![“We were using artificial intelligence
before [the 1956 Dartmouth Workshop],
we just called it Operations Research”
- Herb Simon, 1974 interview
2](https://image.slidesharecdn.com/manionpisaclinicalvalidationofaimay2023-230516141940-dc5a507d/85/Validation-of-Clinical-Artificial-Intelligence-Where-We-Are-and-Where-We-Are-Going-2-320.jpg)
![YEAH, BUT THAT EPIC/SEPSIS SITUATION IS JUST ONE CASE? RIGHT?
15 March 2021
Common pitfalls and recommendations for
using machine learning to detect and
prognosticate for COVID-19 using chest
radiographs and CT scans (Nature Machine
Intelligence)
“Our search identified 2,212 studies, of which 415
were included after initial screening and, after
quality screening, 62 studies were included in this
systematic review. Our review finds that none of
the models identified are of potential clinical
use due to methodological flaws and/or underlying
biases.”
Aug. 16, 2021
At a leading health tech conference, enthusiasm for
machine learning mixes with calls for greater
scrutiny (Stat)
“Throughout the four-day conference [HIMSS], the
largest annual event in health care technology,
industry leaders called for better ways to evaluate
the usefulness of machine learning algorithms,
audit them for bias, and put in place regulations
designed to ensure reliability, fairness, and
transparency.”
SEAN MANION, CHIEF SCIENTIFIC OFFICER, EQUIDEUM HEALTH - SEAN.MANION@EQUIDEUM.COM 6](https://image.slidesharecdn.com/manionpisaclinicalvalidationofaimay2023-230516141940-dc5a507d/85/Validation-of-Clinical-Artificial-Intelligence-Where-We-Are-and-Where-We-Are-Going-6-320.jpg)
Validation of Clinical Artificial Intelligence: Where We Are and Where We Are Going
- 1. Validation of Clinical Artificial Intelligence Where We Are and Where We Are Going Delivered by: Sean Manion, PhD Chief Scientific Officer @ Equideum Health 14 May 2023 Prepared for PISA 2023 1
- 2. “We were using artificial intelligence before [the 1956 Dartmouth Workshop], we just called it Operations Research” - Herb Simon, 1974 interview 2
- 3. TOP 10 IDEAS IN STATISTICS THAT HAVE POWERED THE AI REVOLUTION 1. Hirotugu Akaike (1973). Information Theory and an Extension of the Maximum Likelihood Principle. Proceedings of the Second International Symposium on Information Theory. 2. John Tukey (1977). Exploratory Data Analysis. 3. Grace Wahba (1978). Improper Priors, Spline Smoothing and the Problem of Guarding Against Model Errors in Regression. Journal of the Royal Statistical Society. 4. Bradley Efron (1979). Bootstrap Methods: Another Look at the Jackknife. Annals of Statistics. 5. Alan Gelfand and Adrian Smith (1990). Sampling-based Approaches to Calculating Marginal Densities. Journal of the American Statistical Association. 6. Guido Imbens and Joshua Angrist (1994). Identification and Estimation of Local Average Treatment Effects. Econometrica. 7. Robert Tibshirani (1996). Regression Shrinkage and Selection Via the Lasso. Journal of the Royal Statistical Society. 8. Leland Wilkinson (1999). The Grammar of Graphics. 9. Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio (2014). Generative Adversarial Networks. Proceedings of the International Conference on Neural Information Processing Systems. 10. Yoshua Bengio, Yann LeCun, and Geoffrey Hinton (2015). Deep Learning. Nature. KIM MARTINEAU, COLUMBIA NEWS, 06 JUL 2021; CURATED FROM ANDREW GELMAN & AKI VEHTARI (2021) WHAT ARE THE MOST IMPORTANT STATISTICAL IDEAS OF THE PAST 50 YEARS?, JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION, 116:536, 2087-2097, DOI: 10.1080/01621459.2021.1938081 3
- 4. 4
- 5. HOW GOOD IS HEALTH AI IN A CLINICAL SETTING? March 18, 2019 Reducing Sepsis Mortality by One-Fifth with Epic A predictive model helps clinicians provide faster sepsis treatment (Epic) “A predictive model in Epic helps clinicians at Lee Health identify patterns that indicate adult patients in the early stages of sepsis.” June 21, 2021 The Epic Sepsis Model Falls Short—The Importance of External Validation (JAMA) “Despite generating alerts on 18% of all patients, the ESM did not detect sepsis in 67% of patients with sepsis.” SEAN MANION, CHIEF SCIENTIFIC OFFICER, EQUIDEUM HEALTH - SEAN.MANION@EQUIDEUM.COM 5
- 6. YEAH, BUT THAT EPIC/SEPSIS SITUATION IS JUST ONE CASE? RIGHT? 15 March 2021 Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans (Nature Machine Intelligence) “Our search identified 2,212 studies, of which 415 were included after initial screening and, after quality screening, 62 studies were included in this systematic review. Our review finds that none of the models identified are of potential clinical use due to methodological flaws and/or underlying biases.” Aug. 16, 2021 At a leading health tech conference, enthusiasm for machine learning mixes with calls for greater scrutiny (Stat) “Throughout the four-day conference [HIMSS], the largest annual event in health care technology, industry leaders called for better ways to evaluate the usefulness of machine learning algorithms, audit them for bias, and put in place regulations designed to ensure reliability, fairness, and transparency.” SEAN MANION, CHIEF SCIENTIFIC OFFICER, EQUIDEUM HEALTH - SEAN.MANION@EQUIDEUM.COM 6
