@inproceedings{mahzoon-etal-2026-learning,
title = "Learning to Extract Physician-Determined Diagnosis from Clinical Narratives Using Dual Noisy Labels",
author = "Mahzoon, Vahid and
Talebianaraki, Elmira and
Ojeniyi, Solabomi and
Efferen, Tobin and
Koul, Sheetal and
Vucetic, Zivjena and
Gillespie, Avrum and
Vucetic, Slobodan",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.69/",
pages = "1395--1411",
ISBN = "979-8-89176-462-0",
abstract = "Retrospective studies using electronic health records (EHRs) often require diagnosis labels that reflect whether treating physicians considered a disease to be present. Such physician-determined diagnosis (PDD) labels are expensive to obtain through expert review, while available alternatives provide only noisy proxies. We exploit two complementary proxy labels: administrative diagnostic codes and free-text disease listings in discharge summaries. We introduce a dual-noisy-label learning framework that models PDD as a latent variable and treats these proxies as noisy observations of it, allowing a pretrained language model to learn from their agreement and disagreement without expert-labeled training data. We evaluate the approach on extracting Acute Kidney Injury (AKI) PDD from MIMIC-IV discharge summaries. Based on an expert-adjudicated evaluation set of 150 summaries, our PLM-PDD model achieves 94.6{\%} population-weighted accuracy, compared with 89.4{\%} for the same model trained using administrative coding, and 85.9{\%} and 78.0{\%} for AKI proxies. Largest gains occur in the gray-zone, where the proxies disagree. Ablation experiments show that gray-zone examples and auxiliary ICD supervision contribute to performance, while expert-highlighted evidence receives substantially higher model attention than matched random text. These results demonstrate that complementary noisy EHR labels can be used to learn clinically meaningful diagnosis labels without requiring expert annotation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="mahzoon-etal-2026-learning">
<titleInfo>
<title>Learning to Extract Physician-Determined Diagnosis from Clinical Narratives Using Dual Noisy Labels</title>
</titleInfo>
<name type="personal">
<namePart type="given">Vahid</namePart>
<namePart type="family">Mahzoon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Elmira</namePart>
<namePart type="family">Talebianaraki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Solabomi</namePart>
<namePart type="family">Ojeniyi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tobin</namePart>
<namePart type="family">Efferen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sheetal</namePart>
<namePart type="family">Koul</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zivjena</namePart>
<namePart type="family">Vucetic</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Avrum</namePart>
<namePart type="family">Gillespie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Slobodan</namePart>
<namePart type="family">Vucetic</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Retrospective studies using electronic health records (EHRs) often require diagnosis labels that reflect whether treating physicians considered a disease to be present. Such physician-determined diagnosis (PDD) labels are expensive to obtain through expert review, while available alternatives provide only noisy proxies. We exploit two complementary proxy labels: administrative diagnostic codes and free-text disease listings in discharge summaries. We introduce a dual-noisy-label learning framework that models PDD as a latent variable and treats these proxies as noisy observations of it, allowing a pretrained language model to learn from their agreement and disagreement without expert-labeled training data. We evaluate the approach on extracting Acute Kidney Injury (AKI) PDD from MIMIC-IV discharge summaries. Based on an expert-adjudicated evaluation set of 150 summaries, our PLM-PDD model achieves 94.6% population-weighted accuracy, compared with 89.4% for the same model trained using administrative coding, and 85.9% and 78.0% for AKI proxies. Largest gains occur in the gray-zone, where the proxies disagree. Ablation experiments show that gray-zone examples and auxiliary ICD supervision contribute to performance, while expert-highlighted evidence receives substantially higher model attention than matched random text. These results demonstrate that complementary noisy EHR labels can be used to learn clinically meaningful diagnosis labels without requiring expert annotation.</abstract>
<identifier type="citekey">mahzoon-etal-2026-learning</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.69/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>1395</start>
<end>1411</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Learning to Extract Physician-Determined Diagnosis from Clinical Narratives Using Dual Noisy Labels
%A Mahzoon, Vahid
%A Talebianaraki, Elmira
%A Ojeniyi, Solabomi
%A Efferen, Tobin
%A Koul, Sheetal
%A Vucetic, Zivjena
%A Gillespie, Avrum
%A Vucetic, Slobodan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F mahzoon-etal-2026-learning
%X Retrospective studies using electronic health records (EHRs) often require diagnosis labels that reflect whether treating physicians considered a disease to be present. Such physician-determined diagnosis (PDD) labels are expensive to obtain through expert review, while available alternatives provide only noisy proxies. We exploit two complementary proxy labels: administrative diagnostic codes and free-text disease listings in discharge summaries. We introduce a dual-noisy-label learning framework that models PDD as a latent variable and treats these proxies as noisy observations of it, allowing a pretrained language model to learn from their agreement and disagreement without expert-labeled training data. We evaluate the approach on extracting Acute Kidney Injury (AKI) PDD from MIMIC-IV discharge summaries. Based on an expert-adjudicated evaluation set of 150 summaries, our PLM-PDD model achieves 94.6% population-weighted accuracy, compared with 89.4% for the same model trained using administrative coding, and 85.9% and 78.0% for AKI proxies. Largest gains occur in the gray-zone, where the proxies disagree. Ablation experiments show that gray-zone examples and auxiliary ICD supervision contribute to performance, while expert-highlighted evidence receives substantially higher model attention than matched random text. These results demonstrate that complementary noisy EHR labels can be used to learn clinically meaningful diagnosis labels without requiring expert annotation.
%U https://aclanthology.org/2026.findings-emnlp.69/
%P 1395-1411
Markdown (Informal)
[Learning to Extract Physician-Determined Diagnosis from Clinical Narratives Using Dual Noisy Labels](https://aclanthology.org/2026.findings-emnlp.69/) (Mahzoon et al., Findings 2026)
ACL
- Vahid Mahzoon, Elmira Talebianaraki, Solabomi Ojeniyi, Tobin Efferen, Sheetal Koul, Zivjena Vucetic, Avrum Gillespie, and Slobodan Vucetic. 2026. Learning to Extract Physician-Determined Diagnosis from Clinical Narratives Using Dual Noisy Labels. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 1395–1411, Budapest, Hungary. Association for Computational Linguistics.