《BCG&UCLA：2025人工智能与机器学习在医疗科技领域的崛起研究报告（英文版）（29页）.pdf》由会员分享，可在线阅读，更多相关《BCG&UCLA：2025人工智能与机器学习在医疗科技领域的崛起研究报告（英文版）（29页）.pdf（29页珍藏版）》请在三个皮匠报告上搜索。

1、Artificial Intelligence Stakes a Claim on MedtechMay 2025 By Christian Johnson,Jennifer McCaney,Mahruq Siddiqui,Gunnar Trommer,Erik Adams,Meghna Eichelberger,and Peter Lawyer Contents03 Introduction04 Key Findings05 Innovation:Medtech AI/ML Is Soaring10 Regulation:FDA Road Signs Are Becoming Clearer

2、15 Investment:Are AI/ML Products on the Road to Riches?23 Travelogue:The Road Ahead24 Summary25 Methodology27 About the AuthorsBOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 3From their origins as hard-coded,embedded software in medical devices and equipment,artificia

3、l intelligence and machine learning(AI/ML)now lay claim to over 1,000 US clearances and approvals.After our first collaboration on medtech innovation and regulatory evolution in 2022 documented the rise of AI/ML solutions in medtech,UCLA Biodesign and BCG set out to understand how the technology is

4、evolving within the sector.Our team examined which specialties are most heavily penetrated,how much time is required to bring products to market,where innovative technology hotbeds are located,and how regulatory regimes are managing the load of new applications.We also sought to understand how much

5、has been spent,what sources are funding the great upsurge in innovation,and how these sources of investment are evolving as AI/ML goes mainstream in medtech.AI/ML appears to be well on its way to achieving mainstream acceptance.In the past seven years alone,AI/ML-enabled device authorizations from t

6、he US Food and Drug Administration(FDA)soared from the single digits to 223 in 2023 alone.Radiology continues to dominate the AI/ML race,but other medical specialties such as cardiology are making great strides as well.Software defines success for todays AI/ML toolsand not just the hard-coded variet

7、y that has powered devices and equipment for decades.Today,such products account for about 24%of authorizations,while three-quarters of todays approved products consist of standalone software and algorithms.At the same time,distinct geographic hotspots for AI/ML-enabled device development are visibl

8、e in the San Francisco Bay Area,Tel Aviv,Seoul,Paris,and Shanghai.The FDA has invested heavily in expertise and capacity to keep the innovation machine humming.Still,the median time to clearance for AI/ML is 25%longer than for non-AI/ML,leaving considerable room for improvement.Judicious use by the

9、FDA of third-party reviewers is helping,but the rate of improvement for AI/ML clearance time pales in comparison to that for non-AI/ML products.One area that has not seen a notable rise is AI/ML-based products that feature adaptive algorithms.To date,we have uncovered just three adaptive-logic produ

10、ct authorizations with an FDA-authorized“Predetermined Change Control Plan.”1 Some$14 billion in venture capital(VC)has paced the development of AI/ML-enabled devices from 2010 through Q3 2024,with 3,057 investors backing 387 companies that have collectively produced about half of the 1,016 AI/ML pr

11、oducts cleared by the FDA.Private capital represents about two-thirds of activity in the space,with public companies responsible for just over one-third of AI/ML products.Venture funding has shifted from seed and early stage to Series C and Series D rounds since 2020,with 16 deals netting over$100 m

12、illion each.The sector has seen nearly 60 exits since 2010,including corporate mergers and acquisitions(M&A),initial public offerings(IPOs),and leveraged buyouts(LBOs)with deal sizes ranging from the single millions to multiple billions.Our work underscores how AI/ML software has come of age in the

13、medtech sector.We are excited to provide this analysis to executives of both startup and well-established medtech companies,as well as to regulators,VC professionals,and academics.We believe that this report represents the most comprehensive compendium of global AI/ML-based product authorizations av

14、ailable to date and provides a solid tally of product funding by source.(See“Methodology.”)We realize that there will inevitably be blind spots in and omissions from our list,but we hope that the insights gleaned from this work will more than make up for any limitations.Introduction1.https:/www.fda.

15、gov/medical-devices/software-medical-device-samd/predetermined-change-control-plans-machine-learning-enabled-medical-devices-guiding-principles.BOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 4Key Findings The period from 2015 through Q3 2024 saw growth of more than 35

16、X in AI/ML-enabled devices,with 1,016 products authorized by the FDA to date.Radiology remains the leading application,as image processing and triage software accounts for three-quarters of all authorizations,but penetration of other specialties is growing.Software-as-a-medical-device(SaMD)sets the

17、standard,accounting for 71%of all AI/ML authorizations.Coding hotspots in the US are responsible for about half of AI/ML authorizations,with Israel,France,China,South Korea,and Japan adding another 27%.The median time to approval for AI/ML-enabled devices took 25%longer(about four weeks more)than fo

18、r non-AI/ML products,despite considerable focus and investment by the FDA to improve capacity and bolster expertise.Third-party reviewers offer a marginal advantage for AI/ML-enabled devices,providing a two-week improvement in time-to-clearance in AI/ML versus a three-month improvement for non-AI/ML

19、 products.Just three AI/ML-enabled devices authorized to date contain adaptive logic with an FDA-approved change control program,and no GenAI-enabled devices have received authorization.Two-thirds of AI/ML products were sponsored by private companies at the time of authorization.Half of AI/ML author

20、ized devices came from VC-backed companies,which have cumulatively invested$14 billion in AI/ML-enabled devices since 2010.A total of 16 VC megadeals(deals exceeding$100 million)for AI/ML companies were completed between 2020 and Q3 2024,versus a total of just 8 in the prior five-year period.Exit ac

21、tivity has increased since 2020,with 41 total exits(31 acquisitions,5 IPOs,and 5 LBO/buyouts)from 2020 through Q3 2024,with a combined value of approximately$11 billion,versus 17 exits in the prior decade from 2010 to 2019.BOSTON CONSULTING GROUPARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 4BOS

22、TON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 5The age of AI/ML in the medtech industry has arrived,with annual authorizations ballooning from single digits in the early part of the past decade to a cumulative 1,016 by Q3 2024,the latest reported data.(See Exhibit 1.)Al

23、though they still account for just a small fraction of annual 510(k)clearances,AI/ML-enabled devices are carving out an important niche in the sector.Since September 2023,a dozen De Novo AI/ML-enabled devices have received FDA approvalversus only 20 during the entire prior decade.(See Exhibit 2.)The

