Multi Armed Bandit Based Adaptive Model Selection for Clinical AI Governance An Infrastructure Approach to Safer AI Use in U.S. Healthcare

Multi Armed Bandit Based Adaptive Model Selection for Clinical AI Governance An Infrastructure Approach to Safer AI Use in U.S. Healthcare | 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 Multi Armed Bandit Based Adaptive Model Selection for Clinical AI Governance An Infrastructure Approach to Safer AI Use in U.S. Healthcare Julian Borges This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8651972/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 Background Clinical artificial intelligence systems are increasingly embedded in routine care delivery, including high risk decision contexts. Despite widespread adoption, most clinical AI systems are deployed using static model selection strategies based on aggregate offline validation metrics. Such approaches can obscure context dependent performance variation, permit systematic decision failures to persist during real world use, and limit the ability of health systems to manage safety, operational cost, and equity at deployment time. Objectives To evaluate whether a multi armed bandit based governance framework that incorporates patient context, operational cost constraints, and delayed outcome feedback can function as an adaptive deployment control mechanism for clinical AI systems. Methods We conducted a retrospective simulation study designed to emulate prospective, sequential deployment of clinical AI using structured electronic health record data from U.S. intensive care unit admissions. Multiple independently trained and previously validated mortality prediction models were treated as candidate decision arms within a contextual multi armed bandit framework. Model selection was governed by a fixed confidence action elimination policy. Rewards were defined using a composite function that explicitly balanced predictive performance, operational cost, and safety penalties motivated by prior audit evidence of clinically meaningful failure modes. Outcome feedback was modeled with realistic delays to reflect the temporal availability of clinical endpoints during deployment. Results Across 3,000 sequential ICU admissions, adaptive deployment governed by the multi armed bandit framework achieved lower cumulative regret than static single model deployment, uniform random selection, ensemble averaging, and heuristic switching strategies. Bandit governed deployment demonstrated improved decision safety, reflected by lower false negative rates, while maintaining comparable overall predictive accuracy. The governance layer also reduced mean operational cost per prediction and exhibited more stable performance across age and sex subgroups relative to static deployment. Conclusions Multi armed bandit based governance offers a principled and scalable approach for translating audit identified risks in clinical AI into adaptive deployment control. By managing uncertainty, safety, and operational cost at the level of individual deployment decisions, this framework supports safer, more efficient, and more equitable use of clinical AI systems without requiring model retraining or disruptive workflow changes. Artificial Intelligence and Machine Learning Medical Informatics Biostatistics Other Public Policy Health Economics & Outcomes Research Multi armed bandits clinical artificial intelligence AI governance adaptive model selection contextual bandits deployment safety decision level auditing cost aware machine learning delayed feedback health informatics Full Text Additional Declarations The authors declare no competing interests. Supplementary Files MABGovernanceforClinicalAI.pdf MAB Governance for Clinical AI - Zenodo Repo README.md README 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-8651972","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":577645118,"identity":"135e30d0-487c-47c1-a27c-81e1ab9080c2","order_by":0,"name":"Julian Borges","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAuElEQVRIiWNgGAWjYBACxgYeCINfAkSykaJFcgaxWhgYoFoMbhCrhXlG7sHHFb9s7I1v9xgwfCg7TITDZuQlG57tS0vcdueMAeOMc0RpyTGTbOw5nGB2I8eAmbeNeC3/7Y1nALX8JVpLw48DjBskgFoYidLS8y7ZsLEhOXHGnWMFB3vOpRPWYtiee/Bhwx87e/7ZzRsf/CizJkLLhASgVW0QzgHC6oFAnh+k7g9RakfBKBgFo2CkAgAyuD6nhTFQ5QAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0009-0001-9929-3135","institution":"Boston University","correspondingAuthor":true,"prefix":"","firstName":"Julian","middleName":"","lastName":"Borges","suffix":""}],"badges":[],"createdAt":"2026-01-20 18:15:50","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-8651972/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8651972/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":100781504,"identity":"343512b1-6a62-4814-af49-4b37c6636c1b","added_by":"auto","created_at":"2026-01-21 11:43:49","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":889206,"visible":true,"origin":"","legend":"","description":"","filename":"BorgesJulian2026BanditModelSelectionMimicJAN20.docx","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/885a6e2d32fb31966115914e.docx"},{"id":100781802,"identity":"e3f521e9-e8ff-4049-9060-5b44744056c1","added_by":"auto","created_at":"2026-01-21 11:44:21","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs8651972.json","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/aa6a8256e192cbc98755c88d.json"},{"id":100781959,"identity":"d1200626-e5fe-4cb8-b32f-b84baf6b4b34","added_by":"auto","created_at":"2026-01-21 