not supported.

3 claims go further than the study. 3 other points were not covered by the paper.

  • 3 overstated
  • 3 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

Not supported

Three of six claims overstate the study. Three claims the study doesn't address.

  • 3 overstated
  • 3 not covered
Open claim evidence
3
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 6OverstatedAI tools can help physicians predict treatment and survival outcomes in patients with advanced non-small cell lung cancer treated with immunotherapy, according to a new study published in Nature Medicine.View evidenceHide evidence

Why this verdict

The abstract-level profile supports that I3LUNG developed AI models to predict immunotherapy-related outcomes and that physicians improved predictions in a usability study using an XAI tool. However, the headline frames this as an unhedged causal claim that AI tools can help physicians predict treatment and survival outcomes; the abstract profile does not specify survival outcomes in detail and does not establish patient-level clinical benefit. The headline therefore outruns what the abstract-level evidence can verify.

1

Study evidence

ML and DL models using clinical and blood (CB-only) inputs achieved up to AUC = 0.77 in the independent TEST set.AUC up to 0.77

“I3LUNG ( NCT05537922 ) is currently the largest international, real-world, multimodal, artificial intelligence (AI)-based study, enrolling 2,396 patients.”
2

Study evidence

Lung expert and nonexpert physicians improved their outcome predictions when assisted by the explainable AI (XAI) ML CB-only decision-support tool.

“The clinical usability study showed that lung expert and nonexpert physicians improved their prediction with the explainable AI (XAI) ML CB-only based tool.”
Claim 2 of 6OverstatedThe study reported findings from the I3LUNG project, which the article describes as a large, international trial aimed at improving treatment of metastatic NSCLC by developing AI-based predictive models to determine the best therapeutic approach for each patient.View evidenceHide evidence

Why this verdict

The paper profile supports a large international real-world I3LUNG study developing and validating AI predictive models for NSCLC immunotherapy-related outcomes. But the story’s wording as a large international 'trial' aimed at determining the best therapeutic approach for each patient overstates the abstract-level evidence, which describes observational real-world model development/validation rather than proof of individualized treatment selection.

1

Study evidence

ML and DL models using clinical and blood (CB-only) inputs achieved up to AUC = 0.77 in the independent TEST set.AUC up to 0.77

“I3LUNG ( NCT05537922 ) is currently the largest international, real-world, multimodal, artificial intelligence (AI)-based study, enrolling 2,396 patients.”
Claim 3 of 6OverstatedThe article says the AI models consistently outperformed standard clinical biomarkers; the clinical-and-blood model achieved an AUC of 0.77, while the model that also incorporated imaging and digital pathology achieved an AUC of 0.88.View evidenceHide evidence

As statedAUC 0.77 and 0.88

Why this verdict

The abstract-level profile supports CB-only model AUC up to 0.77 in the independent TEST set and significant superiority over PD-L1/ECOG/NLR/LDH/LIPI in TEST. But the claim’s 'consistently outperformed' framing underplays the external-validation performance drop and the uncertainty around multimodal incremental benefit. The AUC 0.88 for the model incorporating imaging/pathology is not provided in the supplied abstract profile, and the abstract says multimodal gains did not consistently translate to TEST and EXVAL.

1

Study evidence

ML and DL models using clinical and blood (CB-only) inputs achieved up to AUC = 0.77 in the independent TEST set.AUC up to 0.77

“I3LUNG ( NCT05537922 ) is currently the largest international, real-world, multimodal, artificial intelligence (AI)-based study, enrolling 2,396 patients.”
2

Study evidence

CB-only ML and DL models achieved AUC up to 0.77 in the independent TEST set.AUC up to 0.77 (TEST)

“Although multimodal integration with MLEF (CB+CT+DP) was associated with higher performance, its incremental benefit remains uncertain, not translated in TEST and EXVAL.”
Claim 4 of 6Not coveredResearchers enrolled 2,396 patients with advanced NSCLC treated with immunotherapy across six centers in Italy, Germany, Greece, Israel, Spain and the United States, and built two families of AI models using clinical, imaging, pathology, blood, and genomic data.View evidenceHide evidence

As stated2,396 patients; six centers; two model families

Why this verdict

The abstract-level profile supports the 2,396-patient I3LUNG cohort and integration of clinical, blood, CT imaging, digital pathology and genomics into ML/DL fusion models. However, the exact six centers and listed countries are not available in the supplied abstract-level profile, and some details such as the precise patient-stage framing are not fully verifiable at this depth.

