smart-phone-based DSSs have been evaluated in clinical studies , the outcomes have been mixed with some showing a benefit in terms of improving glycemic outcomes , while others do not . 7 , 8 , 9
When combined with CGM sensors that provide real-time glucose estimations , AI-based DSS algorithms can also provide personalized hypoglycemia prediction and prevention . Recent studies have demonstrated that AI-recommended insulin dosage adjustments agree with physician opinion with a level of accuracy approaching that of inter-physician agreement . 7 , 8 , 9
5 ) Translational research - Which requires analysis of abundant databases with numerous variables .
6 ) Pre-emptive Medicine - Pre-emptive medicine is a novel concept proposed in Japan , which aims at delaying the onset , or even preventing the occurrence of chronic diseases , such as diabetes , hypertension , cancer or dementia by using a combination of AI techniques , genomic analysis and environmental interaction data . 3
While physicians do not need to know technical details of AI / ML , they should understand what it can and cannot do . That leads to the question of how to critically appraise machine learning methodologies in clinical practice . The following analyses help the assessment : 10
1 ) Are the end results provided by ML reflecting the true disease state ? Is there a clinical gold standard that the ML technology meets ?
2 ) How is diagnostic accuracy identified ? In medical applications , diagnostic accuracy is usually reported as sensitivity , specificity and area under the curve . It would be contingent on ML technologies to ensure clarity in clinical terms .
3 ) Is the dataset used in model development reflective of the setting in which the model will be applied ? i . e ., can adult blood sugar / insulin dosing datasets be used to provide insulin dosing in children ? Can a model to identify diabetic retinopathy trained on images of severe diabetic retinopathy identify mild or moderate disease accurately ?
4 ) Is the output of an AI / ML technology able to provide differential diagnosis or estimates of confidence ?
5 ) Is the performance of the AI / ML model reproducible and generalizable - across age , gender , ethnicity , populations in different geographic settings ? - so that predictive accuracy is shown to be robust beyond the cohort they have been developed in , or are separate models needed for each ?
With ever increasing availability of wearable and implantable medical devices , the benefits come with tradeoffs . Some of these tradeoffs include legal uncertainty . Who is liable for damage arising from use of AI / ML technologies ? Who is liable when technology fails , leading to damage or when technology driven datasets are hacked ? Consider the insulin pump - an external device that regulates blood sugars almost autonomously but requires significant human input . These human-in-the-loop systems may not truly be
ARTIFICIAL INTELLIGENCE autonomous because of the large number of variables that cannot be anticipated . Pump dysfunction is not uncommon . When users of pumps ignore methods of monitoring blood sugars or pump function , blood sugars can get very high with clinical signs before they notice anything wrong . These individuals may land in the emergency rooms , and in extreme cases , may die . Can all users be expected to understand device functions to fully comprehend the scope of potential downstream risk ?
Cyber security and data privacy is a major problem . Personal medical information and disease data are being captured by both insulin pump manufacturers and hackers . Continuous glucose monitors ( CGM ) that connect to phones and computers automatically capture data and most patients are unaware of their implications . Thus , medical AI raises novel issues at the intersection of privacy law , cyber security obligations and consumer protection . 11
The laws governing these technologies have not been able to keep up the advances in medical technologies and users have a great deal of difficulty being informed of and consent to inherent risk . 11
In conclusion , AI / ML technologies are here to stay . They do make diagnosis , medical decision making , management of disease processes and improving quality of life in patients better . Both physicians and their patients have a need to keep up with them , especially the understanding of what these technologies can do and also recognizing what they cannot do !
References
1
Anyoha , R : The history of Artificial Intelligence : https :// sitn . hms . harvard . edu / flash / 2017 / history-artificial-intelligence /
2
Brown , S : Machine learning , explained : https :// mitsloan . mit . edu / ideas-madeto-matter / machine-learning-explained
3
Gubbi , S et al : Artificial Intelligence and Machine Learning in Endocrinology and Metabolism : The Dawn of a New Era : Front . Endocrinol ., 28 March 2019 | https :// doi . org / 10.3389 / fendo . 2019.00185
4
FDA approved A . I . -based algorithms : https :// medicalfuturist . com / fda-approved-ai-based-algorithms /
5
Hong , N et al : Machine Learning Applications in Endocrinology and Metabolism Research : An Overview : Endocrinol Metab ( Seoul ). 2020 Mar ; 35 ( 1 ): 71 – 84 . Published online 2020 Mar 19 . doi : 10.3803 / EnM . 2020.35.1.71
6
Perakakis , N et al . Non-invasive diagnosis of non-alcoholic steatohepatitis and fibrosis with the use of omics and supervised learning : a proof of concept study . Metabolism . 2019 ; 101:154005 .
7
Tyler , NS et al : An artificial intelligence decision support system for the management of type 1 diabetes Nat Metab . 2020 Jul ; 2 ( 7 ): 612-619 . doi : 10.1038 / s42255-020-0212-y . Epub 2020 Jun1
8Vettoretti
, M et al : Advanced Diabetes Management Using Artificial Intelligence and Continuous Glucose Monitoring Sensors : Sensors ( Basel ). 2020 Jul ; 20 ( 14 ): 3870 . Published online 2020 Jul 10 . doi : 10.3390 / s20143870
9
Rigla , M et al : Artificial Intelligence Methodologies and Their Application to Diabetes : J Diabetes Sci Technol . 2018 Mar ; 12 ( 2 ): 303-310 . doi : 10.1177 / 1932296817710475 . Epub 2017 May 25 .
10Faes
, L et al : A Clinician ’ s Guide to Artificial Intelligence : How to Critically Appraise Machine Learning Studies : Transl Vis Sci Technol . 2020 Feb ; 9 ( 2 ): 7 . Published online 2020 Feb 12 . doi : 10.1167 / tvst . 9.2.7
11
Reyes , C et al : Legal Issues Raised by Medical AI : An Introductory Exploration : https :// www . americanbar . org / groups / business _ law / publications / blt / 2019 / 12 / medical-ai /
Dr . Raghavan is a practicing pediatric endocrinologist in Louisville . She has no financial interest in any of the technologies or companies mentioned in the article .
APRIL 2022 13