11:44:36","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":110571,"visible":true,"origin":"","legend":"","description":"","filename":"rs86519720enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/b50e8d360261baccfdc8a7dd.xml"},{"id":100782188,"identity":"41a3ec52-2feb-4afa-a325-acd223e5dad5","added_by":"auto","created_at":"2026-01-21 11:45:29","extension":"png","order_by":52,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":80212,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/822efb455c56e5ffc4e14d60.png"},{"id":100781685,"identity":"b9b29964-b7f2-4274-9804-3ecda6a34922","added_by":"auto","created_at":"2026-01-21 11:44:09","extension":"png","order_by":53,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":49770,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/e6018ab8c9dfbd955221363b.png"},{"id":100781975,"identity":"2eb42589-34e1-47bc-9067-9b271bb64585","added_by":"auto","created_at":"2026-01-21 11:44:38","extension":"png","order_by":54,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":278826,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/104d3d742d1db7db60a2911c.png"},{"id":100781775,"identity":"bb07bd8b-5623-4d3b-84b1-8631dcb19b0a","added_by":"auto","created_at":"2026-01-21 11:44:19","extension":"png","order_by":55,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":87474,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/0e860dd3b22cf3010f899006.png"},{"id":100781575,"identity":"185e3ede-128d-46e7-b19f-73643285570f","added_by":"auto","created_at":"2026-01-21 11:43:58","extension":"png","order_by":56,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":58705,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/02ab502eb46f648ffef38cf6.png"},{"id":100781406,"identity":"cc4bae67-f466-446a-8fba-59397789f318","added_by":"auto","created_at":"2026-01-21 11:43:29","extension":"png","order_by":57,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":92685,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/b1c828d67bbba3d9520ee2e5.png"},{"id":100782087,"identity":"4741eca7-c14b-459e-ac38-e451e1d92058","added_by":"auto","created_at":"2026-01-21 11:45:02","extension":"png","order_by":58,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":185683,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/e13ff4ac0385f0ff409bede8.png"},{"id":100782187,"identity":"42753b4e-3715-437f-91a1-38bcdabe851a","added_by":"auto","created_at":"2026-01-21 11:45:29","extension":"png","order_by":59,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":33767,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/e96d205d69aa85591edd96c4.png"},{"id":100781574,"identity":"720d06cd-5dd7-4adf-b21c-828d72660e5b","added_by":"auto","created_at":"2026-01-21 11:43:58","extension":"png","order_by":60,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19178,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/2c17409df08589c549a1115b.png"},{"id":100781384,"identity":"4607d0f9-6c48-4f4a-b4a9-ac50b03ccae4","added_by":"auto","created_at":"2026-01-21 11:43:27","extension":"png","order_by":61,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":125125,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/3e664374fc67438e9b98b468.png"},{"id":100781681,"identity":"f0aba436-4708-40bc-90d7-0cb3fbfdb87a","added_by":"auto","created_at":"2026-01-21 11:44:08","extension":"png","order_by":62,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":44745,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/3f78d1ce0075b14522a8b96d.png"},{"id":100781306,"identity":"b05428ec-466d-489b-919b-48c0194ff3fa","added_by":"auto","created_at":"2026-01-21 11:43:13","extension":"png","order_by":63,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":23393,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/bf3905331ec1c2c1bde1d4fb.png"},{"id":100782069,"identity":"32960306-ce75-4543-921a-caaddd37b7e1","added_by":"auto","created_at":"2026-01-21 11:44:56","extension":"png","order_by":64,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":39005,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/a8a8cf7d386cfb9de6ceedee.png"},{"id":100796689,"identity":"e5cc75e4-4199-4b2f-a125-c54adf024525","added_by":"auto","created_at":"2026-01-21 13:45:10","extension":"png","order_by":65,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":49756,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/fa60afe91764ff19951bd9b4.png"},{"id":100781622,"identity":"e8201aa0-514c-441e-ad1d-7b8145718b9d","added_by":"auto","created_at":"2026-01-21 11:44:04","extension":"xml","order_by":66,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":106936,"visible":true,"origin":"","legend":"","description":"","filename":"rs86519720structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/dd69ef03edff72a76df96f50.xml"},{"id":100781716,"identity":"734b03cc-9a3c-4a10-aee5-d66f2f4be8d6","added_by":"auto","created_at":"2026-01-21 11:44:16","extension":"html","order_by":67,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":134115,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/8a83b19d966b5037028679d2.html"},{"id":100777325,"identity":"1e1f737a-59f5-415d-a630-87c8a7c10da2","added_by":"auto","created_at":"2026-01-21 11:09:48","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":850664,"visible":true,"origin":"","legend":"","description":"","filename":"BorgesJulian2026BanditModelSelectionMimicJAN20.