{"id":"6a8ecf698725cf63d16224413a3de8bb7e0038eac63005620553373b023b2c01","title":"arXiv cs.AI: PACE-ECG tests ECG model transfer across healthcare systems","argument":"The authors of PACE-ECG argue that how an ECG model is pretrained can affect how well its representations transfer to clinical prognosis across healthcare systems. Their framework combines contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision. In tests predicting post-myocardial-infarction all-cause mortality and incident heart failure, frozen PACE-ECG representations outperformed DeepECG-SSL in MIMIC-IV and two HEEDB cohorts, while using a smaller encoder and less pretraining data. A matched ablation also found higher external-cohort AUROC with patient-temporal pairing than with generic self-supervision. The authors’ point is not just that the model performed better on these tasks: they argue that in-domain results alone cannot establish clinical generalisability, and that pretraining-objective design matters alongside model and data scale. These results concern the reported cohorts and prediction tasks; the abstract does not establish performance in clinical use.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports results for post-myocardial-infarction all-cause mortality and incident heart failure across MIMIC-IV and two HEEDB cohorts; it does not report sample sizes or confidence intervals. The abstract reports prediction results, but does not establish performance in clinical use.","speaker":{"name":"Cavarra, Riccardo, Lovatelli, Lupo, Ogbomo-Harmitt, Shaheim, Grzelak, Jakub, De Vecchi, Adelaide, King, Andrew, Aslanidi, Oleg","roleAtSource":"Article author"},"source":{"title":"PACE-ECG: Patient-Aware Contrastive ECG Pretraining; Leveraging Longitudinal Structure for Transferable Representations","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.12307","publishedAt":"2026-11-09","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors developed PACE-ECG, combining contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision.","excerpt":"We developed PACE-ECG, an ECG pretraining framework that combines contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision.","url":"https://arxiv.org/abs/2611.12307","relationship":"Source-stated"},{"text":"Across the reported mortality and heart failure tasks, frozen PACE-ECG representations outperformed DeepECG-SSL in-domain and in both external HEEDB cohorts.","excerpt":"Across both downstream tasks, the proposed model outperformed DeepECG-SSL in-domain (AUROC +1.9% for mortality and +4.0% for heart failure) and on external populations in HEEDB-Boston (+5.6% for mortality +10.3% for heart failure) and HEEDB-Atlanta (+6.9% for mortality +11.0% for heart failure)","url":"https://arxiv.org/abs/2611.12307","relationship":"Source-stated"},{"text":"The authors report that PACE-ECG used an encoder 41 times smaller and a fraction of the pretraining data compared with the model it outperformed.","excerpt":"while using an encoder 41 times smaller and a fraction of the pretraining data.","url":"https://arxiv.org/abs/2611.12307","relationship":"Source-stated"},{"text":"In a matched ablation, patient-temporal pairing improved external mortality and heart failure AUROC over generic self-supervision in both HEEDB cohorts.","excerpt":"A matched ablation showed that patient-temporal pairing improved external mortality and heart failure AUROC over generic self-supervision by 7.0% and 9.4% on HEEDB-Boston and 8.9% and 9.3% on HEEDB-Atlanta.","url":"https://arxiv.org/abs/2611.12307","relationship":"Source-stated"},{"text":"The authors argue that in-domain performance alone is insufficient to establish the clinical generalisability of ECG foundation representations.","excerpt":"in-domain performance alone is insufficient to establish the clinical generalisability of ECG foundation representations.","url":"https://arxiv.org/abs/2611.12307","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.12307","checkedAt":"2026-11-11T07:36:10.251Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Would PACE-ECG’s reported advantages over DeepECG-SSL and generic self-supervision hold in other healthcare systems and clinical prognosis tasks?