@inproceedings{bezancon-etal-2026-pico,
title = "{PICO} Entity Datasets and Methodologies in the Context of Randomized Controlled Trials: A Case Study",
author = "Bezan{\c{c}}on, Julien and
Zweigenbaum, Pierre and
Naderi, Nona",
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.2372/",
pages = "50973--50983",
ISBN = "979-8-89176-462-0",
abstract = "Randomized Controlled Trials (RCTs) contain crucial information about the outcomes of clinical experiments. However, the sheer number of RCTs{---}approximately 595,000 asof September 2023{---}makes it difficult to efficiently navigate the available evidence. To address this challenge, medical practitioners rely on the PICO framework to identify relevant information and NLP engineers have developed automated PICO extraction pipelines. In this work, we study four PICO-related datasets created from RCTs and commonly used in the literature. We then compare two PICO extraction pipelines: PICOX reproduction and BIOBERT overlaps. Our findings show that (i) PICO-related datasets can be broadly categorized as either span-based or entity-based, with clear differences in how PICO information is annotated, and (ii) while PICOX reproduction is better suited to extracting text spans containing PICO information, BIOBERT overlaps is more suitable for fine-grained and precise PICO entity extraction. Additionally, we evaluate the performance of both models on nested and over- lapping PICO entities, found in the EBM-NLP dataset. BIOBERT overlaps achieves an entity-level micro-F1 score of 30.70 on these entities, compared with 17.58 for PICOX reproduction."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="bezancon-etal-2026-pico">
<titleInfo>
<title>PICO Entity Datasets and Methodologies in the Context of Randomized Controlled Trials: A Case Study</title>
</titleInfo>
<name type="personal">
<namePart type="given">Julien</namePart>
<namePart type="family">Bezançon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pierre</namePart>
<namePart type="family">Zweigenbaum</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nona</namePart>
<namePart type="family">Naderi</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>Randomized Controlled Trials (RCTs) contain crucial information about the outcomes of clinical experiments. However, the sheer number of RCTs—approximately 595,000 asof September 2023—makes it difficult to efficiently navigate the available evidence. To address this challenge, medical practitioners rely on the PICO framework to identify relevant information and NLP engineers have developed automated PICO extraction pipelines. In this work, we study four PICO-related datasets created from RCTs and commonly used in the literature. We then compare two PICO extraction pipelines: PICOX reproduction and BIOBERT overlaps. Our findings show that (i) PICO-related datasets can be broadly categorized as either span-based or entity-based, with clear differences in how PICO information is annotated, and (ii) while PICOX reproduction is better suited to extracting text spans containing PICO information, BIOBERT overlaps is more suitable for fine-grained and precise PICO entity extraction. Additionally, we evaluate the performance of both models on nested and over- lapping PICO entities, found in the EBM-NLP dataset. BIOBERT overlaps achieves an entity-level micro-F1 score of 30.70 on these entities, compared with 17.58 for PICOX reproduction.</abstract>
<identifier type="citekey">bezancon-etal-2026-pico</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2372/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>50973</start>
<end>50983</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T PICO Entity Datasets and Methodologies in the Context of Randomized Controlled Trials: A Case Study
%A Bezançon, Julien
%A Zweigenbaum, Pierre
%A Naderi, Nona
%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 bezancon-etal-2026-pico
%X Randomized Controlled Trials (RCTs) contain crucial information about the outcomes of clinical experiments. However, the sheer number of RCTs—approximately 595,000 asof September 2023—makes it difficult to efficiently navigate the available evidence. To address this challenge, medical practitioners rely on the PICO framework to identify relevant information and NLP engineers have developed automated PICO extraction pipelines. In this work, we study four PICO-related datasets created from RCTs and commonly used in the literature. We then compare two PICO extraction pipelines: PICOX reproduction and BIOBERT overlaps. Our findings show that (i) PICO-related datasets can be broadly categorized as either span-based or entity-based, with clear differences in how PICO information is annotated, and (ii) while PICOX reproduction is better suited to extracting text spans containing PICO information, BIOBERT overlaps is more suitable for fine-grained and precise PICO entity extraction. Additionally, we evaluate the performance of both models on nested and over- lapping PICO entities, found in the EBM-NLP dataset. BIOBERT overlaps achieves an entity-level micro-F1 score of 30.70 on these entities, compared with 17.58 for PICOX reproduction.
%U https://aclanthology.org/2026.findings-emnlp.2372/
%P 50973-50983
Markdown (Informal)
[PICO Entity Datasets and Methodologies in the Context of Randomized Controlled Trials: A Case Study](https://aclanthology.org/2026.findings-emnlp.2372/) (Bezançon et al., Findings 2026)
ACL