24、 message to the medtech industry is clear:the march toward AI/ML-enabled devices is accelerating and changing the game in this innovation-based market.Radiology Paces the FieldNowhere is this movement more evident than in the field of radiology,which accounts for 75%of all AI/ML authorizations since

25、 2010.The unique attributes of imaging come into play in this field,where computerized algorithms outperform the human eye in looking for patterns in pixels.This holds true across x-ray,computed tomography(CT),ultrasound,endoscopy,and 1,012 other authorizations,all of which involve some form of imag

26、e processing or clinical prioritization and triage.2 The cardiovascular field holds a distant second place,with 70 authorizations in the past five years.Given the similarities between imaging modalities in the cardiac catheterization lab and in radiology suites,the same factors that are driving radi

27、ology toward AI/ML solutions come into play in the cardiovascular space.Nonimaging AI/ML technologysuch as algorithms from multiple companies that detect patterns in heart rhythmshave bolstered the cardiovascular numbers.Other specialties,including neurology,hematology,gastroenterology and urology,a

28、nd ophthalmology have also gained AI/ML approvals.One senior regulatory official commented,“There is similar capability and opportunity in other clinical settings.However,cardiology already has a large number of devices in their space,an organized hospital structure around it,and an urgency to time

29、to treatment.”InnovationMedtech AI/ML Is Soaring2.Total number of authorizations=1,016;technology review excludes four PMA approvals.BOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 6EXHIBIT 1The Rise of AIEXHIBIT 2Regulatory Pathways and FDA Review Times for AI/ML Medt

30、ech DevicesSource:US Food and Drug Administration.Sources:US Food and Drug Administration data;analysis performed by BCG and UCLA Biodesign.Exponential increase in AI/ML approvals,20102024More than 35X increase in AI/ML devices since 2015Number of AI-enabled medical devices authorized by the FDA3180

31、5010015020025022010220112012320136201462015201626201764201880201911420201292021160202222320231692024?35X growthNot shown:11 devices authorizedduring the period 19952010ActualsQ4 2024 data pending FDA reportingMore than 35X increase in AI/ML devices since 2015Source:US Food and Drug Administration.Th

32、e Rise of AIEXHIBIT 1Which regulatory pathways?FDA review timesMost approved AI/ML devices obtain 510(k)clearance within 5 months of FDA submissionDe Novo510(k)PremarketApproval4.4 monthsMedian10 months12 months5 monthsAverage12 months11 months432980510(k)De NovoPremarket ApprovalN=980N=32N=4Sources

33、:US Food and Drug Administration data;analysis performed by BCG and UCLA Biodesign.Regulatory Pathways and FDA Review Times for AI/MLMedtech DevicesEXHIBIT 2BOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 7The Emerging Roadmap for AI/ML DevicesOver the past five years,

34、companies commercializing AI/ML technologies in these specialties appear to be following similar principles.They prioritize software over hardware,integrate the software into existing imaging,monitoring,or surgical planning systems and diagnostic platforms,and use AI to integrate multimodal data acr

35、oss various inputs(such as electrical activity sensors,physical sensors,images,and specimens).The AI/ML logic typically performs one or more functions:Automating high-labor tasks that predict or detect acute events Providing surgical planning or real-time procedural guidance Serving as a decision-su

36、pport tools for diagnostic aidsCategorized by purpose,AI/ML-enabled devices fall into six main classes.(See Exhibit 3.)EXHIBIT 3Emerging Classes of AI/ML Devices in MedtechSource:BCG and UCLA Biodesign.Note:AR=augmented reality;CAD/CAM=computer-aided design/computer-aided manufacturing;EEG=electroen

37、cephalography;EHR=electronic health record;Gi=gastrointestinal;OCT=optical coherence tomography;PSG=polysomnography.Device purposeOpportunityData or inputRole of the AIIntegrationSpecialtiesAutomatedimageanalysis Speed upreview;improveearly detection Digital images(e.g.,fundus,OCT,slides)Detection;s

38、egmentation;classification Software add-onwith imagingsystems Radiology;pathology;GI;dental Signal or eventdetectionAutomatetime-seriesreviewSensor data(EEG,PSG,respiratory,acoustics)Patterndetection;alertsEmbeddedsoftware inmonitorsCardiology;neurology;sleep;anesthesiologyReal-timenavigation Enhanc

39、eproceduralprecisionLive imagingand trackingRegistration;AR overlays(multimodal)Softwareintegrated withsurgical toolsNeurosurgery;orthopedics;dentalRisk predictionForecast riskprogressionMultimodalclinical dataRisk scoring;tracking andpredictionStandalonesoftware/EHRmodulesCardiology;neurology;hospi

40、tal;labsPersonalized procedure planningTailorinterventionsto patientspecificsPre-op imaging;3D scansSimulation;panning Interfaces withCAD/CAMand roboticsOrthopedics;dental;neurosurgeryAdjunctive decentralizeddiagnosticsPerform rapidscreening indecentralizedsettings(point of need)Images;sensors;manua

41、l input;specimenAdaptiveclassificationPortablehome devices;smartphone appsOphthalmology;anesthesiology;labsAI/ML-enableddevice examplesGI Genius;IDx-DR;Paige Prostate;Dental Monitoring encevis;autoscore;EnsoSleep,Tyto InsightsClearPointSystem;X-Guide;Precision AIPlanning BrainSee;SepsisImmunoScore;K

42、idneyIntelX;Acumen HPIPrecision AISurgical PlanningSystem;UnitedOrthopedic KneePatient SpecificInstrumentationNotal HomeOCT;DreaMedAdvisor;TytoCareCrackle Detection;Healthy.ioSource:BCG and UCLA Biodesign.Note:AR=augmented reality;CAD/CAM=computer-aided design/computer-aided manufacturing;EEG=electr

43、oencephalography;EHR=electronic health record;Gi=gastrointestinal;OCT=optical coherence tomography;PSG=polysomnography.Emerging Classes of AI/ML Devices in MedtechEXHIBIT 3BOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 8Software Is the New DeviceWhen it comes to the t

44、ypes of AI/ML-enabled devices currently receiving FDA clearance,we see another profound skewthis time toward standalone software,more commonly referred to as software as a medical device(SaMD).SaMD represented 699 of the 980 510k clearances and 22 of the 32 De Novo AI/ML clearances that occurred in