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: PACE-ECG tests ECG model transfer across healthcare systems","summary":"The authors of PACE-ECG argue that how an ECG model is pretrained can affect how well its representations transfer to clinical prognosis across healthcare systems. Their framework combines contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision. In tests predicting post-myocardial-infarction all-cause mortality and incident heart failure, frozen PACE-ECG representations outperformed DeepECG-SSL in MIMIC-IV and two HEEDB cohorts, while using a smaller encoder and less pretraining data. A matched ablation also found higher external-cohort AUROC with patient-temporal pairing than with generic self-supervision. The authors’ point is not just that the model performed better on these tasks: they argue that in-domain results alone cannot establish clinical generalisability, and that pretraining-objective design matters alongside model and data scale. These results concern the reported cohorts and prediction tasks; the abstract does not establish performance in clinical use.","question":"Would PACE-ECG’s reported advantages over DeepECG-SSL and generic self-supervision hold in other healthcare systems and clinical prognosis tasks?","claims":[{"text":"The authors developed PACE-ECG, combining contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision.","excerpt":"We developed PACE-ECG, an ECG pretraining framework that combines contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision.","support":"source-stated"},{"text":"Across the reported mortality and heart failure tasks, frozen PACE-ECG representations outperformed DeepECG-SSL in-domain and in both external HEEDB cohorts.","excerpt":"Across both downstream tasks, the proposed model outperformed DeepECG-SSL in-domain (AUROC +1.9% for mortality and +4.0% for heart failure) and on external populations in HEEDB-Boston (+5.6% for mortality +10.3% for heart failure) and HEEDB-Atlanta (+6.9% for mortality +11.0% for heart failure)","support":"source-stated"},{"text":"The authors report that PACE-ECG used an encoder 41 times smaller and a fraction of the pretraining data compared with the model it outperformed.","excerpt":"while using an encoder 41 times smaller and a fraction of the pretraining data.","support":"source-stated"},{"text":"In a matched ablation, patient-temporal pairing improved external mortality and heart failure AUROC over generic self-supervision in both HEEDB cohorts.","excerpt":"A matched ablation showed that patient-temporal pairing improved external mortality and heart failure AUROC over generic self-supervision by 7.0% and 9.4% on HEEDB-Boston and 8.9% and 9.3% on HEEDB-Atlanta.","support":"source-stated"},{"text":"The authors argue that in-domain performance alone is insufficient to establish the clinical generalisability of ECG foundation representations.","excerpt":"in-domain performance alone is insufficient to establish the clinical generalisability of ECG foundation representations.","support":"source-stated"}],"limitations":["The abstract reports results for post-myocardial-infarction all-cause mortality and incident heart failure across MIMIC-IV and two HEEDB cohorts; it does not report sample sizes or confidence intervals.","The abstract reports prediction results, but does not establish performance in clinical use."],"mediaId":"6a93f2e0bd65639a8a13efbe3de403ed8eae859a0f09f16b87bc8ee2d81ce4c1","correctionReason":null,"id":"6a8ecf698725cf63d16224413a3de8bb7e0038eac63005620553373b023b2c01","publicationId":"a8fb2460-84e3-4c2b-8db3-54255fa5d024","version":1,"sourceUrl":"https://arxiv.org/abs/2611.12307","sourceTitle":"PACE-ECG: Patient-Aware Contrastive ECG Pretraining; Leveraging Longitudinal Structure for Transferable Representations","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-09","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-10T08:39:56.111Z","createdAt":"2026-11-10T14:33:54.351Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"6a93f2e0bd65639a8a13efbe3de403ed8eae859a0f09f16b87bc8ee2d81ce4c1","kind":"article","url":"https://arxiv.org/abs/2611.12307","sourceUrl":"https://arxiv.org/abs/2611.12307","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-10T14:33:07.862Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.12307","checkedAt":"2026-11-11T07:36:10.251Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"6a93f2e0bd65639a8a13efbe3de403ed8eae859a0f09f16b87bc8ee2d81ce4c1","confidence":0.9,"provider":"typesafe","model":"jev-1.13.0","inputHash":"b92c1eaabdb9019764062531cc77733946363dd7ee975055a07d43e250b20af2","decidedAt":"2026-11-10T14:33:08.472Z"},"provenance":{"textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observationId":"c5a7fbc8a8e8d04d7b3e878636e6eddc7e66c7351903a6b56465bb6c99445373","retrievedAt":"2026-11-10T14:33:07.799Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"87a0671c-5573-4675-8bdf-d8802706180b"},"reviewer":{"model":"gpt-6-luna","runId":"80d4df45-c4a9-4029-8c08-fdcd323b209e"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-10T14:33:54.351Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Cavarra, Riccardo","url":null,"kind":"unknown"},{"name":"Lovatelli, Lupo","url":null,"kind":"unknown"},{"name":"Ogbomo-Harmitt, Shaheim","url":null,"kind":"unknown"},{"name":"Grzelak, Jakub","url":null,"kind":"unknown"},{"name":"De