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1_covered_03b038ca-a955-49fd-aa94-bdeeb7663ac6.pdf"},{"id":100782189,"identity":"403cfa72-cdb3-4912-b85e-91f662aaeaca","added_by":"auto","created_at":"2026-01-21 11:45:29","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":398554,"visible":true,"origin":"","legend":"\u003cp\u003eMAB Governance for Clinical AI - Zenodo Repo\u003c/p\u003e","description":"","filename":"MABGovernanceforClinicalAI.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/8e670d439fee7287807d4e15.pdf"},{"id":100781382,"identity":"b2954ea6-2d08-4e70-a366-2f3a516f69fe","added_by":"auto","created_at":"2026-01-21 11:43:27","extension":"md","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":10209,"visible":true,"origin":"","legend":"\u003cp\u003eREADME\u003c/p\u003e","description":"","filename":"README.md","url":"https://assets-eu.researchsquare.com/files/rs-8651972/v1/db3199bc096d7556328df385.md"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eMulti Armed Bandit Based Adaptive Model Selection for Clinical AI Governance An Infrastructure Approach to Safer AI Use in U.S. Healthcare\u003c/p\u003e","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","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":"Multi armed bandits, clinical artificial intelligence, AI governance, adaptive model selection, contextual bandits, deployment safety, decision level auditing, cost aware machine learning, delayed feedback, health informatics","lastPublishedDoi":"10.21203/rs.3.rs-8651972/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8651972/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eClinical artificial intelligence systems are increasingly embedded in routine care delivery, including high risk decision contexts. Despite widespread adoption, most clinical AI systems are deployed using static model selection strategies based on aggregate offline validation metrics. Such approaches can obscure context dependent performance variation, permit systematic decision failures to persist during real world use, and limit the ability of health systems to manage safety, operational cost, and equity at deployment time.\u003c/p\u003e\u003ch2\u003eObjectives\u003c/h2\u003e \u003cp\u003eTo evaluate whether a multi armed bandit based governance framework that incorporates patient context, operational cost constraints, and delayed outcome feedback can function as an adaptive deployment control mechanism for clinical AI systems.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003e We conducted a retrospective simulation study designed to emulate prospective, sequential deployment of clinical AI using structured electronic health record data from U.S. intensive care unit admissions. Multiple independently trained and previously validated mortality prediction models were treated as candidate decision arms within a contextual multi armed bandit framework. Model selection was governed by a fixed confidence action elimination policy. Rewards were defined using a composite function that explicitly balanced predictive performance, operational cost, and safety penalties motivated by prior audit evidence of clinically meaningful failure modes. Outcome feedback was modeled with realistic delays to reflect the temporal availability of clinical endpoints during deployment.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eAcross 3,000 sequential ICU admissions, adaptive deployment governed by the multi armed bandit framework achieved lower cumulative regret than static single model deployment, uniform random selection, ensemble averaging, and heuristic switching strategies. Bandit governed deployment demonstrated improved decision safety, reflected by lower false negative rates, while maintaining comparable overall predictive accuracy. The governance layer also reduced mean operational cost per prediction and exhibited more stable performance across age and sex subgroups relative to static deployment.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eMulti armed bandit based governance offers a principled and scalable approach for translating audit identified risks in clinical AI into adaptive deployment control. By managing uncertainty, safety, and operational cost at the level of individual deployment decisions, this framework supports safer, more efficient, and more equitable use of clinical AI systems without requiring model retraining or disruptive workflow changes.\u003c/p\u003e","manuscriptTitle":"Multi Armed Bandit Based Adaptive Model Selection for Clinical AI Governance An Infrastructure Approach to Safer AI Use in U.S. Healthcare","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-21 10:23:42","doi":"10.21203/rs.3.rs-8651972/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":"a1b2541f-21ce-4dee-8b64-0ef225e0c118","owner":[],"postedDate":"January 21st, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":61457945,"name":"Artificial Intelligence and Machine Learning"},{"id":61457946,"name":"Medical Informatics"},{"id":61457947,"name":"Biostatistics"},{"id":61457948,"name":"Other Public Policy"},{"id":61457949,"name":"Health Economics \u0026 Outcomes Research"}],"tags":[],"updatedAt":"2026-01-21T10:23:42+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-21 10:23:42","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8651972","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8651972","identity":"rs-8651972","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}