45、the past five years.Whether software qualifies as SaMD depends on whether it satisfies two criteria:The software provides a patient benefit,and a malfunction of the technology would present a risk to patients.The software can operate independently of other machinery and equipment.If both conditions

46、hold,the product is considered SaMD.If it does not meet the second condition,the technology is considered software in a medical device(SiMD),and the device itself(rather than the software)must receive FDA authorization.3Much of the remainder of AI/ML clearances and approvals(27%)combine software and

47、 hardware,often in the form of precoded software loaded onto a dedicated platform.Some of the earliest AI/ML successes have involved capital equipment,frequently leveraging a machines memory cache with intelligent software to interpret new data and assist with diagnosis and treatment.Examples range

48、from ultraportable imaging solutions to real-time invasive imaging to procedural guidance capital equipment systems with recent devices.Hyperfines The Swoop ultra-low field magnetic resonance portable bedside MRI system,for instance,expands access to high-quality brain imaging across multiple settin

49、gs of care,and Medtronics GI Genius is the first intelligent computer-aided endoscopy system that can detect polyps in real-time to accelerate time to detection,reduce diagnostic variability,and improve accuracy of diagnosing colorectal cancer.AI/ML Is Still Embryonic in In Vitro DiagnosticsThough o

50、nly 22 AI/ML clearances and approvals in our sample came from the in vitro diagnostics(IVD)space,these products share two unique traits:They can miniaturize lab-scale diagnostics.They deliver lab-quality results at the point of care or need.AI has transformed large,high-throughput lab platforms acro

51、ss hematology,immunology,and microbiology into portable benchtop devices,thereby enabling immediate,affordable,accurate tests for cancer,chronic diseases,and infections,even as over-the-counter options.This shift spans the diagnostic chain for IVD from lab to hospital to home.For example,traditional

52、 players such as CellaVision have adapted their digital platforms(for example,the CellaVision DC-1 analyzer)for use in small,independent labs and distributed networks.More recently,startups such as Israels Healthy.io have leveraged smartphone cameras and colorimetric detection to perform at-home tes

53、ts for conditions such as urinary tract infections,renal health,and wound care,allowing patients to report results via telemedicine and receive prescriptions without visiting a centralized lab or waiting a week for mail-in lab results.Medtech AI/ML Coding Hotspots Are Popping UpGiven the prominence

54、of software in AI/ML-enabled devices,innovation tends to cluster in recognized coding hotspots around the world.(See Exhibit 4.)California spikes as the most prominent development center,accounting for 13%of the global total,and the San Francisco Bay Area has contributed 95 of the Golden States 127

55、AI/ML successes.Across the country,other areas that have achieved a strong showing for AI/ML clearances tend to be global or regional headquarters and R&D center locations for major players in the diagnostic imaging field such as GE in Wisconsin,Siemens in Pennsylvania,Philips in Tennessee,and Canon

56、 in California.Outside the US,five nationsIsrael,France,China,South Korea,and Japanaccount for about 27%of all AI/ML clearances and approvals.Israels contribution is especially notable,with Tel Avivbased Aidoc and Zebra Medical Vision combining for 34 of the countrys 81 US AI/ML authorizations.An Is

57、raeli medtech AI leader explained why his country was able to punch above its weight:“Young adults get brought into the military and are trained as data scientists.You get a great cohort of people who were trained.before even starting at university.Also,a lot of people are passionate about health ca

58、re.You want to connect to a higher purpose and be doing something meaningful.This mindset is ingrained in us.”Some cities outside the US have a diverse collection of companies that produce AI/ML devices(Seoul,for example,hosts 14 companies that are responsible for 29 devices);others tend to center o

59、n the overseas headquarters of global medical device companies,including United-Imaging in Shanghai,China;Canon Medical Systems in Japan;and Samsung in South Korea.3.Bernhard Knappe et al.,“Software as a Medical Device(SaMD):What It Is and Why it Matters,”Orthogonal,October 8,2024.BOSTON CONSULTING

60、GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 9EXHIBIT 4Global AI HotspotsSources:US Food and Drug Administration data;BCG and UCLA Biodesign.The US leads in number of AI devices(49%all time);Israel leads outside the US;South Korea has entered the top 5Canada(37)Sweden(20)UK(31)France

61、(53)Israel(81)Netherlands(29)Japan(44)South Korea(47)China(50)US(496)1,016 AI/ML developed worldwide 49%(496)originated in the US 10%(127)originated in California Israel leads outside the US with 81 AI/ML(8%)South Korea knocked Japan out of the top 5 producers of AI/ML as of 2023 Germany ranks 12th(

62、18 total)but skyrocketed with 10 in 2023 aloneInsightsNumber of AI/ML devicesPercentage of total(%)US49649Israel818France535China505South Korea475Japan444Canada374UK313Netherlands293Sweden202Rest of the world 12813Top 10 AI/ML hubs(as of September 2024)Sources:US Food and Drug Administration data;BC

63、G and UCLA Biodesign.Global AI HotspotsEXHIBIT 4BOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 10Medtechs digital age began just before the new millennium,with the confluence of three events:the FDA issued its first quality guidelines for software;the agency created t

64、he De Novo approval pathway for Class I and Class II devices with no predicate;and the first AI/ML-enabled device received clearance.The product that ushered in the AI/ML era was a device called PAPNET Testing System,which in 1995 received Premarket Approval(PMA)to detect abnormalities and lesions m

65、issed during manual microscopic examination of pap smears.4 Over the next decade,regulators from around the world came together to hash out SaMD guidelines and definitions,and the FDA created the Breakthrough Device program.By 2017,some 50 standalone AI software products had received FDA authorizati

66、on,setting the stage for the agency to promulgate a regulatory framework for updates and modifications of cleared AI/ML products.In the fall of 2024,the FDA convened a Digital Health Advisory Committee to explore regulatory measures for AI/ML-based SaMD and generative AI(GenAI)developmentthe goal be

67、ing a protocol that“outlines a holistic approach to total product life cycle oversight to further the enormous potential that these technologies have to improve patient care while delivering safe and effective software functionality that improves the quality of care that patients receive.”5Regulatio

68、nFDA Road Signs Are Becoming Clearer4.www.accessdata.fda.gov/cdrh_docs/pdf/p940029.pdf.5.US Food&Drug Administration,Center for Device and Radiological Health.Artificial Intelligence/Machine Learning AI/ML-Based Software as a Medical Device SaMD Action Plan.January 2021.BOSTON CONSULTING GROUP +UCLA

69、ARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 11Medtech Executives Are Optimistic,but Wary of RoadblocksWith the total number of AI/ML-based devices now exceeding 1,000,the FDA continues to update its guidelines for total life cycle management of AI/ML-enabled devices.Medtech companies navigatin

70、g this space approach the topic with understandable caution.A UCLA Biodesign/BCG survey of 52 medtech executives and directors showed that eight out of ten respondents rated FDA regulatory requirements as a critical challenge complicating efforts to bring their innovations to market.(See Exhibit 5.)

