JAMAHealthForumTM

ResearchLetter|AIinHealthPolicy

AdoptionofArtificialIntelligenceintheHealthCareSector

ThuyD.Nguyen,PhD;ChristopherM.Whaley,PhD;KosaliSimon,PhD;NeilMehta,BS;HaoYu,PhD;RyanK.McBain,PhD,MPH;AteevMehrotra,MD,MPH;JonathanH.Cantor,PhD

Introduction

Adoptionofartificialintelligence(AI)isexpectedtohaveimportantimplicationsforhealthcare

deliveryandworkforce.1AIcanshapepatienteducationandengagement,streamline

documentation,andassistclinicianswithinformationsynthesis.2However,therearelimiteddataonreal-timetrendsofAIadoptioninthehealthcaresectorandhowthiscompareswithothersectorsoftheeconomy.3Tofillthisknowledgegap,wecomparedAIuseinhealthcarevsothersectorsfrom2023to2025.

Methods

WeanalyzedtheUSCensusBureau’sBusinessTrendsandOutlookSurvey(BTOS)toexamine

changesinAIusefromSeptember2023toMay2025.4InaccordancewiththeCommonRule,thiscross-sectionalstudywasexemptfromethicsreviewandinformedconsentbecauseitused

deidentified,publiclyavailableBTOSdata.Wefollowedthe

STROBE

reportingguideline.

TheBTOSasks1.2millionparticipatingfirmsannuallyabouttheircurrentuseofAIforbusinesspurposes.Specifically,firmsareasked,“Inthelasttwoweeks,didthisbusinessuse…[AI]in

producinggoodsorservices?(ExamplesofAI:machinelearning,naturallanguageprocessing,virtualagents,voicerecognition,etc.).”TheNorthAmericanIndustryClassificationSystemcodesusedtoidentifyeachsectorandhealthcaresubsectorsaredescribedintheeMethodsin

Supplement1

.WereportedunadjustedbiweeklytrendsinthepercentageoffirmsrespondingyestousingAIinthe

healthcarevsselectothersectors.

WeperformedpiecewiselinearregressionwiththenlhockeyprograminStata(StataCorp)toidentifythebreakpointinbiweeklyAIusetrends.WethenconductedaninterruptedtimeseriesanalysistoassesschangesintheslopeofAIuse.AnalyseswereconductedwithStata,version18.0.Two-sidedhypothesistestswereusedwithasignificancelevelof.05(P<.05).

Results

BetweenSeptember2023andMay2025,themean(SD)AIuseinhealthcarebyfirmswas5.9%

(1.6%;119300firm-levelresponses)andincreasedovertime.In2025,AIuseinhealthcare(8.3%)wasstilllowerthaninothersectors,suchasfinanceandinsurance(11.6%);education(15.1%);

professional,scientific,andtechnicalservices(19.2%);andinformationservices(23.2%)(Figure,A).TheestimatedbreakpointinAIusetrendsinthehealthcaresectorwasDecember30,2024,to

January12,2025(Figure,B;Table).Atthistransition,theslopeshiftedsignificantlyfromnearlyflatin2023through2024(biweeklypercentageincrease:0.005%;95%CI,0.004%-0.007%)to

graduallyincreasing(0.03%;95%CI,0.02%-0.03%)—a481.5%change.

Withinhealthcare,thelargestgainswereforoutpatientandambulatorycare(Figure,C),wherethepercentageoffirmsusingAIincreasedfrom4.6%in2023to8.7%in2025.Nursingand

residentialcarefacilitiesexperiencedmorelimitedgrowth:3.1%in2023to4.5%in2025.

+

Supplementalcontent

Authoraffiliationsandarticleinformationarelistedattheendofthisarticle.

OpenAccess.ThisisanopenaccessarticledistributedunderthetermsoftheCC-BYLicense.

JAMAHealthForum.2025;6(11):e255029.doi:10.1001/jamahealthforum.2025.5029November21,20251/4

JAMAHealthForum.2025;6(11):e255029.doi:10.1001/jamahealthforum.2025.5029(Reprinted)November21,20252/4

JAMAHealthForum|ResearchLetterAdoptionofArtificialIntelligenceintheHealthCareSector

Discussion

TimelyestimatesofAIuseamongUSfirmsindicatethat,whileAIadoptioninhealthcarelagsbehindothersectors,ithasbeenrapidlyincreasingsince2023,particularlyamongoutpatientand

ambulatoryorganizations.FutureresearchisnecessarytounderstandthereasonsandconsequencesoflowerrateofAIadoptioninhealthcare,particularlyincertainsubsectors,suchasnursingand

residentialcarefacilities.

