瑕瑜互见
然而,许多医疗行业的专家担心,基于AI的产品,将无法与这样的评价相提并论。一些医生和消费者权益团体担心,“快速失败,随后修复”(fail fast and fix it later)的方式,会将患者置于风险之中。相关人士认为,监管机构并未采取足够的措施,来确保消费者的安全。
去年在7月,调研结构Gartner公司发布的一份报告中表示,仅在第三季度,医疗人工智能就吸引了16亿美元的风投资金,“几乎处于预期炒作的高峰期……随着现实的考验,可能会陷入泡沫幻灭后的低谷。”
“黑箱”
即便是开发人员,往往也不知道AI系统如何得出结论,这样的学习识别数据模式的AI系统,通常被称为AI“黑箱”。威斯康星大学麦迪逊分校法学与生物伦理学教授Pilar Ossorio博士表示,鉴于AI有许多风险未知,相关的AI医学领域,需要仔细监督。但是,大多数AI器械,并不需要FDA批准。2016年底,美国国会通过的《21世纪治愈法案》(The 21st Century Cures Act)。该法案第3060节,对医用软件监管做出澄清,使多个类型的医疗软件免于接受联邦机构审评,其中包括某些健身应用、电子病历记录,以及可帮助医生做出医疗决策的工具。依据该法案,2019年9月27日,FDA发布依据《21世纪治愈法案》第3060节,对相应的医用软件政策做出变更的指南文件。
缺乏针对性研究?
去年12月,美国国家科学院发布题为《医疗人工智能:希望、炒作、承诺、危险》的报告(Artificial Intelligence in Health Care: The Hope, the Hype, the Promise, the Peril)。依据该报告披露,目前正在使用的约32万种医疗应用程序,尚无这些应用程序能否真正改善健康状况的针对性研究。
但是,西雅图艾伦人工智能研究所(the Allen Institute for AI)首席执行官Oren Etzioni博士表示,人工智能开发商有经济动因,来确保其医疗产品安全。“如果快速失败,意味着很多人会丧生。我认为,我们不需要快速失败。如果发生使用者死亡,或受到严重伤害,包括投资者在内,没有谁会感到轻松。”
近年来,由于允许销售危险医疗器械,FDA受到抨击。根据国际调查记者同盟(International Consortium of Investigative Journalists)披露,在过去十年中,使用存在危险的医疗器械,与8万例死亡和170万例伤害存在关联。这些器械中的很多,通过有争议的510(k)通道流程获批,根据相关规定,只要公司认为与现有器械类似,就可以让公司销售未经临床测试的存在“中度风险”的产品。2011年,美国国家医学科学院委员会得出结论,认为510(k)程序存在根本性缺陷,因此FDA应该将其淘汰,重起炉灶。但取而代之的是,FDA正在使用该流程来批准AI器械。
监管重点:“在于评估公司,而不是产品”
纽约西奈山医院(Mount Sinai Hospital)的医生们,希望借助AI,帮助使用胸部X光片来预测哪些患者有患上肺炎的高风险。尽管该系统可以根据在该医院拍摄的X光片做出准确预测,但使用同样的技术,对其他医院拍摄的图像进行测试,结果却失败了。最终,研究人员意识到,计算机只是学会了辨别该医院在病床边用便携式胸部X机拍摄的片子,与放射科所拍摄X光片之间,存在差别。对于那些病重而无法离开房间的患者,医生倾向于使用便携式胸部X光检查,既然如此,这些患者发生肺部感染的风险更高,也就不足为奇。
本文题图:Pixabay
参考资料
[1] IBM. MD Anderson Taps IBM Watson to Power "Moon Shots" Mission Aimed at Ending Cancer, Starting with Leukemia. Oct 18, 2013. Retrieved Jan 10, 2020 from https://www-03.ibm.com/press/us/en/pressrelease/42214.wss
[2] TMF. The Top 12 Health Chatbots.May 29, 2018. Retrieved Jan 10, 2020 from https://medicalfuturist.com/top-12-health-chatbots/
[3] AliveCor. AliveCor Named No.1 Artificial Intelligence Company in Fast Company’s 2018 Most Innovative Companies Ranking. Feb 20, 2018. Retrieved Jan 10, 2020 from https://www.alivecor.com/press/press_release/alivecor-named-no-1-artificial-intelligence-company-in-fast-companys-2018-most-innovative-companies/
[4] FASTCOMPANY. The World's Most Innovative Companies. Retrieved Jan 10, 2019 https://www.fastcompany.com/most-innovative-companies/2018.
