{"id":"35be7986046b5d54cdb87a485a64dc48525cc321882fa4d27893079b26916354","title":"Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses","argument":"The authors report that instruction-based fine-tuning on public medical classification data improved open-weight LLMs’ coding of German tumor diagnoses. Their motivation is practical: smaller models may support privacy-preserving automation, but they often struggle with coding accuracy in German-language contexts. On coded diagnoses from a local tumor documentation system, ICD-10-GM accuracy increased from 1.4–24% to 41–58%; partial accuracy, counting three-character codes, increased from 31–74% to 73–83%. ICD-O-3 topography coding also improved, but remained considerably less accurate. Malformed code outputs fell to 0% for all models, and tumor-diagnosis recognition reached 99%. The authors also report that accuracy correlated positively with model size, while fine-tuning narrowed the gaps between small and large models. In their comparison, Qwen3’s reasoning mode generally performed worse than fine-tuning and was over 100 times slower. The results suggest that public catalogues can provide training material for this documentation task, but the reported evaluation is tied to the local test data and does not establish performance in other settings.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract says the evaluation used coded diagnoses from a local tumor documentation system; it does not establish whether the reported gains hold for diagnoses from other settings. The authors estimate an upper limit for ICD-10 coding performance based on data quality, so reported accuracy is framed against that estimated limit rather than an independently established ceiling.","speaker":{"name":"Lenz, Stefan, Rosario, Lakisha Ortiz, Vollmar, Georg, Ustjanzew, Arsenij, Alickovic, Fatma, Kindler, Thomas, Panholzer, Torsten","roleAtSource":"Article author"},"source":{"title":"Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2510.13624","publishedAt":"2025-10-15","dateBasis":"Date basis not provided."},"evidence":[{"text":"The study evaluates instruction-based fine-tuning of open-weight LLMs for coding German tumor diagnoses, using coded diagnoses from a local tumor documentation system as test data.","excerpt":"This study investigates whether instruction-based fine-tuning on public datasets improves the coding accuracy of open-weight LLMs for German tumor diagnosis texts. The evaluation uses coded diagnoses from the local tumor documentation system as test data.","url":"https://arxiv.org/abs/2510.13624","relationship":"Source-stated"},{"text":"The authors report improved ICD-10-GM accuracy and partial accuracy after fine-tuning.","excerpt":"ICD-10-GM accuracy rose from 1.4-24% to 41-58%, and partial accuracy from 31-74% to 73-83%.","url":"https://arxiv.org/abs/2510.13624","relationship":"Source-stated"},{"text":"The authors report that ICD-O-3 topography coding improved, but remained less accurate than ICD-10-GM coding.","excerpt":"The accuracy of ICD-O-3 topography coding also improved but started and remained considerably lower with an exact accuracy of 22-40% and a partial accuracy of 56-67% after fine-tuning.","url":"https://arxiv.org/abs/2510.13624","relationship":"Source-stated"},{"text":"The authors report that Qwen3 reasoning mode generally performed worse than fine-tuning and took over 100 times longer.","excerpt":"The reasoning mode in Qwen3 generally yielded a lower performance than fine-tuning and was over 100 times slower.","url":"https://arxiv.org/abs/2510.13624","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2510.13624","checkedAt":"2026-11-08T18:50:10.084Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Do the reported coding gains hold on tumor diagnoses beyond the local documentation system used as test data?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses","summary":"The authors report that instruction-based fine-tuning on public medical classification data improved open-weight LLMs’ coding of German tumor diagnoses. Their motivation is practical: smaller models may support privacy-preserving automation, but they often struggle with coding accuracy in German-language contexts. On coded diagnoses from a local tumor documentation system, ICD-10-GM accuracy increased from 1.4–24% to 41–58%; partial accuracy, counting three-character codes, increased from 31–74% to 73–83%. ICD-O-3 topography coding also improved, but remained considerably less accurate. Malformed code outputs fell to 0% for all models, and tumor-diagnosis recognition reached 99%. The authors also report that accuracy correlated positively with model size, while fine-tuning narrowed the gaps between small and large models. In their comparison, Qwen3’s reasoning mode generally performed worse than fine-tuning and was over 100 times slower. The results suggest that public catalogues can provide training material for this documentation task, but the reported evaluation is tied to the local test data and does not establish performance in other settings.","question":"Do the reported coding gains hold on