71、At the top of their list for AI/ML was the issue of ensuring data privacy and security.As one R&D leader summed it up,“Data privacy and security becomes murky with some AI devices,and its a huge barrier.”The FDAs efforts to establish a predictable regulatory environment that prioritizes upholding pa

72、tient safety while allowing innovation to flourish are numerous and noteworthy.AI/ML is one of 20 ongoing regulatory research programs at the agency whose goal is to create the least burdensome comprehensive evaluation of safety and effectiveness of these products.Currently,the program focuses on me

73、thods and metrics for training algorithms and testing data;minimizing bias;establishing standards for gauging the performance,safety,and effectiveness of continuously learning algorithms;assessing emerging clinical applications;and monitoring postmarket effects.In addition to addressing its immediat

74、e regulatory challenges,the FDA must also grapple with harmonizing its system with international regulatory regimes.6Radar Gun:Clocking Submission-to-Authorization Time for AI/MLThe UCLA Biodesign/BCG team evaluated 1,016 successful AI/ML FDA authorizations received by 387 companies through Q3 2024.

75、The team noted the approval type and number of products(980 510(k)clearances,32 De Novo,and 4 PMA),time to first approval,and time to subsequent approvals.From this group,the team generated a comparative sample of 659 510(k)cleared AI/ML-enabled devices as well as a matching control group of 6,347 s

76、tandard(that is,not AI/ML)510(k)cleared devices,basing the matches on products sharing the same product code classification(for example,product code LLZ=System,Image Processing,Radiological).The team compared the times from FDA submission to authorization for AI/ML-enabled devices and for standard d

77、evices through 2023(the last complete year of reported authorizations).EXHIBIT 5Challenges That Medtech Companies FaceSource:BCG and UCLA Biodesign.Regulations,data privacy,clinical efficacy,and adoption are the biggest challenges for AI/ML devicesQ.What are some of the main challenges you face in b

78、ringing AI/ML-enabled medical devices to market?80605650484842383228Navigating regulatory requirementsEnsuring data privacy and securityDemonstrating clinical efficacy and safetyGaining clinician acceptance and adoptionDemonstrating financial valueTransitioning current products toAI/ML-enabled capab

79、ilitiesIntegrating with existing health care IT systemsSurvey respondents(%)Securing funding or investmentGetting to reimbursement/coverageManaging and analyzing large data setsThere is frustration about the standards,but its understandable.FDA experts are smart,and they want to collaborate on AI de

80、vices,but they realized they must bring in more AI and ML expert reviewers and now there are more stakeholders that you must satisfy to get approval.“CTO at an international medtech companyData privacy and security becomes murky with some AI devices,and its a huge barrier.FDA gets very nervous if yo

81、ur device retains the data for any reason such as self-learning.“R&D lead and director at aninternational medtech companyCustomer adoption is a major challenge because every user will do their own personal verification.They wont just get excited if we say its AI-enabledthat era has passed.In the end

82、,clinicians only look at the outcomes and assess if the device actually makes their job easier.“President and CEO at aninternational medtech companySource:BCG and UCLA Biodesign.Challenges That Medtech Companies FaceEXHIBIT 56.Christian Johnson et al.,Interstates and Autobahns:Global Medtech Innovat

83、ion in the Digital Age.Boston:Boston Consulting Group and UCLA Biodesign Center,March 2022.BOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 12Despite the difficulties outlined above,AI/ML-enabled devices do not appear to suffer an undue regulatory penalty in the US.(See

84、 Exhibit 6.)Median clearance time for AI/ML-enabled devices ran about 25%approximately four weekslonger than standard devices.Foreign applicants for AI/ML product authorization fared slightly better than their domestic counterparts,with median clearance times of 129 days for foreign applicants and 1

85、35 days for US applicants.In contrast,US applicants for standard devices received authorization roughly two weeks sooner than foreign companies did.Company and Regulatory Experience Affect Authorization TimingNot surprisingly,median clearance time for AI/ML-enabled versus standard medical devices ru

86、ns longer for first-time approvals than for subsequent successes.(See Exhibit 7.)Median first-time AI/ML clearances took about five weeks longer than median first-time clearances for standard products,but the delay for subsequent authorization dropped to just 22 days.The learning curve for companies

87、 submitting applications and for the regulators reviewing their submissions most likely accounts for the speedier timeline on subsequent applications.The most significant regulatory gap occurred in the performance of third-party reviewers.In principle,third-party reviewers bring subject matter exper

88、tise and much-needed capacity to the task of helping the FDA manage its regulatory workload.The relative performance,as measured by time to authorization,is quite stark for non-AI/ML approvals.Third-party reviewers authorize standard products in a median time of just 29 days versus 114 for their FDA

89、 staff counterparts.However,the performance differential drops sharply for AI/ML-enabled products,where median time to authorization required 115 days for third-party reviewers versus 134 days for FDA staff.These findings raise an obvious question:can industry expect median clearance time to improve

90、 significantly as more third-party reviewers come on board and gain familiarity with AI/ML regulatory frameworks?The chief technology officer of one international medtech company expressed his hope for exactly that outcome as aggregate experience grows,saying,“There is frustration about the standard

91、s,but its understandable.FDA.realized that they must bring in more AI and ML expert reviewers,and now there are more stakeholders that you must satisfy to get approval.”But Specific Product Type Does Not Matter as MuchThe type of AI/ML product under review plays only a minor role in median approval

92、time.(See Exhibit 8.)The median time to clearance for AI/ML products in the sample was 133 days versus 106 for standard products.Interestingly,Pulsed Doppler Ultrasound applicantsboth AI/ML and standardsignificantly outperformed these benchmarks at 98 and 67 days,respectively.Median AI/ML clearance

93、time across five other categories ranged from 112 to 140 days versus non-AI/ML performance of 91 to 129 days.The remainder of our sample clocked in at a median of 146 days for AI/ML versus 117 days for their standard product comparators.EXHIBIT 6Statistical Analysis:510(k)Clearance Timelines by Geog

94、raphySource:UCLA Biodesign.Note:IQR=interquartile range.1A p-value of less than 0.05 is considered statistically significant.510(k)AI/ML clearancesNMean(standarddeviation)Median(IQR)Non-AI/ML clearancesOverallUSNon-US659152.9(94.6)133.0(87.0209.0)318151.8(89.7)135.5(89.0202.0)341153.9(99.1)129.0(86.