Vecchi, Adelaide","url":null,"kind":"unknown"},{"name":"King, Andrew","url":null,"kind":"unknown"},{"name":"Aslanidi, Oleg","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Cavarra, Riccardo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Cavarra, Riccardo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"Lovatelli, Lupo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lovatelli, Lupo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"Ogbomo-Harmitt, Shaheim","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Ogbomo-Harmitt, Shaheim","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"Grzelak, Jakub","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Grzelak, Jakub","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"De Vecchi, Adelaide","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"De Vecchi, Adelaide","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"King, Andrew","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"King, Andrew","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"Aslanidi, Oleg","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Aslanidi, Oleg","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}}],"participationCoverage":{"total":7,"returned":7,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.12307","snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/6a8ecf698725cf63d16224413a3de8bb7e0038eac63005620553373b023b2c01/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-pace-ecg-tests-ecg-model-transfer-across-healthcare-6a8ecf69","formats":{"html":"/readings/arxiv-cs-ai-pace-ecg-tests-ecg-model-transfer-across-healthcare-6a8ecf69","markdown":"/readings/arxiv-cs-ai-pace-ecg-tests-ecg-model-transfer-across-healthcare-6a8ecf69.md","json":"/readings/arxiv-cs-ai-pace-ecg-tests-ecg-model-transfer-across-healthcare-6a8ecf69.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-pre-training","name":"Pre-training","explanation":"A model's initial training on a large dataset, before further adaptation for particular uses.","referenceUrl":"https://developers.google.com/machine-learning/glossary#pre-training"},{"id":"term-confidence-interval","name":"Confidence interval","explanation":"A range of values consistent with the data at a stated level of confidence.","referenceUrl":"https://en.wikipedia.org/wiki/Confidence_interval"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"6a8ecf698725cf63d16224413a3de8bb7e0038eac63005620553373b023b2c01","revision":"a8fb2460-84e3-4c2b-8db3-54255fa5d024","revisionKind":"publication-id","title":"arXiv cs.AI: PACE-ECG tests ECG model transfer across healthcare systems","url":"https://escapemeridian.com/readings/arxiv-cs-ai-pace-ecg-tests-ecg-model-transfer-across-healthcare-6a8ecf69"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.12307","revision":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","revisionKind":"sha256-source-text","title":"PACE-ECG: Patient-Aware Contrastive ECG Pretraining; Leveraging Longitudinal Structure for Transferable Representations","url":"https://arxiv.org/abs/2611.12307"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"6a8ecf698725cf63d16224413a3de8bb7e0038eac63005620553373b023b2c01","title":"arXiv cs.AI: PACE-ECG tests ECG model transfer across healthcare systems","argument":"The authors of PACE-ECG argue that how an ECG model is pretrained can affect how well its representations transfer to clinical prognosis across healthcare systems. Their framework combines contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision. In tests predicting post-myocardial-infarction all-cause mortality and incident heart failure, frozen PACE-ECG representations outperformed DeepECG-SSL in MIMIC-IV and two HEEDB cohorts, while using a smaller encoder and less pretraining data. A matched ablation also found higher external-cohort AUROC with patient-temporal pairing than with generic self-supervision. The authors’ point is not just that the model performed better on these tasks: they argue that in-domain results alone cannot establish clinical generalisability, and that pretraining-objective design matters alongside model and data scale. These results concern the reported cohorts and prediction tasks; the abstract does not establish performance in clinical use.