tumor diagnoses beyond the local documentation system used as test data?","claims":[{"text":"The study evaluates instruction-based fine-tuning of open-weight LLMs for coding German tumor diagnoses, using coded diagnoses from a local tumor documentation system as test data.","excerpt":"This study investigates whether instruction-based fine-tuning on public datasets improves the coding accuracy of open-weight LLMs for German tumor diagnosis texts. The evaluation uses coded diagnoses from the local tumor documentation system as test data.","support":"source-stated"},{"text":"The authors report improved ICD-10-GM accuracy and partial accuracy after fine-tuning.","excerpt":"ICD-10-GM accuracy rose from 1.4-24% to 41-58%, and partial accuracy from 31-74% to 73-83%.","support":"source-stated"},{"text":"The authors report that ICD-O-3 topography coding improved, but remained less accurate than ICD-10-GM coding.","excerpt":"The accuracy of ICD-O-3 topography coding also improved but started and remained considerably lower with an exact accuracy of 22-40% and a partial accuracy of 56-67% after fine-tuning.","support":"source-stated"},{"text":"The authors report that Qwen3 reasoning mode generally performed worse than fine-tuning and took over 100 times longer.","excerpt":"The reasoning mode in Qwen3 generally yielded a lower performance than fine-tuning and was over 100 times slower.","support":"source-stated"}],"limitations":["The abstract says the evaluation used coded diagnoses from a local tumor documentation system; it does not establish whether the reported gains hold for diagnoses from other settings.","The authors estimate an upper limit for ICD-10 coding performance based on data quality, so reported accuracy is framed against that estimated limit rather than an independently established ceiling."],"mediaId":"0f9fedc78eee621d3886fc465b0c18d7892db13fc882919e5ef87ad6ecacff42","correctionReason":null,"id":"35be7986046b5d54cdb87a485a64dc48525cc321882fa4d27893079b26916354","publicationId":"6d248e78-4b39-46e4-89e9-1b7698cbbc48","version":1,"sourceUrl":"https://arxiv.org/abs/2510.13624","sourceTitle":"Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2025-10-15","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:01:00.921Z","createdAt":"2026-11-08T18:18:03.986Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"0f9fedc78eee621d3886fc465b0c18d7892db13fc882919e5ef87ad6ecacff42","kind":"article","url":"https://arxiv.org/abs/2510.13624","sourceUrl":"https://arxiv.org/abs/2510.13624","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-08T18:10:52.826Z","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/2510.13624","checkedAt":"2026-11-08T18:50:10.084Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"0f9fedc78eee621d3886fc465b0c18d7892db13fc882919e5ef87ad6ecacff42","confidence":0.88,"provider":"typesafe","model":"jev-1.13.0","inputHash":"2541addb86dd540353c08aff5e66d4d4cafc093cb18415fc30bcb4db9a84e77e","decidedAt":"2026-11-08T18:10:53.849Z"},"provenance":{"textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observationId":"1f8c8de260199e207c8e543783db03f920e049cb523febb410663ceaad60ecf5","retrievedAt":"2026-11-08T18:10:52.654Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"0450e4c6-4671-45fb-8599-63bc33b8bbcf"},"reviewer":{"model":"gpt-6-luna","runId":"6e870116-d5df-43cf-85d1-5e16a1dab4ef"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-08T18:18:03.986Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Lenz, Stefan","url":null,"kind":"unknown"},{"name":"Rosario, Lakisha Ortiz","url":null,"kind":"unknown"},{"name":"Vollmar, Georg","url":null,"kind":"unknown"},{"name":"Ustjanzew, Arsenij","url":null,"kind":"unknown"},{"name":"Alickovic, Fatma","url":null,"kind":"unknown"},{"name":"Kindler, Thomas","url":null,"kind":"unknown"},{"name":"Panholzer, Torsten","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Lenz, Stefan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lenz, Stefan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Rosario, Lakisha Ortiz","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Rosario, Lakisha Ortiz","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Vollmar, Georg","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Vollmar, Georg","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Ustjanzew, Arsenij","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Ustjanzew, Arsenij","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Alickovic, Fatma","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Alickovic, Fatma","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Kindler, Thomas","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Kindler, Thomas","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Panholzer, Torsten","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Panholzer, Torsten","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}}],"participationCoverage":{"total":7,"returned":7,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2510.13624","snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z","basis":"publisher-page-metadata","sourceLinks":47,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/35be7986046b5d54cdb87a485a64dc48525cc321882fa4d27893079b26916354/source"}},"canonicalUrl":"https://escapemeridian.com/readings/instruction-tuning-large-language-models-with-medical-35be7986","formats":{"html":"/readings/instruction-tuning-large-language-models-with-medical-35be7986","markdown":"/readings/instruction-tuning-large-language-models-with-medical-35be7986.md","json":"/readings/instruction-tuning-large-language-models-with-medical-35be7986.