95、0210.0)Np-value1Mean(standarddeviation)Median(IQR)6,347136.2(112.4)106.0(56.0186.0)3,558129.2(106.3)100.0(55.0171.0)2,789145.2(119.2)115.0(57.0204.0)0.00010.00010.0014Overall summary:AI/ML devices4 weeks(27 days)longer for approval of AI/ML devices than non-AI/ML devicesGeographic(US vs.non-US)5 wee

96、ks longer in the US for approval of AI/ML devices than non-AI/ML devices6 days longer for approval of non-US AI/ML devices than of US AI/ML devicesSource:UCLA Biodesign.Note:IQR=interquartile range.1A p-value of less than 0.05 is considered statistically significant.Statistical Analysis:Regulatory C

97、learance TimelinesEXHIBIT 6BOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 13EXHIBIT 7Statistical Analysis:510(k)Clearance Timelines by ExperienceEXHIBIT 8Statistical Analysis:510(k)Clearance Timelines by Product TypeSource:UCLA Biodesign.Note:IQR=interquartile range.1

98、A p-value of less than 0.05 is considered statistically significant.Source:UCLA Biodesign.Note:CT=computed tomography;IQR=interquartile range;MRI=magnetic resonance imaging.1A p-value of less than 0.05 is considered statistically significant.510(k)AI/ML clearancesNMean(standarddeviation)Median(IQR)N

99、on-AI/ML clearancesFirstapprovalsSubsequentapprovalsThird-partyreviewerNo third-partyreviewer146189.7(93.4)186.5(120.0252.0)513142.4(92.4)121.0(79.0191.0)56129.3(106.1)115.5(47.5165.0)Np-value1Mean(standarddeviation)Median(IQR)1,083179.5(126.4)152.0(86.0254.0)5,264127.3(107.2)99.0(52.0168.0)48149.4(

100、52.7)29.0(21.056.0)0.02020.00010.0001603155.0(93.3)134.0(87.0212.0)5,866143.3(113.1)114.0(61.0195.0)0.0001First approvals:AI/ML devices5 weeks longer than non-AI/ML devicesSubsequent approvals:AI/ML devices2.5 weeks longer than non-AIML devicesThird-party review:AI/ML devices 2.5 weeks longer with n

101、o third-party reviewer within AI/ML group5 months longer(144 days)than non-AIML devices with third-party reviewer3 weeks longer with no third-party reviewer than non-AIML devicesSource:UCLA Biodesign.Note:IQR=interquartile range.1A p-value of less than 0.05 is considered statistically significant.St

102、atistical Analysis:Regulatory Clearance TimelinesEXHIBIT 7510(k)AI/ML clearancesNon-AI/ML clearancesProduct typePulsedDoppler,ultrasoundX-ray,CTImageprocessing,radiologyNuclearMRIComputer-assistedtriage notificationsoftwareComputer-assistedprioritizationsoftware for lesionsAutomatedimage processing,

103、radiologyOtherN395712534402779258ProductcodeIYNJAKLLZLNHQASQFMQIHOtherMean(standarddeviation)106.6(48.2)147.7(85.1)141.3(101.6)140.7(89.5)122.7(83.0)159.3(85.6)171.4(114.0)166.5(91.7)Median(IQR)98.0(84.0139.0)140.0(81.0227.0)118.0(71.0190.0)131.5(60.0188.0)111.5(74.5160.0)137.0(92.0214.0)140.0(101.0

104、231.0)146.0(96.0232.0)N6562801,000280103134,105Mean(standarddeviation)87.1(73.3)144.8(96.4)124.8(104.4)113.2(87.6)96.5(41.3)223.7(95.1)154.6(82.4)147.8(119.4)Median(IQR)67.5(32.5113.0)129.0(77.5189.0)91.0(50.0170.0)91.0(57.0141.0)97.0(72.0119.0)198.0(144.0329.0)123.0(92.0254.0)117.0(60.0205.0)p-valu

105、e10.00160.59010.00730.04310.53900.8756$1 billion)20172019 1.United-Imaging:$5 billion2.Tempus AI:$3.1 billion3.23&Me:$2.75 billion20202022 1.Tempus AI:$10 billion2.23&Me:$2.8 billion3.Athelas:$1.5 billion2023+1(no unicorns)1.Paige:$650 million2.Rapid AI:$600 million3.Overjet:$550 million$0$100$200$3

106、00$400$500$600201020112012201320142015201620172018201920202021202223&Me23&Me;United-Imaging23&Me;Butterfly Net;Heartflow;Tempus AIHeartflow;Tempus AI23&Me;Tempus AIShakunAthelas;Biofourmis;Viz.ai;Tempus AI20232024$10Averagedeal size($millions)$4$7$4$10$2$3$1$10$2$7$2$6$2$18$5$18$3$13$4$22$8Mediandea

107、l size($millions)$26$14$35$14$20$8$17$10Sources:BCG and UCLA Biodesign;FDA;Pitchbook,January 2025.Note:23&Me first reached unicorn status in 2015;Butterfly Net,Heartflow,and Tempus AI first reached unicorn status in 2018.1Valuation data incomplete for 2023 and later is incomplete for devices not yet

108、 submitted for FDA clearance.Average Postdeal Valuation vs.Deal SizeBOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 22EXHIBIT 16Exit ActivitySources:BCG and UCLA Biodesign;FDA;Pitchbook,January 2025.Note:IPO=initial public offering,LBO=leveraged buyout;M&A=mergers and