- 7. Generalizability types de Hond, A.A.H., Shah, V.B., Kant, I.M.J. et al. Perspectives on validation of clinical predictive algorithms. npj Digit. Med. 6, 86 (2023). https://doi.org/10.1038/s41746-023-00832-9
- 8. “Machine learning in intensive care medicine: ready for take-of?” Fleuren et al 2020, Intensive Care Med (2020) 46:1486–1488 https://doi.org/10.1007/s00134-020- 06045-y Clinical Validation of AI = 1%
- 9. 9 “If errors of inaccuracy and errors of hypersensitivity always seemed to be in opposite directions, on what basis could we make a compromise between those errors? The answer was that we could make such a compromise only on a statistical basis.” - Norbert Wiener (w/ Julian Bigelow), “Extrapolation, Interpolation and Smoothing of Stationary Time Series” (1942/1949) regarding their Anti-Aircraft Predictor
- 10. Illustration of varying degrees of relevance for protocol items across common study types Ciolino, J. D., Spino, C., Ambrosius, W. T., Khalatbari, S., Cayetano, S. M., Lapidus, J. A., Nietert, P. J., Oster, R. A., Perkins, S. M., Pollock, B. H., Pomann, G. M., Price, L. L., Rice, T. W., Tosteson, T. D., Lindsell, C. J., & Spratt, H. (2021). Guidance for biostatisticians on their essential contributions to clinical and translational research protocol review. Journal of clinical and translational science, 5(1), e161.
- 11. Is Machine Learning Generalizable Research? HHS has defined "research" as a systematic investigation, including research development, testing and evaluation, designed to develop or contribute to generalizable knowledge. SEAN MANION, CHIEF SCIENTIFIC OFFICER, EQUIDEUM HEALTH - SEAN.MANION@EQUIDEUM.COM 11
- 12. Belmont Report 12 SEAN MANION, CHIEF SCIENTIFIC OFFICER, EQUIDEUM HEALTH - SEAN.MANION@EQUIDEUM.COM
- 13. Institutional Review Board (IRB) • Review Scientific, Medical, and Ethical aspects of research • Made of experts along with at least one member whose interest is non- scientific, and another not affiliated with the university/trial site • Some research is exempted, but this generally requires an IRB rep to determine, not the investigator 13 SEAN MANION, CHIEF SCIENTIFIC OFFICER, EQUIDEUM HEALTH - SEAN.MANION@EQUIDEUM.COM
- 14. AI-Institutional Review Board (AIRB)? Should some medical AI be reviewed or exempted by an IRB? Should IRBs be utilized for expanded ethical review for AI? Should a parallel structure be developed? ◦ Future of Privacy Forum, “Designing an Artificial Intelligence Research Review Committee” (2019) ◦ AI Global, “Responsible AI Guidelines: Independent Review” (2020) ◦ Ethical Machines – Reid Blackman, Harvard Business Review Press (2022) 14 SEAN MANION, CHIEF SCIENTIFIC OFFICER, EQUIDEUM HEALTH - SEAN.MANION@EQUIDEUM.COM
- 15. For Health, Blockchain is necessary but not sufficient. EmergingTechnologyConvergence:the EpicenterofHealthIndustryValueCreation Trusted Privacy-in-Depth A bundle of emerging privacy tech deployable via blockchain networks: public, private, and hybrid. Data Federation and Decentralized AI • Federated analytics and learning, secure inference • Incentivized training data sourcing with verifiable provenance, data confidence fabrics • Digital twins with self-sovereign AIs (avatars) in the Metaverse, in silico research on digital twin simulations Blockchain Networks Public, Private, and Hybrid Distributed ledgers Web2 - Today’s Modern Distributed Web Smart Contracts Secure Automation Across Organizational Boundaries Optional Tokenization Secured and Transferred, Fungible and Non-Fungible Decentralized Apps (“dApps”) Web3 User Experience, Wallets, Identities, Verifiable Credentials Tightly-Coupled Off-Chain “Blockchain” or “web3” 15 New Software Modalities New Hardware Modalities
- 16. Source Data Federated Data Architectures Metadata Technical Descriptions of Source Data Enterprise Source Data has been exponentially expanding for the last ~25 years. We have already reached the cybersecurity, bioethical, legal, and logistical extremes of data centralization strategies, while “data hungry” AI’s continue to starve. The growth rate is increasing from the edge with 5G and IoT, along with our new understanding of social determinant data as health-relevant data. Continuous feeds of all this data will soon be under the custody of the consumer, enabling radical empowerment through self-sovereign AI. Derived Data Analytical Findings, Insights Value Blockchain is the core enabler of federated compute. 16 ©2022 EQUIDEUM HEALTH | ALL RIGHTS RESERVED. Blockchain makes it possible to bring the compute to the data, vs. the data to the compute… in a trusted, scalable, and secure manner. It is no longer necessary to centralize Source Data to create Derived Data. Enterprise and Patient-Sovereign
- 17. 17 ©2022 Equideum Health | All rights reserved. Additional Information: https://equideum.health/ sean.manion@equideum.com https://www.linkedin.com/in/sean-manionphd/