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports results for post-myocardial-infarction all-cause mortality and incident heart failure across MIMIC-IV and two HEEDB cohorts; it does not report sample sizes or confidence intervals. The abstract reports prediction results, but does not establish performance in clinical use.","speaker":{"name":"Cavarra, Riccardo, Lovatelli, Lupo, Ogbomo-Harmitt, Shaheim, Grzelak, Jakub, De Vecchi, Adelaide, King, Andrew, Aslanidi, Oleg","roleAtSource":"Article author"},"source":{"title":"PACE-ECG: Patient-Aware Contrastive ECG Pretraining; Leveraging Longitudinal Structure for Transferable Representations","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.12307","publishedAt":"2026-11-09","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors developed PACE-ECG, combining contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision.","excerpt":"We developed PACE-ECG, an ECG pretraining framework that combines contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision.","url":"https://arxiv.org/abs/2611.12307","relationship":"Source-stated"},{"text":"Across the reported mortality and heart failure tasks, frozen PACE-ECG representations outperformed DeepECG-SSL in-domain and in both external HEEDB cohorts.","excerpt":"Across both downstream tasks, the proposed model outperformed DeepECG-SSL in-domain (AUROC +1.9% for mortality and +4.0% for heart failure) and on external populations in HEEDB-Boston (+5.6% for mortality +10.3% for heart failure) and HEEDB-Atlanta (+6.9% for mortality +11.0% for heart failure)","url":"https://arxiv.org/abs/2611.12307","relationship":"Source-stated"},{"text":"The authors report that PACE-ECG used an encoder 41 times smaller and a fraction of the pretraining data compared with the model it outperformed.","excerpt":"while using an encoder 41 times smaller and a fraction of the pretraining data.","url":"https://arxiv.org/abs/2611.12307","relationship":"Source-stated"},{"text":"In a matched ablation, patient-temporal pairing improved external mortality and heart failure AUROC over generic self-supervision in both HEEDB cohorts.","excerpt":"A matched ablation showed that patient-temporal pairing improved external mortality and heart failure AUROC over generic self-supervision by 7.0% and 9.4% on HEEDB-Boston and 8.9% and 9.3% on HEEDB-Atlanta.","url":"https://arxiv.org/abs/2611.12307","relationship":"Source-stated"},{"text":"The authors argue that in-domain performance alone is insufficient to establish the clinical generalisability of ECG foundation representations.","excerpt":"in-domain performance alone is insufficient to establish the clinical generalisability of ECG foundation representations.","url":"https://arxiv.org/abs/2611.12307","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.12307","checkedAt":"2026-11-11T07:36:10.251Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Would PACE-ECG’s reported advantages over DeepECG-SSL and generic self-supervision hold in other healthcare systems and clinical prognosis tasks?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: PACE-ECG tests ECG model transfer across healthcare systems","summary":"The authors of PACE-ECG argue that how an ECG model is pretrained can affect how well its representations transfer to clinical prognosis across healthcare systems. Their framework combines contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision. In tests predicting post-myocardial-infarction all-cause mortality and incident heart failure, frozen PACE-ECG representations outperformed DeepECG-SSL in MIMIC-IV and two HEEDB cohorts, while using a smaller encoder and less pretraining data. A matched ablation also found higher external-cohort AUROC with patient-temporal pairing than with generic self-supervision. The authors’ point is not just that the model performed better on these tasks: they argue that in-domain results alone cannot establish clinical generalisability, and that pretraining-objective design