109、acquisitions.M&A activity was strong during the pandemicand has remained strong afterward After 2017,a rising wave of exits accompanied thesurge in AI/ML medical devices cleared by the FDA024681012TypeYearCompanyAmountIPO20172019PrediLife$4.08 million20202022Spectral AI$15.6 millionLunit$28 millionU

110、nited Imaging$1.6 billion2023+Ceribell$180 millionAcquirerBuyout/LBO 20172019The Esaote Group$354 millionMultiple20202022Therenva Ziehm ImagingScanMed$20 millionDirectMed Imaging2023+RadFormation$6.9 millionBVP ForgeLimbus AIRadFormationSonio$92 millionSamsung Medison20172019Behold.aiSimon Raslingha

111、mWithingsMr.Eric CarreelQuantitative InsightsQlarity ImagingM&A20202022(21 total)Excel Medical$19.2 millionHill Rom HoldingsPreventice$936 millionBoston Scientific7D Surgical$120 millionSea SpineGauss$160 millionStrykerZebra Medical Vision$110 millionNanoxBK Medical Holding$1.4 billionGE HealthcareC

112、athworks$585 millionMedtronic2023+(12 total)Caption Care$150 millionGE HealthcareDIA Imaging Analysis$100 millionPhilipsAthela$6 billionCommuneVolpara Health$292 millionLunit20101120111201212013220141201522016112017120181122019120201420211920221262023152024Q1Q3227Buyout/LBOIPOM&ASources:BCG and UCLA

113、 Biodesign;FDA;Pitchbook,January 2025.Note:IPO=initial public offering,LBO=leveraged buyout;M&A=mergers and acquisitions.Exit ActivityEXHIBIT 16More VC Investors Are Heading to the Exit RampFor VC investors,the ultimate indicator of success is a great exit.From 2010 to 2023,Pitchbook tallied 58 medt

114、ech AI/ML exits representing paydays for VC investors.(See Exhibit 16.)Large public companies such as GE Healthcare,Philips,Medtronic,Stryker,Boston Scientific,and Hill-Rom accounted for the outright majority of buyers,but the largest individual deal was the$6 billion takeover of Athelas by Commune

115、in October 2023.In addition to M&A,VC players chalked up exits with seven LBOs and four IPOs involving companies with new AI/ML-enabled devices.Strong exit activity for AI/ML-related products suggests sustained confidence in the technologys transformative potentiala sentiment underscored by the rece

116、nt UCLA Biodesign/BCG survey of 50 medtech executives and board directors.Two-thirds of respondents said they were somewhat more or much more positive about the outlook for AI/ML-enabled devices than they had been just 12 months prior.None claimed to be much more negative,and just 10%indicated that

117、their outlook for the technology was slightly more pessimistic.As the president and CEO of one medtech company put it,“My confidence has grown because of the awareness in the market.I was startled at how quickly the industry adopted AI devices.”BOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE ST

118、AKES A CLAIM ON MEDTECH 23You Are Now Arriving at the Future of AI/MLAs a technology,AI is still in its infancy,as ongoing development and breakthroughs are announced daily.With just over 1,000 devices authorized for the US market cumulatively,AI/ML-enabled products are a tiny fraction of the 3,000

119、or so products that the FDA clears each year across 510k,De Novo,PMA,and other pathways.As experience with AI/ML grows,so will the backlog of marketed devices.How these products perform will determine whether early promises of greater efficiency,clinical improvement,and patient satisfaction come to

120、fruitionalong with the attendant financial rewards for the companies that bring these products to market.Road Hazards:Data Privacy and Ownership Rights AI/ML technology requires access to underlying patient and clinical data,raising thorny issues about what patients,clinics,and companies can claim a

121、s their own.There is an obvious need for de-identified patient data with clear usage rights established,but there is as yet no straightforward way to ensure its provenance.The problem compounds as more AI/ML-enabled devices hit the market because so many of them generate their own data stream.Compan

122、ies developing AI/ML-enabled devices need to be able to define where their data came from,how they obtained it,and how they used it.Otherwise,they may face downstream legal battle if their inventions are challenged in court.Road Construction:A New BillA Senate bill proposed by Democratic Senator Mat

123、t Heinrich of New Mexico and Republican Senator Marsha Blackburn of Tennessee would offer a reimbursement pathway for health care services provided by AI/ML-enabled devices.8 The senators hope to promote widespread adoption of AI devices in the clinic,resulting in better patient outcomes and greater

124、 efficiency.Current Centers for Medicare and Medicaid Services(CMS)guidelines do not stipulate standard or consistent billing procedures for providers using AI/ML algorithms.The proposed bill would create an Ambulatory Payment Classification(APC)for a period of five years,allowing CMS to determine w

125、hether to assign a permanent code.The bill comes during a period of tremendous upheaval and uncertainty in the federal government,however,and whether it will gain traction remains to be seen.International Road Signs Are HazyInternational regulatory harmonization is always a challenge,with different

126、schools of thought on how to regulate AI on display across the US,the EU,and China.As described by Anu Bradford in Foreign Affairs,the US can be characterized as taking a pro-business approach,versus a privacy-first paradigm in the EU and more reliance on government control of content in China.9 The

127、se differences will be difficult to resolve,especially for smaller,VC-backed companies,resulting in delays or outright forfeiture of overseas markets.Time will tell whether local variation will suffice or whether independent new product development processes will be required to access different mark

128、ets.Adaptive Logic and GenAI:From Side Trip to Main RouteThe potential for adaptive logic to power broad clinical improvements and individualized patient therapy is a lofty goal worth pursuing.The technological hill to climb is steep,but regulatory issues are even more forbidding.With proper guideli

129、nes(for example,clean training data,clear boundaries,and fail-safes)to prevent AI hallucination,greater experience with PCCPs,and a fit-for-purpose postmarketing surveillance process,we anticipate that AI/ML-enabled devices with adaptive logic will provide the next phase of growth for this exciting

130、technology.As for GenAI,the hill it has to climb is even steeper,but one avenue that appears quite promising is the use of synthetic data to train large language models.TravelogueThe Road Ahead8.Health Tech Investment Act.S.Bill,119th Congress,2025.9.Anu Bradford,“The Race to Artificial Intelligence

131、”Foreign Affairs,June 27,2023.BOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 24Although the first FDA-recognized device was launched in 1995,AI/ML-enabled devices did not become commonplace until the early 2020s.Claiming 1,016 authorized products by Q3 2024,AI/ML devi