Figure.TrendsinArtificialIntelligence(AI)UseinHealthCarevsNon-HealthCareSectors

Firms,%

30

25

20

15

10

5

0

AFirmscurrentlyusingAI

Informationservices

Education

Financeandinsurance

Professional,scientific,andtechnicalservices

Healthcare

SepOctNovDec11964

JanJanFebMarAprMayJunJulAugSepOctNovDecDecJanFebMarAprMay129262522201715129742302724242119

2023

20242025

Firstdateofbiweeklydatacollection

B

ChangeinAIusetrajectoryinhealthcare,2023-2024vs2025

15

Firms,%

10

5

0

Actual

Estimated

SepOctNovDec11964

JanJanFebMarAprMayJunJulAugSepOctNovDecDecJanFebMarAprMay129262522201715129742302724242119

2023

20242025

Firstdateofbiweeklydatacollection

C

FirmscurrentlyusingAIinhealthcarebysubsector

15

10

Firms,%

5

0

A,Theresponseestimateisplottedforthepercentageoffirmsansweringyestothequestion:“Inthelasttwoweeks,didthisbusinessuse…[AI]inproducinggoodsorservices?(ExamplesofAI:machinelearning,naturallanguageprocessing,virtualagents,voicerecognition,etc.).”B,Theestimatedvaluesarecalculatedfrom

ordinaryleastsquaresregressionsofthepercentageoffirmsusingAIonbiweeklytrendsbeforeandaftertheestimatedbreakpointonDecember30,2024.C,TheUSCensusBureausuppressedpercentageestimates

fornursingandresidentialcarefacilitiesforspecificperiods,aswellasforhospitalsacrossallperiods,duetoconfidentialityreasons.

Outpatientandambulatory

Nursingandresidential

SepOctNovDec11964

JanJanFebMarAprMayJunJulAugSepOctNovDecDecJanFebMarAprMay129262522201715129742302724242119

2023

20242025

Firstdateofbiweeklydatacollection

JAMAHealthForum|ResearchLetterAdoptionofArtificialIntelligenceintheHealthCareSector

JAMAHealthForum.2025;6(11):e255029.doi:10.1001/jamahealthforum.2025.5029(Reprinted)November21,20253/4

Table.ChangesinPercentageof119300Firm-LevelResponsesonCurrentUseofArtificialIntelligence(AI)intheHealthCareSectora

FirmscurrentlyusingAI,value(95%CI),%

Intercept

Beforebreakpointb

6.5(6.1-6.8)

Afterbreakpointb

6.3(5.7-6.9)

Changeinlevel

?0.2

Pvalue

.56

Biweeklychange

Beforebreakpoint

0.005(0.004-0.007)

Afterbreakpoint

0.03(0.02-0.03)

Changeinbiweeklyslope

0.02

Pvalue

<.001

Dependentvariablemean(SD)

5.9(1.6)

TheAIadoptionrateobservedinthisstudyislowerthanestimatesinpreviousnationalstudiesfocusingonwell-resourcedorganizations5andhospitals.6Forexample,Poonetal5examinedstagesofAIadoptioninhealthcareorganizationsfromdevelopmenttofulldeploymentandfoundthat0%to48%oftheseorganizationsreportedfullimplementationamongusecasecategories.Incontrast,theBTOStracksuseofAIinproducinggoodsandservicesbysurveyingabroaderrangeof

organizationsacrossvarioussectorsandemploymentsizes,includingsmallerfirms.Astudy

limitationisthatBTOSsubestimatesforhospitalsforthestudyperiodaresuppressedbytheUS

CensusBureauduetoconfidentialityreasons.RapidadoptionofAIinhealthcarehighlightsthe

urgentneedforactivemonitoringandeffectiveregulationstoensuresafeandefficientdeploymentofAIinpatientcare.

ARTICLEINFORMATION

AcceptedforPublication:September10,2025.

Published:November21,2025.doi

:10.1001/jamahealthforum.2025.5029

OpenAccess:Thisisanopenaccessarticledistributedunderthetermsofthe

CC-BYLicense

.?2025NguyenTDetal.JAMAHealthForum.

