@inproceedings{yuan-etal-2026-diagnosis,
title = "Diagnosis Before Treatment? Investigating Intermediate-Variable Reasoning in {LLM}s",
author = "Yuan, Zhishan and
Almeida, Tiago and
Hao, Shudong and
Zhu, Zining and
Dawlabani, Nickolas Elias and
Pervaiz, Ghazia and
Ning, Yue",
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.344/",
pages = "7220--7235",
ISBN = "979-8-89176-462-0",
abstract = "Many expert-domain decisions involve intermediate variables, yet standard LLM evaluation often measures only final-answer accuracy. We present a controlled study of intermediate-variable dependence in medical treatment selection, using diagnosis as an explicit variable between patient phenotype and treatment. We construct a six-option multiple-choice benchmark from PrimeKG-derived phenotype{--}disease{--}drug chains and evaluate six LLMs using behavioral diagnosis interventions, held-out QLoRA path supervision, external validation, and auxiliary layer-wise localization analysis. Providing the gold diagnosis improves average treatment accuracy from 49.9{\%} to 84.9{\%}, while incorrect predicted diagnoses reduce accuracy from 85.5{\%} to 23.1{\%}. A shuffled-diagnosis control yields 18.2{\%} accuracy, suggesting that the effect depends on diagnosis correctness rather than merely adding disease-like text. Held-out QLoRA experiments show that path supervision improves treatment selection, with diagnosis-first full-chain and multitask objectives yielding the strongest gains. Center-of-Gravity localization further gives a modest diagnosis-first signal, which we interpret as a localization pattern rather than evidence of faithful clinical reasoning. Overall, in this controlled setting, final-answer accuracy can obscure failures in intermediate-variable dependence, while structured supervision can partially improve diagnosis-to-treatment behavior."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yuan-etal-2026-diagnosis">
<titleInfo>
<title>Diagnosis Before Treatment? Investigating Intermediate-Variable Reasoning in LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zhishan</namePart>
<namePart type="family">Yuan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tiago</namePart>
<namePart type="family">Almeida</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shudong</namePart>
<namePart type="family">Hao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zining</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nickolas</namePart>
<namePart type="given">Elias</namePart>
<namePart type="family">Dawlabani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ghazia</namePart>
<namePart type="family">Pervaiz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Ning</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>Many expert-domain decisions involve intermediate variables, yet standard LLM evaluation often measures only final-answer accuracy. We present a controlled study of intermediate-variable dependence in medical treatment selection, using diagnosis as an explicit variable between patient phenotype and treatment. We construct a six-option multiple-choice benchmark from PrimeKG-derived phenotype–disease–drug chains and evaluate six LLMs using behavioral diagnosis interventions, held-out QLoRA path supervision, external validation, and auxiliary layer-wise localization analysis. Providing the gold diagnosis improves average treatment accuracy from 49.9% to 84.9%, while incorrect predicted diagnoses reduce accuracy from 85.5% to 23.1%. A shuffled-diagnosis control yields 18.2% accuracy, suggesting that the effect depends on diagnosis correctness rather than merely adding disease-like text. Held-out QLoRA experiments show that path supervision improves treatment selection, with diagnosis-first full-chain and multitask objectives yielding the strongest gains. Center-of-Gravity localization further gives a modest diagnosis-first signal, which we interpret as a localization pattern rather than evidence of faithful clinical reasoning. Overall, in this controlled setting, final-answer accuracy can obscure failures in intermediate-variable dependence, while structured supervision can partially improve diagnosis-to-treatment behavior.</abstract>
<identifier type="citekey">yuan-etal-2026-diagnosis</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.344/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>7220</start>
<end>7235</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Diagnosis Before Treatment? Investigating Intermediate-Variable Reasoning in LLMs
%A Yuan, Zhishan
%A Almeida, Tiago
%A Hao, Shudong
%A Zhu, Zining
%A Dawlabani, Nickolas Elias
%A Pervaiz, Ghazia
%A Ning, Yue
%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 yuan-etal-2026-diagnosis
%X Many expert-domain decisions involve intermediate variables, yet standard LLM evaluation often measures only final-answer accuracy. We present a controlled study of intermediate-variable dependence in medical treatment selection, using diagnosis as an explicit variable between patient phenotype and treatment. We construct a six-option multiple-choice benchmark from PrimeKG-derived phenotype–disease–drug chains and evaluate six LLMs using behavioral diagnosis interventions, held-out QLoRA path supervision, external validation, and auxiliary layer-wise localization analysis. Providing the gold diagnosis improves average treatment accuracy from 49.9% to 84.9%, while incorrect predicted diagnoses reduce accuracy from 85.5% to 23.1%. A shuffled-diagnosis control yields 18.2% accuracy, suggesting that the effect depends on diagnosis correctness rather than merely adding disease-like text. Held-out QLoRA experiments show that path supervision improves treatment selection, with diagnosis-first full-chain and multitask objectives yielding the strongest gains. Center-of-Gravity localization further gives a modest diagnosis-first signal, which we interpret as a localization pattern rather than evidence of faithful clinical reasoning. Overall, in this controlled setting, final-answer accuracy can obscure failures in intermediate-variable dependence, while structured supervision can partially improve diagnosis-to-treatment behavior.
%U https://aclanthology.org/2026.findings-emnlp.344/
%P 7220-7235
Markdown (Informal)
[Diagnosis Before Treatment? Investigating Intermediate-Variable Reasoning in LLMs](https://aclanthology.org/2026.findings-emnlp.344/) (Yuan et al., Findings 2026)
ACL