[5] FDA. FDA permits marketing of clinical decision support software for alerting providers of a potential stroke in patients. Feb 13, 2018. Retrieved Feb 14, 2018 from https://www.fda.gov/news-events/press-announcements/fda-permits-marketing-clinical-decision-support-software-alerting-providers-potential-stroke
[6] Michael Walter. Viz.ai artificial intelligence stroke software helping doctors win race against time. 2019. Retrieved Jan 10, 2020 from https://www.radiologybusiness.com/sponsored/22221/topics/artificial-intelligence/vizai-artificial-intelligence-stroke-software
[7] Keane, P.A., Topol, E.J. With an eye to AI and autonomous diagnosis. npj Digital Med 1, 40 (2018) doi:10.1038/s41746-018-0048-y
[8] IDx. Autonomous AI that instantly detects disease. 2018. Retrieved Jan 10, 2020 from https://www.eyediagnosis.co/
[9] Andis Robeznieks. This ophthalmologist is doing health care AI the right way . Aug 8, 2019. Retrieved Jan 10, 2020 from https://www.ama-assn.org/practice-management/digital/ophthalmologist-doing-health-care-ai-right-way
[10] TMF. FDA Approvals for Smart Algorithms in Medicine in One Giant Infographic. June 6, 2019. Retrieved Jan 10, 2020 from https://medicalfuturist.com/fda-approvals-for-algorithms-in-medicine/
[11] Scott Gottlieb. FDA Announces New Steps to Empower Consumers and Advance Digital Healthcare. July 27, 2017. Retrieved July 30, 2017 from https://www.fda.gov/news-events/fda-voices-perspectives-fda-leadership-and-experts/fda-announces-new-steps-empower-consumers-and-advance-digital-healthcare
[12] Scott Gottlieb. Statement from FDA Commissioner Scott Gottlieb, M.D. on steps toward a new, tailored review framework for artificial intelligence-based medical devices. April 2, 2019. Retrieved April 4, 2019 from https://www.fda.gov/news-events/press-announcements/statement-fda-commissioner-scott-gottlieb-md-steps-toward-new-tailored-review-framework-artificial
[13] FDA. Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) - Discussion Paper and Request for Feedback. Nov 5, 2019. Retrieved Nov 6, 2019 from https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-device
[14] FDA. Digital Health Innovation Action Plan. Jul 27, 2017. Retrieved Jul 28, 2017 from https://www.fda.gov/news-events/fda-voices-perspectives-fda-leadership-and-experts/fda-announces-new-steps-empower-consumers-and-advance-digital-healthcare
[15] GARTNER. Interpreting technology hype. Gartner Hype Cycle. 2020. Retrieved Jan 10, 2020 from https://www.gartner.com/en/research/methodologies/gartner-hype-cycle
[16] Panetta, K. 2017. Top Trends in the Gartner Hype Cycle for Emerging Technologies, 2017. Retrieved Jan 10, 2020 from Gartner. https://www.gartner.com/smarterwithgartner/top-trends-in-the-gartner-hype-cycle-for-emerging-technologies-2017/
[17] Wang P, Berzin TM, Glissen Brown Jr., et al. Real-time automatic detection system increases colonoscopic polyp and adenoma detection rates: a prospective randomised controlled study. Gut. 2019 Oct; 68(10):1813-1819. doi: 10.1136/gutjnl-2018-317500.
[18] Ioana A. Cristea, Eli M. Cahan, John P. A. Ioannidis. Stealth research: Lack of peer‐reviewed evidence from healthcare unicorns. Eur J Clin Invest. 2019;49:e13072. https://doi.org/10.1111/eci.13072
[19] Kate Levchuk. Is Black Box Human Better Than Black Box AI? Oct 15, 2018. Is Black Box Human Better Than Black Box AI? Retrieved Oct 20, 2018 from https://www.forbes.com/sites/cognitiveworld/2018/10/15/black-box-human-or-black-box-ai-a-talk-with-kim-larsen-deutsche-telekom/#594f432e1932
[20] GPO. The 21st Century Cures Act. Dec 13, 2016. Retrieved Jan 10, 2020 from https://www.congress.gov/114/plaws/publ255/PLAW-114publ255.pdf
[21] FDA CDRH, CBER. Changes to Existing Medical Software Policies Resulting from Section 3060 of the 21st Century Cures Act. Sept 27, 2019. Retrieved Sept 29, 2019 from https://www.fda.gov/media/109622/download
[22] Matheny, M., S. Thadaney Israni, M. Ahmed, and D. Whicher (Eds). Artificial Intelligence in Health Care: The Hope, the Hype, the Promise, the Peril. NAM Special Publication. Washington, DC: National Academy of Medicine. Dec, 2019. Retrieved Jan 10, 2020 from https://nam.edu/wp-content/uploads/2019/12/AI-in-Health-Care-PREPUB-FINAL.pdf