1

Study evidence

ML and DL models using clinical and blood (CB-only) inputs achieved up to AUC = 0.77 in the independent TEST set.AUC up to 0.77

“I3LUNG ( NCT05537922 ) is currently the largest international, real-world, multimodal, artificial intelligence (AI)-based study, enrolling 2,396 patients.”
2

Study evidence

CB-only ML and DL models achieved AUC up to 0.77 in the independent TEST set.AUC up to 0.77 (TEST)

“Although multimodal integration with MLEF (CB+CT+DP) was associated with higher performance, its incremental benefit remains uncertain, not translated in TEST and EXVAL.”
Claim 5 of 6Not coveredIn a test of human-AI collaboration, 20 physicians reviewing 100 real patient cases improved sensitivity for identifying responders from an AUC of 0.72 to 0.87 when given AI support, with the biggest gains among non-lung-cancer specialists and better inter-physician agreement.View evidenceHide evidence

As statedAUC 0.72 to 0.87; 20 physicians; 100 cases

Why this verdict

The abstract-level profile supports the qualitative finding that lung expert and nonexpert physicians improved predictions using an explainable AI CB-only decision-support tool. It does not provide the claimed details: 20 physicians, 100 cases, AUC change from 0.72 to 0.87, sensitivity for responders, largest gains among non-lung-cancer specialists, or improved inter-physician agreement. Those specifics cannot be verified from the supplied abstract-level evidence.

1

Study evidence

Lung expert and nonexpert physicians improved their outcome predictions when assisted by the explainable AI (XAI) ML CB-only decision-support tool.

“The clinical usability study showed that lung expert and nonexpert physicians improved their prediction with the explainable AI (XAI) ML CB-only based tool.”
Claim 6 of 6Not coveredThe article says the current study reported the project’s retrospective phase, and that I3LUNG is now prospectively enrolling more than 2,000 patients across the same six international centers.View evidenceHide evidence

As statedmore than 2,000 patients

Why this verdict

The abstract-level profile supports that prospective validation of the decision-support system is ongoing in more than 2,000 patients. It does not verify the full framing that the current report is specifically the project’s retrospective phase or that prospective enrollment is across the same six international centers.

1

Study evidence

ML and DL models using clinical and blood (CB-only) inputs achieved up to AUC = 0.77 in the independent TEST set.AUC up to 0.77

“I3LUNG ( NCT05537922 ) is currently the largest international, real-world, multimodal, artificial intelligence (AI)-based study, enrolling 2,396 patients.”
2

Study evidence

Lung expert and nonexpert physicians improved their outcome predictions when assisted by the explainable AI (XAI) ML CB-only decision-support tool.

“The clinical usability study showed that lung expert and nonexpert physicians improved their prediction with the explainable AI (XAI) ML CB-only based tool.”
Then look for missing context

Context layer

What the story left out

Important study details the story did not include.

  • External validation performance declined, with EXVAL AUCs ranging from 0.55 to 0.72, indicating possible population heterogeneity and limited generalizability.

    This is an interpretation-changing limitation in the paper profile. The story caveats mention current biomarker limitations and prospective enrollment but do not acknowledge the abstract-reported external-validation performance drop.

    From observational cohort model development and validation; comparative multimodal fusion evaluation

  • Multimodal integration with CB+CT+digital pathology was associated with higher performance in development, but its incremental benefit was uncertain and did not translate consistently to TEST or EXVAL.

    The story emphasizes multimodal model performance, including an AUC not available in the abstract profile, but does not reflect the paper’s key caveat that multimodal incremental benefit was uncertain and not consistently replicated in held-out/external validation.

    From comparative multimodal fusion evaluation

  • The paper profile describes observational real-world model validation and usability testing, not evidence that AI use improves patient survival, changes treatment outcomes, or determines the best therapy for individual patients.

    Although the story mentions the retrospective phase, its headline and lead use stronger clinical decision-support and individualized-treatment language than the abstract-level evidence supports.

    From observational cohort model development and validation; clinical usability study (assisted vs unassisted physician predic

4 things the story did carry across
  • I3LUNG is a large international real-world cohort/model-development and validation study enrolling 2,396 NSCLC patients and integrating clinical, blood, CT, digital pathology and genomics data.
  • CB-only ML/DL models achieved AUC up to 0.77 in the independent TEST set and significantly outperformed PD-L1, ECOG PS, NLR, LDH and LIPI in TEST.
  • The clinical usability study found that lung expert and nonexpert physicians improved their predictions when using an explainable AI CB-only decision-support tool.
  • Prospective validation of the AI decision-support system is ongoing in more than 2,000 patients.
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

secondary data

1Lead resultsecondary dataDevelop and validate real-world multimodal AI models (clinical+blood, CT imaging, digital pathology, genomics) to predict immunotherapy-related outcomes in NSCLC, including internal testing and external validation, and compare against standard biomarkers/scores (PD-L1, ECOG PS, NLR, LDH, LIPI).observational cohort model development and validationExpand