132、ces are carving out an increasing share of the medtech landscape.More importantly,this technology is changing how innovation occurs,how regulatory bodies operate,and how funding flows into the medtech sector.Innovation remains heavily concentrated in the radiology sector,where image processing techn

133、ology has been a boon to clinicians in interpreting patterns in microscopic data.Other specialties are gaining headway,too,especially cardiology,given its heavy reliance on image,neurological,and acoustic processing.Occupying center stage in all AI/ML innovation is software.SaMD accounts for 71%of a

134、uthorizations to date,with software/hardware combinations responsible for the balance.For this reason,the innovators behind AI/ML-enabled devices tend to be heavily concentrated in coding hotspots in the US,Israel,France,China,and South Korea.The FDA has continuously invested and adapted to accommod

135、ate the rapid influx of new AI/ML-enabled devices.Throughout the past decade,the agency has brought on staff and engaged third-party reviewers with deep software expertiseand as recently as January 2025,the FDA issued additional guidance for total life cycle management of AI/ML-enabled devices.As a

136、result,AI/ML-enabled products now experience only a four-week penalty in median time to authorization versus standard products(133 days versus 106 days).US applicants lag slightly behind foreign players in median time to approval(135.5 days versus 129 days).The use of third-party reviewers,which col

137、lapses median time to authorization for standard products from 114 to 29 days,has less impact on AI/ML-enabled devices,which experience a decrease of just 26 days(155 days versus 129 days).Another area where regulation is progressing slowly is adaptive learning.To date,just 3 of the 1,016 authorized

138、 AI/ML-enabled devices contain self-learning algorithms.Still,medtech executives are excited about the prospects for their AI/ML-enabled devices,and 96%describe themselves as at least somewhat confident about their likelihood of receiving FDA approval and achieving successful commercialization.Their

139、 confidence is echoed by the trajectory of AI/ML-enabled device funding,which has skyrocketed since 2010.VC leads the way with$14 billion raised across 3,057 unique investors in 387 companies.Early-stage investments in the single millions characterized the early 2010s,but more recently VC has funded

140、 a string of megadeals exceeding$100 millionthe largest of which attracted$506 million.Although the absolute number of VC deals has fallen from a high point in 2022,absolute investment remains close to prepandemic numbers.Public corporations account for 40%of the AI/ML-enabled devices on the market,

141、with large imaging companies setting the pace for the entire industry.Meanwhile,strong exits continue to power interest in the space,as the five largest deals(M&A and IPO)netted a collective$10.6 billion for their creators.We believe that this report includes the most comprehensive and current infor

142、mation on approval and funding trends for AI/ML-enabled devices.In such a fast-moving space,there will undoubtedly be errors and omissions in our data set.Nonetheless,we hope that our work will guide medtech innovators,regulators,and investors as they chart their course in the AI/ML-enabled device l

143、andscape.We will continue to monitor progress in this dynamic field,which has already changed the face of the medtech industry and promises to make an even deeper impact in the years to come.SummaryBOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 25In this first-of-a-ki

144、nd study,we examined the AI/ML-enabled medical device landscape in the US from 1995 to 2024.Our team compared successful authorizations for AI/ML-enabled devices and comparable products to understand overall growth and penetration trends,time from submission to authorization,and potential drivers of

145、 variance.In addition,we identified aggregate investment levels for AI/ML-enabled medical technologies,classified investor types,and examined IPO,M&A,and merger activity for companies developing AI/ML-enabled devices.Analytical Methods1.Creation of Database for FDA-Authorized AI/ML-Enabled Devices.T

146、he UCLA Biodesign team and the UCLA Biostatistics department developed,cleaned,and analyzed a novel proprietary database that merges and integrates product and regulatory data from the FDAs Medical Device database,the FDAs Artificial Intelligence and Machine Learning(AI/ML)-Enabled Medical Devices l

147、ist,Pitchbooks data set on public and private company and capital data,and a multitude of other publicly available and private medical device data sets to curate a total data set covering 1,016 AI/ML-enabled medical devices(including 980 510(k)clearances,32 De Novo clearances,and 4 PMA clearances)fr

148、om 387 companies during the time period from 1995 to Q3 2024when the FDA last released AI/ML data.2.Comparative Analysis of Time to Authorization for AI/ML-Enabled and Standard Devices.UCLA Biodesign started with the recent data set of 1,016 AI/ML-enabled medical devices from 387 companies that it r

149、eleased in Q3 2024.We analyzed this list to uncover overall trends and establish benchmarks for insights including annual FDA market authorizations for 510(k),De Novo,and PMA pathways,average and median times in FDA review for various pathways and medical specialties,and segmentation and characteriz

150、ation of AI/ML-enabled devices by geography,product code,technology type(a categorization established by UCLA Biodesign),medical specialty,and other segments.Then,to compare time to authorization between AI/ML-enabled medical devices and standard(that is,non-AI/ML-enabled)devices within the same pro

151、duct code classification(for example,product code LLZ=System,Image Processing,Radiological),we distilled and matched a study group of 659 AI/ML-enabled medical devices that obtained 510(k)clearance from 2010 to 2023 to that for a control group(N=6,347)of standard(that is,non-AI/ML-enabled)510(k)clea

152、red medical devices within the same product code classifications and time period.3.Aggregate Venture Capital Funding for AI/ML-Enabled Devices and Exit Activity.We filtered and distilled the integrated database to isolate study and control group companies whose funding activity was recorded by Pitch

153、book from 2010 to Q3 2024.The team tabulated capital raised by 133 new ventures from time of incorporation to time of positive FDA decision(510(k)clearance,De Novo granting,or Premarket Approval(PMA).Where available,the team captured funding rounds,the size of each deal,pre-and post-valuations,and o

154、ther funding variables.We used a subset of this group,consisting of 107 companies that recorded their first AI/ML-enabled device authorization,to assess how much capital is required to develop,test,validate,and obtain FDA market authorization for a medical device.In addition,the team compiled availa

155、ble data on VC exit activity,capturing deal size for IPOs,mergers,LBOs,and acquisitions by investor type.4.Medtech Executive Survey.UCLA Biodesign and BCG prepared and conducted a four-question online survey of 52 C-suite-or vice-president-level executives with experience in AI/ML-enabled medical de

156、vice development,regulation,and commercialization.The survey sought to understand these leaders experiences,sentiments,perceived needs and gaps,and recommendations for the advancement of AI/ML-enabled devices.We conducted the survey in November 2024 and supplemented it with a series of in-depth inte