CorrespondingAuthor:ThuyD.Nguyen,PhD,DepartmentofHealthManagementandPolicy,UniversityofMichiganSchoolofPublicHealth,1415WashingtonHeights,AnnArbor,MI48109

(thuydn@

).

AuthorAffiliations:DepartmentofHealthManagementandPolicy,UniversityofMichigan,AnnArbor(Nguyen);DepartmentofHealthServices,Policy,andPractice,BrownUniversitySchoolofPublicHealth,Providence,RhodeIsland(Whaley,Mehta);O’NeillSchoolofPublicandEnvironmentalAffairs,IndianaUniversity,Bloomington

(Simon);DepartmentofPopulationMedicine,HarvardMedicalSchoolandHarvardPilgrimHealthCareInstitute,Boston,Massachusetts(Yu);RANDCorporation,Arlington,Virginia(McBain);DepartmentofHealthServices,

PolicyandPractice,BrownUniversitySchoolofPublicHealth,Providence,RhodeIsland(Mehrotra);RANDCorporation,SantaMonica,California(Cantor).

AuthorContributions:DrNguyenhadfullaccesstoallofthedatainthestudyandtakesresponsibilityfortheintegrityofthedataandtheaccuracyofthedataanalysis.

Conceptanddesign:Nguyen,Whaley,Simon,Mehrotra,Cantor.

Acquisition,analysis,orinterpretationofdata:Nguyen,Whaley,Mehta,Yu,McBain,Cantor.

Draftingofthemanuscript:Nguyen,Mehta,Cantor.

Criticalreviewofthemanuscriptforimportantintellectualcontent:Nguyen,Whaley,Simon,Yu,McBain,Mehrotra,Cantor.

Statisticalanalysis:Nguyen,Whaley,Mehta,Cantor.

Obtainedfunding:Whaley,Yu.

Administrative,technical,ormaterialsupport:Whaley,McBain.

Supervision:Yu,McBain,Mehrotra.

aTheestimatedmeans(intercepts)andbiweeklychangeslopewerecalculatedusingordinaryleastsquaresregressionofthepercentageoffirms

reportingcurrentuseofAIonbiweeklytrends.

bTheestimatedbreakpointwasDecember30,2024,toJanuary12,2025.

JAMAHealthForum|ResearchLetterAdoptionofArtificialIntelligenceintheHealthCareSector

JAMAHealthForum.2025;6(11):e255029.doi:10.1001/jamahealthforum.2025.5029(Reprinted)November21,20254/4

ConflictofInterestDisclosures:DrWhaleyreportedreceivinggrantsfromtheNationalInstituteonAging(NIA)duringtheconductofthestudy;grantsfromtheRobertWoodJohnsonFoundation,ArnoldVentures,the

CommonwealthFund,andPatientRightsAdvocates;andpersonalfeesfromICERandAnalysisGroupoutsidethesubmittedwork.DrYureportedreceivinggrantsfromtheNationalInstituteofMentalHealth(NIMH)duringtheconductofthestudyandgrantsfromtheNIMH,theNationalInstituteonAlcoholAbuseandAlcoholism,the

NationalInstituteofNursingResearch,andtheNationalInstituteonMinorityHealthandHealthDisparitiesoutsidethesubmittedwork.DrMehrotrareportedreceivinggrantsfromtheNationalInstitutesofHealthduringthe

conductofthestudyandpersonalfeesfromBlackOpalVenturesAdvisoryBoardoutsidethesubmittedwork.

DrCantorreportedreceivinggrantsfromtheNIMHduringtheconductofthestudy;grantsfromtheNIA;andpersonalfeesfromChestnutHealth,TheAspenInstitute,GovernmentAccountabilityOffice,andtheNIMH

outsidethesubmittedwork.Nootherdisclosureswerereported.

Funding/Support:ThisstudywasfundedbygrantR01AG073286fromtheNIA(DrWhaley)andgrantR01MH132551fromtheNIMH(DrYu).

RoleoftheFunder/Sponsor:Thefundershadnoroleinthedesignandconductofthestudy;collection,

management,analysis,andinterpretationofthedata;preparation,review,orapprovalofthemanuscript;anddecisiontosubmitthemanuscriptforpublication.

DataSharingStatement:See

Supplement2

.

AdditionalContributions:WethankMaddiePotter,BS,O'NeilSchoolof