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-llm","name":"Large language model","explanation":"A language model with many learned numerical settings, used to process and generate language.","referenceUrl":"https://developers.google.com/machine-learning/glossary#large-language-model"},{"id":"term-fine-tuning","name":"Fine-tuning","explanation":"Further training of an existing model to adapt it to a particular task or use.","referenceUrl":"https://developers.google.com/machine-learning/glossary#fine-tuning"},{"id":"term-open-weights","name":"Open weights","explanation":"A model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.","referenceUrl":"https://opensource.org/ai/open-weights"}],"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":"35be7986046b5d54cdb87a485a64dc48525cc321882fa4d27893079b26916354","revision":"6d248e78-4b39-46e4-89e9-1b7698cbbc48","revisionKind":"publication-id","title":"Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses","url":"https://escapemeridian.com/readings/instruction-tuning-large-language-models-with-medical-35be7986"},"source":{"type":"source","id":"https://arxiv.org/abs/2510.13624","revision":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","revisionKind":"sha256-source-text","title":"Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses","url":"https://arxiv.org/abs/2510.13624"},"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":"35be7986046b5d54cdb87a485a64dc48525cc321882fa4d27893079b26916354","title":"Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses","argument":"The authors report that instruction-based fine-tuning on public medical classification data improved open-weight LLMs’ coding of German tumor diagnoses. Their motivation is practical: smaller models may support privacy-preserving automation, but they often struggle with coding accuracy in German-language contexts. On coded diagnoses from a local tumor documentation system, ICD-10-GM accuracy increased from 1.4–24% to 41–58%; partial accuracy, counting three-character codes, increased from 31–74% to 73–83%. ICD-O-3 topography coding also improved, but remained considerably less accurate. Malformed code outputs fell to 0% for all models, and tumor-diagnosis recognition reached 99%. The authors also report that accuracy correlated positively with model size, while fine-tuning narrowed the gaps between small and large models. In their comparison, Qwen3’s reasoning mode generally performed worse than fine-tuning and was over 100 times slower. The results suggest that public catalogues can provide training material for this documentation task, but the reported evaluation is tied to the local test data and does not establish performance in other settings.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract says the evaluation used coded diagnoses from a local tumor documentation system; it does not establish whether the reported gains hold for diagnoses from other settings. The authors estimate an upper limit for ICD-10 coding performance based on data quality, so reported accuracy is framed against that estimated limit rather than an independently established ceiling.","speaker":{"name":"Lenz, Stefan, Rosario, Lakisha Ortiz, Vollmar, Georg, Ustjanzew, Arsenij, Alickovic, Fatma, Kindler, Thomas, Panholzer, Torsten","roleAtSource":"Article author"},"source":{"title":"Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2510.13624","publishedAt":"2025-10-15","dateBasis":"Date basis not provided."},"evidence":[{"text":"The study evaluates instruction-based fine-tuning of open-weight LLMs for coding German tumor diagnoses, using coded diagnoses from a local tumor documentation system as test data.","excerpt":"This study investigates whether instruction-based fine-tuning on public datasets improves the coding accuracy of open-weight LLMs for German tumor diagnosis texts. The evaluation uses coded diagnoses from the local tumor documentation system as test data.","url":"https://arxiv.org/abs/2510.13624","relationship":"Source-stated"},{"text":"The authors report improved ICD-10-GM accuracy and partial accuracy after fine-tuning.","excerpt":"ICD-10-GM accuracy rose from 1.4-24% to 41-58%, and partial accuracy from 31-74% to 73-83%.","url":"https://arxiv.org/abs/2510.13624","relationship":"Source-stated"},{"text":"The authors report that ICD-O-3 topography coding improved, but remained less accurate than ICD-10-GM coding.","excerpt":"The accuracy of ICD-O-3 topography coding also improved but started and remained considerably lower with an exact accuracy of 22-40% and a partial accuracy of 56-67% after fine-tuning.","url":"https://arxiv.org/abs/2510.13624","relationship":"Source-stated"},{"text":"The authors report that Qwen3 reasoning mode generally performed worse than fine-tuning and took over 100 times longer.","excerpt":"The reasoning mode in Qwen3 generally yielded a lower performance than fine-tuning and was over 100 times slower.","url":"https://arxiv.org/abs/2510.13624","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2510.13624","checkedAt":"2026-11-08T18:50:10.084Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Do the reported coding gains hold on tumor diagnoses beyond the local documentation system used as test data?