When Chain-of-Thought Backfires: Evaluating Prompt Sensitivity in Medical Language Models

When Chain-of-Thought Backfires: Evaluating Prompt Sensitivity in Medical Language Models | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article When Chain-of-Thought Backfires: Evaluating Prompt Sensitivity in Medical Language Models Binesh Sadanandan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8759042/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Large language models are increasingly deployed in medical settings, yet their sensitivity to prompt formatting remains poorly characterized. We evaluate MedGemma (4B and 27B variants) on MedMCQA (4,183 questions) and PubMedQA (1,000 questions). Our experiments reveal concerning findings: chain- of-thought prompting decreases accuracy by 5.7% compared to direct answering; few-shot examples degrade performance by 11.9% while increasing position bias from 0.14 to 0.47; shuffling answer options causes the model to change predictions 59.1% of the time with accuracy dropping up to 27.4 percentage points; and truncating context to 50% causes accuracy to plummet below the no-context baseline. These results demonstrate that prompt engineering techniques validated on general-purpose models do not transfer to domain-specific medical LLMs. Artificial Intelligence and Machine Learning Full Text Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8759042","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":583914379,"identity":"678cbf39-4f8e-4784-b1f7-5ab3eb7a4d8a","order_by":0,"name":"Binesh Sadanandan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA90lEQVRIiWNgGAWjYBACAyCWYDCwAZEgcAAqzkZIS0EayVo+HCZBizn/4Yc3fhicT9wu3cD24GPbHXmD470HGD6UHcapxXJGmrFlj8HtxJ1zDrAbzmx7ZrjhzLkExhnncGsxuMFgJsED1LLhRgKbNO+2w4wzZ+QYMPO24dFy/vg3yT8G5yBa/m47bD9z/hsD5r/4tBzIMZPmMTgA0cK47XBiP9BSZkY8Wixn5BRbyxgkG2+4kdgm2fvvcHI/T47BwZ5z6Ti1mPMf33jzzR872Q03ko9J/Dhz2LaN/Yzhgx9l1ji1IAHGBjjzADHqR8EoGAWjYBTgBgD8Dl4y+xIJ0QAAAABJRU5ErkJggg==","orcid":"","institution":"University of New Haven","correspondingAuthor":true,"prefix":"","firstName":"Binesh","middleName":"","lastName":"Sadanandan","suffix":""}],"badges":[],"createdAt":"2026-02-01 22:43:50","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-8759042/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8759042/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":101731998,"identity":"0a1cf1c1-0245-4694-a0e2-07a429f6e948","added_by":"auto","created_at":"2026-02-03 06:20:53","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":261992,"visible":true,"origin":"","legend":"","description":"","filename":"MedMCQARobustnessStudy.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8759042/v1_covered_35ef55a2-9970-4a8d-b5f4-a048afbd2c91.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eWhen Chain-of-Thought Backfires: Evaluating Prompt Sensitivity\u003c/p\u003e\n\u003cp\u003ein Medical Language Models\u003c/p\u003e","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"University of New Haven","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8759042/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8759042/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eLarge language models are increasingly deployed in medical settings, yet their sensitivity to prompt formatting remains poorly characterized. We evaluate MedGemma (4B and 27B variants) on MedMCQA (4,183 questions) and PubMedQA (1,000 questions). Our experiments reveal concerning findings: chain- of-thought prompting decreases accuracy by 5.7% compared to direct answering; few-shot examples degrade performance by 11.9% while increasing position bias from 0.14 to 0.47; shuffling answer options causes the model to change predictions 59.1% of the time with accuracy dropping up to 27.4 percentage points; and truncating context to 50% causes accuracy to plummet below the no-context baseline. These results demonstrate that prompt engineering techniques validated on general-purpose models do not transfer to domain-specific medical LLMs.\u003c/p\u003e","manuscriptTitle":"When Chain-of-Thought Backfires: Evaluating Prompt Sensitivity\nin Medical Language Models","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-03 06:20:47","doi":"10.21203/rs.3.rs-8759042/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fc1dd66e-8fd7-4a94-aea3-68e07e2f009b","owner":[],"postedDate":"February 3rd, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":62116564,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2026-02-03T06:20:47+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-03 06:20:47","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8759042","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8759042","identity":"rs-8759042","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}