@inproceedings{rohr-etal-2026-deepicd,
title = "{D}eep{ICD}-R1: Medical Reasoning through Hierarchical Rewards and Unsupervised Distillation",
author = {R{\"o}hr, Tom and
Steffek, Thomas Maximilian Josef and
Teucher, Roman and
Bressem, Keno and
Figueroa, Alexei and
Grundmann, Paul and
Troeger, Peter and
Gers, Felix Alexander and
L{\"o}ser, Alexander},
editor = "Piperidis, Stelios and
Bel, N{\'u}ria and
van den Heuvel, Henk and
Ide, Nancy and
Krek, Simon and
Toral, Antonio",
booktitle = "Proceedings of the Fifteenth Language Resources and Evaluation Conference",
month = may,
year = "2026",
address = "Palma de Mallorca, Spain",
publisher = "ELRA Language Resource Association",
url = "https://aclanthology.org/2026.lrec-1.843/",
doi = "10.63317/5ntn3gnmv9cy",
pages = "10764--10775",
abstract = "Large language models (LLMs) show strong reasoning abilities, but full retraining for the medical domain is often infeasible because of lacking data or compute resources. We present DeepICD-R1, a framework for efficient medical reasoning fine-tuning that unites hierarchical rewards with distilled supervision. We reformulate ICD-10-CM prediction as a reinforcement learning problem and design a hierarchical outcome-based reward that reflects the ICD code structure across chapter, category, and full-code levels. In parallel, we publish a large-scale distilled dataset of over 90k reasoning traces derived from MIMIC-IV admission notes, integrating clinical validation and official coding guidelines. Fine-tuning smaller instruction-tuned LLMs with this data and GRPO reinforcement yields consistent gains in diagnostic accuracy and reasoning coherence. Extensive ablations confirm that hierarchical supervision and verifiable outcome rewards enable competitive, domain-specialized reasoning models without additional pretraining, providing a reproducible foundation for clinical NLP research. Keywords: Clinical NLP, Large Reasoning Model, GRPO, Supervised Fine-Tuning"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="rohr-etal-2026-deepicd">
<titleInfo>
<title>DeepICD-R1: Medical Reasoning through Hierarchical Rewards and Unsupervised Distillation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tom</namePart>
<namePart type="family">Röhr</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thomas</namePart>
<namePart type="given">Maximilian</namePart>
<namePart type="given">Josef</namePart>
<namePart type="family">Steffek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Roman</namePart>
<namePart type="family">Teucher</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Keno</namePart>
<namePart type="family">Bressem</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alexei</namePart>
<namePart type="family">Figueroa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Paul</namePart>
<namePart type="family">Grundmann</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peter</namePart>
<namePart type="family">Troeger</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Felix</namePart>
<namePart type="given">Alexander</namePart>
<namePart type="family">Gers</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alexander</namePart>
<namePart type="family">Löser</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Fifteenth Language Resources and Evaluation Conference</title>
</titleInfo>
<name type="personal">
<namePart type="given">Stelios</namePart>
<namePart type="family">Piperidis</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Núria</namePart>
<namePart type="family">Bel</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Henk</namePart>
<namePart type="family">van den Heuvel</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nancy</namePart>
<namePart type="family">Ide</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Simon</namePart>
<namePart type="family">Krek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Antonio</namePart>
<namePart type="family">Toral</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA Language Resource Association</publisher>
<place>
<placeTerm type="text">Palma de Mallorca, Spain</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Large language models (LLMs) show strong reasoning abilities, but full retraining for the medical domain is often infeasible because of lacking data or compute resources. We present DeepICD-R1, a framework for efficient medical reasoning fine-tuning that unites hierarchical rewards with distilled supervision. We reformulate ICD-10-CM prediction as a reinforcement learning problem and design a hierarchical outcome-based reward that reflects the ICD code structure across chapter, category, and full-code levels. In parallel, we publish a large-scale distilled dataset of over 90k reasoning traces derived from MIMIC-IV admission notes, integrating clinical validation and official coding guidelines. Fine-tuning smaller instruction-tuned LLMs with this data and GRPO reinforcement yields consistent gains in diagnostic accuracy and reasoning coherence. Extensive ablations confirm that hierarchical supervision and verifiable outcome rewards enable competitive, domain-specialized reasoning models without additional pretraining, providing a reproducible foundation for clinical NLP research. Keywords: Clinical NLP, Large Reasoning Model, GRPO, Supervised Fine-Tuning</abstract>
<identifier type="citekey">rohr-etal-2026-deepicd</identifier>
<identifier type="doi">10.63317/5ntn3gnmv9cy</identifier>
<location>
<url>https://aclanthology.org/2026.lrec-1.843/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>10764</start>
<end>10775</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T DeepICD-R1: Medical Reasoning through Hierarchical Rewards and Unsupervised Distillation
%A Röhr, Tom
%A Steffek, Thomas Maximilian Josef
%A Teucher, Roman
%A Bressem, Keno
%A Figueroa, Alexei
%A Grundmann, Paul
%A Troeger, Peter
%A Gers, Felix Alexander
%A Löser, Alexander
%Y Piperidis, Stelios
%Y Bel, Núria
%Y van den Heuvel, Henk
%Y Ide, Nancy
%Y Krek, Simon
%Y Toral, Antonio
%S Proceedings of the Fifteenth Language Resources and Evaluation Conference
%D 2026
%8 May
%I ELRA Language Resource Association
%C Palma de Mallorca, Spain
%F rohr-etal-2026-deepicd
%X Large language models (LLMs) show strong reasoning abilities, but full retraining for the medical domain is often infeasible because of lacking data or compute resources. We present DeepICD-R1, a framework for efficient medical reasoning fine-tuning that unites hierarchical rewards with distilled supervision. We reformulate ICD-10-CM prediction as a reinforcement learning problem and design a hierarchical outcome-based reward that reflects the ICD code structure across chapter, category, and full-code levels. In parallel, we publish a large-scale distilled dataset of over 90k reasoning traces derived from MIMIC-IV admission notes, integrating clinical validation and official coding guidelines. Fine-tuning smaller instruction-tuned LLMs with this data and GRPO reinforcement yields consistent gains in diagnostic accuracy and reasoning coherence. Extensive ablations confirm that hierarchical supervision and verifiable outcome rewards enable competitive, domain-specialized reasoning models without additional pretraining, providing a reproducible foundation for clinical NLP research. Keywords: Clinical NLP, Large Reasoning Model, GRPO, Supervised Fine-Tuning
%R 10.63317/5ntn3gnmv9cy
%U https://aclanthology.org/2026.lrec-1.843/
%U https://doi.org/10.63317/5ntn3gnmv9cy
%P 10764-10775
Markdown (Informal)
[DeepICD-R1: Medical Reasoning through Hierarchical Rewards and Unsupervised Distillation](https://aclanthology.org/2026.lrec-1.843/) (Röhr et al., LREC 2026)
ACL
- Tom Röhr, Thomas Maximilian Josef Steffek, Roman Teucher, Keno Bressem, Alexei Figueroa, Paul Grundmann, Peter Troeger, Felix Alexander Gers, and Alexander Löser. 2026. DeepICD-R1: Medical Reasoning through Hierarchical Rewards and Unsupervised Distillation. In Proceedings of the Fifteenth Language Resources and Evaluation Conference, pages 10764–10775, Palma de Mallorca, Spain. ELRA Language Resource Association.