matters alongside model and data scale. These results concern the reported cohorts and prediction tasks; the abstract does not establish performance in clinical use.","question":"Would PACE-ECG’s reported advantages over DeepECG-SSL and generic self-supervision hold in other healthcare systems and clinical prognosis tasks?","claims":[{"text":"The authors developed PACE-ECG, combining contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision.","excerpt":"We developed PACE-ECG, an ECG pretraining framework that combines contrastive learning across temporally neighbouring recordings from the same patient with clinically guided multitask supervision.","support":"source-stated"},{"text":"Across the reported mortality and heart failure tasks, frozen PACE-ECG representations outperformed DeepECG-SSL in-domain and in both external HEEDB cohorts.","excerpt":"Across both downstream tasks, the proposed model outperformed DeepECG-SSL in-domain (AUROC +1.9% for mortality and +4.0% for heart failure) and on external populations in HEEDB-Boston (+5.6% for mortality +10.3% for heart failure) and HEEDB-Atlanta (+6.9% for mortality +11.0% for heart failure)","support":"source-stated"},{"text":"The authors report that PACE-ECG used an encoder 41 times smaller and a fraction of the pretraining data compared with the model it outperformed.","excerpt":"while using an encoder 41 times smaller and a fraction of the pretraining data.","support":"source-stated"},{"text":"In a matched ablation, patient-temporal pairing improved external mortality and heart failure AUROC over generic self-supervision in both HEEDB cohorts.","excerpt":"A matched ablation showed that patient-temporal pairing improved external mortality and heart failure AUROC over generic self-supervision by 7.0% and 9.4% on HEEDB-Boston and 8.9% and 9.3% on HEEDB-Atlanta.","support":"source-stated"},{"text":"The authors argue that in-domain performance alone is insufficient to establish the clinical generalisability of ECG foundation representations.","excerpt":"in-domain performance alone is insufficient to establish the clinical generalisability of ECG foundation representations.","support":"source-stated"}],"limitations":["The abstract reports results for post-myocardial-infarction all-cause mortality and incident heart failure across MIMIC-IV and two HEEDB cohorts; it does not report sample sizes or confidence intervals.","The abstract reports prediction results, but does not establish performance in clinical use."],"mediaId":"6a93f2e0bd65639a8a13efbe3de403ed8eae859a0f09f16b87bc8ee2d81ce4c1","correctionReason":null,"id":"6a8ecf698725cf63d16224413a3de8bb7e0038eac63005620553373b023b2c01","publicationId":"a8fb2460-84e3-4c2b-8db3-54255fa5d024","version":1,"sourceUrl":"https://arxiv.org/abs/2611.12307","sourceTitle":"PACE-ECG: Patient-Aware Contrastive ECG Pretraining; Leveraging Longitudinal Structure for Transferable Representations","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-09","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-10T08:39:56.111Z","createdAt":"2026-11-10T14:33:54.351Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"6a93f2e0bd65639a8a13efbe3de403ed8eae859a0f09f16b87bc8ee2d81ce4c1","kind":"article","url":"https://arxiv.org/abs/2611.12307","sourceUrl":"https://arxiv.org/abs/2611.12307","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-10T14:33:07.862Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.12307","checkedAt":"2026-11-11T07:36:10.251Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"6a93f2e0bd65639a8a13efbe3de403ed8eae859a0f09f16b87bc8ee2d81ce4c1","confidence":0.9,"provider":"typesafe","model":"jev-1.13.0","inputHash":"b92c1eaabdb9019764062531cc77733946363dd7ee975055a07d43e250b20af2","decidedAt":"2026-11-10T14:33:08.472Z"},"provenance":{"textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observationId":"c5a7fbc8a8e8d04d7b3e878636e6eddc7e66c7351903a6b56465bb6c99445373","retrievedAt":"2026-11-10T14:33:07.799Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"87a0671c-5573-4675-8bdf-d8802706180b"},"reviewer":{"model":"gpt-6-luna","runId":"80d4df45-c4a9-4029-8c08-fdcd323b209e"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-10T14:33:54.351Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Cavarra, Riccardo","url":null,"kind":"unknown"},{"name":"Lovatelli, Lupo","url":null,"kind":"unknown"},{"name":"Ogbomo-Harmitt, Shaheim","url":null,"kind":"unknown"},{"name":"Grzelak, Jakub","url":null,"kind":"unknown"},{"name":"De