157、rviews to provide additional context for the teams findings.Limitations of the StudyAlthough we believe our approach and analysis provides many useful insights,readers should be aware of some important limitations of the study,including the following:Incomplete Record of AI/ML-Enabled Devices.The FD

158、A updates and publishes information on AI/ML-enabled devices on a quarterly or semiannual basis,sometimes reclassifying authorized products in arrears.As a result,our database may not include all AI/ML-enabled devices in the US market since 1995.Incomplete Record of Venture Capital Investment.Our an

159、alysis of VC funding relies on published totals from Pitchbook,which may or may not be complete.We have explicitly omitted insights on 2024 VC investment when examining year-on-year trends because the FDA-authorized AI/ML-enabled device list has not yet been updated for Q4 2024.Imperfect Classificat

160、ion of Devices.We based our segmentation of technologies on the review of device descriptions and technical descriptions in the 510(k),De Novo,and PMA submissions and summary letters on FDAs website,for which general categorizations may or may not be accurate.MethodologyBOSTON CONSULTING GROUP +UCLA

161、ARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 26 Limited Geographic Analysis.We based our regional distribution of companies on the location of each company at time of regulatory submission,whether it was a private venture,a global headquarters of a multinational company,or a subsidiary or joint

162、 venture of a company.We did not present a geographic distribution based on the parent companys global headquarters.Management of Outliers.We used interquartile range to manage outliers,and we applied a Wilcoxon two-tailed test to compare approval times between the AI/ML and control data sets.Althou

163、gh we obtained averages as part of the study,we calculated and cited medians throughout the discussion to minimize skew and reduce the impact of outliers.Areas for Future StudyThis work sheds light on the speed of innovation,regulatory timeframes,and aggregate VC funding levels for AI/ML-enabled dev

164、ices.However,other intriguing questions would bear further investigation:Device Success Rates.The study did not tabulate total submissions for AI/ML-enabled devices and their comparators and,therefore,provides no guidance on relative success rates.Aggregate Investment in AI/ML-Enabled Devices.This s

165、tudy compiled only private investment from VC,plus capital provided by grants,foundations,downstream mergers,acquisitions,and IPOs.Specifically missing is the cost of developing AI/ML-enabled devices in-house for publicly traded companies and how that cost compares with the cost of developing standa

166、rd devices.Adaptive AI-Enabled and GenAI-Enabled Device Experience.To date,only 3 of the 1,016 FDA-authorized AI/ML-enabled devices include adaptive(self-learning)devices and none include generative(self-creating)AI software.As the industry evolves along these vectors,future studies could document t

167、he innovation process,regulatory experience,and cost to bring such technology to market.BOSTON CONSULTING GROUP +UCLAARTIFICIAL INTELLIGENCE STAKES A CLAIM ON MEDTECH 27About the Authors Christian Johnson is founder and CEO of ScoutMedix,a strategy and advisory consulting firm focused on the medtech

168、 industry,and a frequent collaborator with UCLA Biodesign.You may contact him by email at .Mahruq Siddiqui is a principal at M3S Labs,advising medtech and life sciences companies on digital technology strategy and implementation.She is currently collaborating with UCLA Biodesign on an AI/ML medtech

169、research study.You may contact her by email at .Erik Adams is a managing director and partner in BCGs San Diego office.You may contact him by email at .Peter Lawyer is a senior advisor to BCG,a former senior partner and managing director with the firm,and the founder of its medtech sector.You may co

170、ntact him by email at .Jennifer McCaney is chief innovation officer at UCLA Health,executive director of UCLA Biodesign,and an adjunct associate professor at the UCLA David Geffen School of Medicine and Anderson School of Management.You may contact her by email at jmccaneyanderson.ucla.edu.Gunnar Tr

171、ommer is a managing director and partner in the Manhattan Beach office of Boston Consulting Group.He serves as health care leader for BCG X in North America and as global medtech leader for BCG X.You may contact him by email at .Meghna Eichelberger is a partner and associate director for medical tec

172、hnologies in BCGs Boston office.She is topic leader for clinical,market access,and regulatory issues for BCGs medtech sector.You may contact her by email at .For Further ContactIf you would like to discuss this report,please contact the authors.AcknowledgmentsThe authors would like to thank Kalyn Ri

173、cciuti(medtech knowledge analyst,BCG,Boston),Matt Sternberg(lead product manager,BCG X,Chicago),Nami Rokhgar(project leader,BCG,Los Angeles),and Jeff Dean(principal,BCG,San Diego)for their analytical work in support of this effort.In addition,we thank Kerri Miller and Miriam Mburu of BCG global mark

174、eting for their publication assistance.UCLA UCLA Biodesign is a healthcare technology innovation hub at the University of California Los Angeles.Uniting stakeholders across the healthcare ecosystem,UCLA Biodesign seeks to transform medicine through the development and translation of novel technologi

175、es.The advancement of industry research and thought leadership are central to UCLA Biodesigns mission.UCLA Biodesign collaborates with industry partners and the medical community to support innovations that will deliver improved value and outcomes to patients worldwide.An annual innovation fellowshi

176、p at UCLA Biodesign supports training and leadership development in collaboration with the UCLA David Geffen School of Medicine,UCLA Anderson School of Management,UCLA Clinical and Translational Science Institute,and UCLA Health.Boston Consulting Group Boston Consulting Group partners with leaders i

177、n business and society to tackle their most important challenges and capture their greatest opportunities.BCG was the pioneer in business strategy when it was founded in 1963.Today,we work closely with clients to embrace a transformational approach aimed at benefiting all stakeholdersempowering orga

178、nizations to grow,build sustainable competitive advantage,and drive positive societal impact.Our diverse,global teams bring deep industry and functional expertise and a range of perspectives that question the status quo and spark change.BCG delivers solutions through leading-edge management consulti

179、ng,technology and design,and corporate and digital ventures.We work in a uniquely collaborative model across the firm and throughout all levels of the client organization,fueled by the goal of helping our clients thrive and enabling them to make the world a better place.For information or permission to reprint,please contact BCG at .To find the latest BCG content and register to receive e-alerts on this topic or others,please visit .Follow Boston Consulting Group on LinkedIn,Facebook and X(formerly Twitter).Boston Consulting Group 2025.All rights reserved.5/25