[23] Theodore T. Lee, Aaron S. Kesselheim. U.S. Food and Drug Administration Precertification Pilot Program for Digital Health Software: Weighing the Benefits and Risks. Ann Intern Med. 2018;168(10):730-732. DOI: 10.7326/M17-2715
[24] FDA. Overview of Device Regulation. Aug 31, 2018. Retrieved Jan 10, 2020 from https://www.fda.gov/medical-devices/device-advice-comprehensive-regulatory-assistance/overview-device-regulation
[25] ICIJ. Everything You Need to Know About the Implant Files. Nov 30, 2018. https://www.icij.org/investigations/implant-files/what-you-need-to-know-about-the-implant-files/
[26] The National Academies of Sciences, Engineering, Medicine. Devices and the Public’s Health: The FDA 510(k) Clearance Process at 35 Years. July 29, 2011. Retrieved Jan 10, 2020 from http://www.nationalacademies.org/hmd/Reports/2011/Medical-Devices-and-the-Publics-Health-The-FDA-510k-Clearance-Process-at-35-Years.aspx
[26]Hwang TJ, Kesselheim AS, Vokinger KN. Lifecycle Regulation of Artificial Intelligence- and Machine Learning-Based Software Devices in Medicine. JAMA. 2019; 322(23):2285-2286. doi:10.1001/jama.2019.16842
[27] FDA. 510(k) Premarket Notification. K173542. Jan 25, 2018. Retrieved Jan 10, 2020 from https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K173542
[28] FDA. 510(k) Premarket Notification. K101749. Aug 26, 2010. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K101749
[29] FDA. Letter to Arterys Inc. ℅ Mr. John Axerio-Cilies. Jan 25, 2018. Retrieved Jan 10, 2020 from https://www.accessdata.fda.gov/cdrh_docs/pdf17/K173542.pdf
[30] Karas Cross. Response to Letter to FDA on regulation of software as medical device. June 19, 2019. Retrieved Jan 10, 2020 from https://www.warren.senate.gov/imo/media/doc/2019.06.19%20Response%20to%20Letter%20to%20FDA%20on%20regulation%20of%20software%20as%20medical%20device.pdf
[31] Ronquillo JG, Zuckerman DM. Software-Related Recalls of Health Information Technology and Other Medical Devices: Implications for FDA Regulation of Digital Health. Milbank Q. 2017;95(3):535–553. doi:10.1111/1468-0009.12278
[32] FDA. FDA permits marketing of artificial intelligence-based device to detect certain diabetes-related eye problems. Apr 12, 2018. Retrieved Jan 10, 2020 from https://www.fda.gov/news-events/press-announcements/fda-permits-marketing-artificial-intelligence-based-device-detect-certain-diabetes-related-eye
[33] Kathleen C. Frasera, Jed A. Meltzerb,Frank Rudzicz. Linguistic Features Identify Alzheimer‘s Disease in Narrative Speech. Journal of Alzheimer’s Disease 49 (2016) 407–422 DOI 10.3233/JAD-150520
[34] Jennifer Couzin-Frankel. Medicine contends with how to use artificial intelligence. Science.
21 Jun 2019: Vol. 364, Issue 6446, pp. 1119-1120 DOI: 10.1126/science.364.6446.1119
[35] John R. Zech , Marcus A. Badgeley , Manway Liu, et al. Variable generalization performance of a deep learning model to detect pneumonia in chest radiographs: A cross-sectional study. PLOS Medicine. Nov 6, 2018. https://doi.org/10.1371/journal.pmed.1002683
[36] American Kidney Fund. Acute kidney injury. 2020. Retrieved Jan 10, 2020 from https://www.kidneyfund.org/kidney-disease/kidney-problems/acute-kidney-injury.html#what_causes_acute_kidney_injury
[37] Mustafa Suleyman, Dominic King. Using AI to give doctors a 48-hour head start on life-threatening illness. Jul 31, 2019. Retrieved Jan 10, 2019 from https://deepmind.com/blog/article/predicting-patient-deterioration
[38] Tomašev, N., Glorot, X., Rae, J.W. et al. A clinically applicable approach to continuous prediction of future acute kidney injury. Nature 572, 116–119 (2019) doi:10.1038/s41586-019-1390-1
[39] Connell, A., Montgomery, H., Martin, P. et al. Evaluation of a digitally-enabled care pathway for acute kidney injury management in hospital emergency admissions. npj Digit. Med. 2, 67 (2019) doi:10.1038/s41746-019-0100-6
[40] Sauranbh Jha. The Rise and Rise of Quantitative Cassandras. Oct 7, 2019.Retrieved Jan 10, 2019 from https://thehealthcareblog.com/blog/2019/10/07/the-rise-and-rise-of-quantitative-cassandras/
[41] Lynne Shallcross.Artificial intelligence has come to medicine. Are patients being put at risk? Jan 3, 2020. Retrieved Jan 4, 2020 from https://www.latimes.com/business/story/2020-01-03/artificial-intelligence-healthcare
[42] 屈原.《楚辞⋅卜居》:詹尹乃释策而谢曰:“夫尺有所短,寸有所长;物有所不足,智有所不明;数有所不逮,神有所不通。用君之心,行君之意。龟策诚不能知此事。”