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses","summary":"The authors report that instruction-based fine-tuning on public medical classification data improved open-weight LLMs’ coding of German tumor diagnoses. Their motivation is practical: smaller models may support privacy-preserving automation, but they often struggle with coding accuracy in German-language contexts. On coded diagnoses from a local tumor documentation system, ICD-10-GM accuracy increased from 1.4–24% to 41–58%; partial accuracy, counting three-character codes, increased from 31–74% to 73–83%. ICD-O-3 topography coding also improved, but remained considerably less accurate. Malformed code outputs fell to 0% for all models, and tumor-diagnosis recognition reached 99%. The authors also report that accuracy correlated positively with model size, while fine-tuning narrowed the gaps between small and large models. In their comparison, Qwen3’s reasoning mode generally performed worse than fine-tuning and was over 100 times slower. The results suggest that public catalogues can provide training material for this documentation task, but the reported evaluation is tied to the local test data and does not establish performance in other settings.","question":"Do the reported coding gains hold on tumor diagnoses beyond the local documentation system used as test data?","claims":[{"text":"The study evaluates instruction-based fine-tuning of open-weight LLMs for coding German tumor diagnoses, using coded diagnoses from a local tumor documentation system as test data.","excerpt":"This study investigates whether instruction-based fine-tuning on public datasets improves the coding accuracy of open-weight LLMs for German tumor diagnosis texts. The evaluation uses coded diagnoses from the local tumor documentation system as test data.","support":"source-stated"},{"text":"The authors report improved ICD-10-GM accuracy and partial accuracy after fine-tuning.","excerpt":"ICD-10-GM accuracy rose from 1.4-24% to 41-58%, and partial accuracy from 31-74% to 73-83%.","support":"source-stated"},{"text":"The authors report that ICD-O-3 topography coding improved, but remained less accurate than ICD-10-GM coding.","excerpt":"The accuracy of ICD-O-3 topography coding also improved but started and remained considerably lower with an exact accuracy of 22-40% and a partial accuracy of 56-67% after fine-tuning.","support":"source-stated"},{"text":"The authors report that Qwen3 reasoning mode generally performed worse than fine-tuning and took over 100 times longer.","excerpt":"The reasoning mode in Qwen3 generally yielded a lower performance than fine-tuning and was over 100 times slower.","support":"source-stated"}],"limitations":["The abstract says the evaluation used coded diagnoses from a local tumor documentation system; it does not establish whether the reported gains hold for diagnoses from other settings.","The authors estimate an upper limit for ICD-10 coding performance based on data quality, so reported accuracy is framed against that estimated limit rather than an independently established ceiling."],"mediaId":"0f9fedc78eee621d3886fc465b0c18d7892db13fc882919e5ef87ad6ecacff42","correctionReason":null,"id":"35be7986046b5d54cdb87a485a64dc48525cc321882fa4d27893079b26916354","publicationId":"6d248e78-4b39-46e4-89e9-1b7698cbbc48","version":1,"sourceUrl":"https://arxiv.org/abs/2510.13624","sourceTitle":"Instruction-Tuning Large Language Models with Medical Classification Data Improves ICD Coding of German Tumor Diagnoses","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2025-10-15","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:01:00.921Z","createdAt":"2026-11-08T18:18:03.986Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"0f9fedc78eee621d3886fc465b0c18d7892db13fc882919e5ef87ad6ecacff42","kind":"article","url":"https://arxiv.org/abs/2510.13624","sourceUrl":"https://arxiv.org/abs/2510.13624","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-08T18:10:52.826Z","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/2510.13624","checkedAt":"2026-11-08T18:50:10.084Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"0f9fedc78eee621d3886fc465b0c18d7892db13fc882919e5ef87ad6ecacff42","confidence":0.88,"provider":"typesafe","model":"jev-1.13.0","inputHash":"2541addb86dd540353c08aff5e66d4d4cafc093cb18415fc30bcb4db9a84e77e","decidedAt":"2026-11-08T18:10:53.849Z"},"provenance":{"textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observationId":"1f8c8de260199e207c8e543783db03f920e049cb523febb410663ceaad60ecf5","retrievedAt":"2026-11-08T18:10:52.654Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"0450e4c6-4671-45fb-8599-63bc33b8bbcf"},"reviewer":{"model":"gpt-6-luna","runId":"6e870116-d5df-43cf-85d1-5e16a1dab4ef"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-08T18:18:03.986Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Lenz, Stefan","url":null,"kind":"unknown"},{"name":"Rosario, Lakisha