In plain English

I3LUNG (NCT05537922) is a large (n=2,396) international real-world cohort study that developed and evaluated multimodal AI models (clinical+blood, CT, digital pathology, genomics) to predict immunotherapy-related outcomes in NSCLC. The authors trained machine-learning (ML) and deep-learning (DL) models with early- and intermediate-fusion approaches, reported discrimination (AUC) in an independent internal TEST set and in external validation (EXVAL), benchmarked model performance against PD-L1 and clinical/hematologic prognostic scores, and performed a clinical usability assessment of an explainable AI decision-support tool.

Key findings

  • ML and DL models using clinical and blood (CB-only) inputs achieved up to AUC = 0.77 in the independent TEST set.AUC up to 0.77
  • Model discrimination decreased in external validation cohorts (EXVAL), with reported AUC range 0.55–0.72.AUC range 0.55–0.72
“I3LUNG ( NCT05537922 ) is currently the largest international, real-world, multimodal, artificial intelligence (AI)-based study, enrolling 2,396 patients.”
What this piece can’t prove
  • External validation performance declined (AUC 0.55–0.72), indicating possible population heterogeneity and limited generalizability.
  • Incremental benefit of multimodal fusion is reported as uncertain and not consistently observed in TEST and EXVAL.

2 further details could not be confirmed from the summary.

2human in vivoEvaluate clinical usability of an explainable AI (XAI) decision-support tool by assessing whether expert and nonexpert physicians improve their outcome predictions when using the tool.clinical usability study (assisted vs unassisted physician predictions)Expand

In plain English

A clinical usability study reported that both lung expert and nonexpert physicians improved their outcome predictions when using an explainable AI (XAI) decision-support tool based on a machine-learning model trained on clinical and blood (CB) data.

Key findings

  • Lung expert and nonexpert physicians improved their outcome predictions when assisted by the explainable AI (XAI) ML CB-only decision-support tool.
“The clinical usability study showed that lung expert and nonexpert physicians improved their prediction with the explainable AI (XAI) ML CB-only based tool.”
What this piece can’t prove
  • No methodological details provided on study design (randomization, blinding, crossover, or parallel assignment).
  • Unclear how clinician participants were recruited or characterized beyond 'expert' vs 'nonexpert'.
  • No description of the prediction task, evaluation metrics, or whether training/practice with the tool occurred before measurement.

1 further detail could not be confirmed from the summary.

3secondary dataAssess the incremental benefit (or lack thereof) of multimodal integration (early fusion and intermediate fusion) versus CB-only models, including whether gains translate to independent test and external validation cohorts.comparative multimodal fusion evaluationExpand

In plain English

The abstract reports that integrating clinical+blood (CB) data with CT and digital pathology (DP) using an early-fusion machine-learning approach (MLEF) was associated with higher performance than CB-only models, but that this incremental benefit was uncertain and did not translate to the independent internal TEST set or to external validation (EXVAL). CB-only ML and DL models achieved AUC up to 0.77 in the TEST set, with lower AUCs in EXVAL (range 0.55–0.72). Details on effect sizes for the multimodal incremental gains, comparison metrics between MLEF and DLIF, and evaluation stratified by outcome are not provided in the abstract.

Key findings

  • CB-only ML and DL models achieved AUC up to 0.77 in the independent TEST set.AUC up to 0.77 (TEST)
  • Performance decreased in external validation (EXVAL), with AUCs reported in the range 0.55–0.72.AUC range 0.55–0.72 (EXVAL)
“Although multimodal integration with MLEF (CB+CT+DP) was associated with higher performance, its incremental benefit remains uncertain, not translated in TEST and EXVAL.”
What this piece can’t prove
  • The abstract indicates non-translation of multimodal gains to TEST and EXVAL but does not specify whether this reflects overfitting, cohort heterogeneity, or other factors; causal reasons are not detailed.

2 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.

Papers considered

The selected paper, plus nearby candidates.

PubMed, Crossref, Europe PMC · 39 candidate papers

Candidate

NATURE AND CULTURE: AN ECOLINGUISTICS’ ANALYSIS OF IS A RIVER ALIVE BY ROBERT MACFARLANE

Scholarly Journal · 2026 · Crossref

Source (opens in a new tab)

And 33 more candidates considered.