Vecchi, Adelaide","url":null,"kind":"unknown"},{"name":"King, Andrew","url":null,"kind":"unknown"},{"name":"Aslanidi, Oleg","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Cavarra, Riccardo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Cavarra, Riccardo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"Lovatelli, Lupo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lovatelli, Lupo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"Ogbomo-Harmitt, Shaheim","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Ogbomo-Harmitt, Shaheim","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"Grzelak, Jakub","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Grzelak, Jakub","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"De Vecchi, Adelaide","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"De Vecchi, Adelaide","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"King, Andrew","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"King, Andrew","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}},{"actor":{"name":"Aslanidi, Oleg","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Aslanidi, Oleg","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12307","locator":"head/meta[name=author|citation_author]"},"assertionId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T07:36:10.251Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","sourceUrl":"https://arxiv.org/abs/2611.12307","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z"}}],"participationCoverage":{"total":7,"returned":7,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.12307","snapshotId":"95ff9779b40727b6f9644b8562257a497e1d64860ad316db42ba617e214c5334","contentSha256":"f35df3cd952e376304392c61abd81f657f05053ed74c9161ab9bad01b1858057","textHash":"9bdb200a492a1619f44a246d74d49a98103740d6d567301273935fa326c88f4d","observedAt":"2026-11-11T07:36:10.251Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/6a8ecf698725cf63d16224413a3de8bb7e0038eac63005620553373b023b2c01/source"}}}],"audits":[{"publicationId":"a8fb2460-84e3-4c2b-8db3-54255fa5d024","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The draft attributes the framework, results and conclusions to the paper's authors. The title names arXiv cs.AI as the venue. Comparative and generalisability claims stay framed as the authors' reported findings and argument, not as Escape Meridian's view.","excerpt":"Authors:Riccardo Cavarra, Lupo Lovatelli, Shaheim Ogbomo-Harmitt, Jakub Grzelak, Adelaide De Vecchi, Andrew King, Oleg Aslanidi"},{"dimension":"summary","status":"uncertain","explanation":"The summary faithfully conveys the authors' argument, and the title and question fit. A general reader, however, meets several central terms with no plain-language explanation: ECG, contrastive learning, frozen representations, AUROC, matched ablation, self-supervision, post-myocardial-infarction (after a heart attack), and the MIMIC-IV and HEEDB cohorts. Without them the main point is hard to follow: training on a patient's own nearby recordings helped predictions carry over to other hospitals. Smallest repair: keep the terms exactly, but add or confirm term-library tooltips for them. Also add one plain clause saying what the predictions are (death after a heart attack, new heart failure).","excerpt":null},{"dimension":"date","status":"pass","explanation":"originalPublished is 2026-11-09. This matches the arXiv submission and v1 date. The feed date of 2026-11-10 is correctly not used.","excerpt":"[Submitted on 9 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"All claims match the abstract. They cover the framework design, the in-domain and external gains over DeepECG-SSL, the encoder 41 times smaller with a fraction of the pretraining data, the ablation gains, and the authors' conclusion. The limitations are fair: the abstract gives no sample sizes or confidence intervals and no clinical-use evidence. The question about other systems and tasks follows from the evidence without presupposing an answer. No invented benefits or costs.","excerpt":"These findings indicate that pretraining-objective design greatly influences downstream transfer alongside model and data scale"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the arXiv abstract page, the original source. No better source media was found.","excerpt":null}],"createdAt":"2026-11-10T14:33:54.351Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}