Ortiz","url":null,"kind":"unknown"},{"name":"Vollmar, Georg","url":null,"kind":"unknown"},{"name":"Ustjanzew, Arsenij","url":null,"kind":"unknown"},{"name":"Alickovic, Fatma","url":null,"kind":"unknown"},{"name":"Kindler, Thomas","url":null,"kind":"unknown"},{"name":"Panholzer, Torsten","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Lenz, Stefan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lenz, Stefan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Rosario, Lakisha Ortiz","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Rosario, Lakisha Ortiz","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Vollmar, Georg","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Vollmar, Georg","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Ustjanzew, Arsenij","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Ustjanzew, Arsenij","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Alickovic, Fatma","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Alickovic, Fatma","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Kindler, Thomas","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Kindler, Thomas","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}},{"actor":{"name":"Panholzer, Torsten","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Panholzer, Torsten","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2510.13624","locator":"head/meta[name=author|citation_author]"},"assertionId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T18:50:10.084Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","sourceUrl":"https://arxiv.org/abs/2510.13624","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z"}}],"participationCoverage":{"total":7,"returned":7,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2510.13624","snapshotId":"95a7bcbea1229b0147bd7affdd34801f35db40ff428bf3ddc8488a8f9092dca3","contentSha256":"080f0265e6a92fd687c51c84a7769dbfc72d7d20669ae75e0899316fff8c7d00","textHash":"4c54e1fa11069a139cd588f3476cc9fecb8ab72eedb7a661349c44a5d473e153","observedAt":"2026-11-08T18:50:10.084Z","basis":"publisher-page-metadata","sourceLinks":47,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/35be7986046b5d54cdb87a485a64dc48525cc321882fa4d27893079b26916354/source"}}}],"audits":[{"publicationId":"6d248e78-4b39-46e4-89e9-1b7698cbbc48","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes the results and motivation to the paper's authors with phrases such as 'The authors report' and 'In their comparison.' It does not present the source's claims as established fact or as Escape Meridian's view.","excerpt":"Authors:Stefan Lenz, Lakisha Ortiz Rosario, Georg Vollmar, Arsenij Ustjanzew, Fatma Alickovic, Thomas Kindler, Torsten Panholzer"},{"dimension":"summary","status":"uncertain","explanation":"The figures are faithful, but a lay reader is not told what ICD coding is (assigning standard diagnosis codes, not programming). ICD-10-GM, ICD-O-3 topography, open-weight and instruction-tuning also need tooltips. The summary drops the authors' estimated ceiling of 60-79% exact and 81-94% partial, which changes how the 41-58% result reads. It also omits the 500,000+ question-answer pairs built from public catalogues and the released dataset and checkpoints. Smallest repair: add one plain sentence defining ICD coding and one stating the estimated ceiling.","excerpt":"the upper limit for ICD-10 coding performance was estimated at 60-79% for exact and 81-94% for partial (three-character codes only) derivation"},{"dimension":"date","status":"pass","explanation":"originalPublished is 2025-10-15, which matches the v1 submission date. The 5 Nov 2026 date is a later revision and the feed date is later still, so neither should be displayed.","excerpt":"Submitted on 15 Oct 2025 (v1), last revised 5 Nov 2026 (this version, v2)"},{"dimension":"support","status":"pass","explanation":"Each claim matches its abstract excerpt. The comparison of ICD-O-3 with ICD-10-GM follows from 'started and remained considerably lower.' The limitation about local test data and the question about generalization are framed as unknowns rather than as negative claims. The closing inference about public catalogues mirrors the authors' own stated potential.","excerpt":"Our findings highlight the potential of leveraging public catalogues to build instruction datasets that improve LLMs in medical documentation tasks."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified arXiv abstract page for this paper. It is the direct original source, it is relevant, and it is the only supplied candidate.","excerpt":null}],"createdAt":"2026-11-08T18:18:03.986Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}