1 of 14

Discussion of proposed manuscript, possibly from the VPWG:��A Practical Approach to Digital Twins�in Medicine

James Sluka

Intelligent Systems Engineering and Biocomplexity Institute

Indiana University

2 of 14

A Practical Approach to Digital Twins in Medicine: Outline

  • The goal of the paper is to define the domain of medical digital twins (DT) as well as the major benefits, opportunities, and challenges.
  • There are a range of understandings, definitions, and applications for DT. In our VP WG seminars, there is no consensus of what a "Digital Twin" is.
      • Is it a giant computing infrastructure that will cost billions to develop?
      • Or can usable examples be developed on a much smaller financial scale?
      • Is it AI based, or mechanism based?
      • How does it relate to "Personalized Medicine"?
      • What data is readily available and what data will require technological development to measure?
      • How is data shared and communicated?
      • Where is data stored?
      • How is the data converted into actionable insights in patient care?
    • We will attempt to enumerate the domains and range for Medical Digital Twins, outline the challenges, opportunities and benefits and provide a definition of the components.

​

3 of 14

A Practical Approach to Digital Twins in Medicine: Outline

  • The goal of the paper is to define the domain of medical digital twins (DT) as well as the major benefits, opportunities, and challenges.
  • There are a range of understandings, definitions, and applications for DT. In our VP WG seminars, there is no consensus of what a "Digital Twin" is.
      • Is it a giant computing infrastructure that will cost billions to develop?
      • Or can usable examples be developed on a much smaller financial scale?
      • Is it AI based, or mechanism based?
      • How does it relate to "Personalized Medicine"?
      • What data is readily available and what data will require technological development to measure?
      • How is data shared and communicated?
      • Where is data stored?
      • How is the data converted into actionable insights in patient care?
    • We will attempt to enumerate the domains and range for Medical Digital Twins, outline the challenges, opportunities and benefits and provide a definition of the components.

​

All of these questions can have multiple answers!

4 of 14

A Practical Approach to Digital Twins in Medicine: Target Journal?

  • A “Current Opinion” in medicine or human health journal?
  • Regardless of final submission journal, will release via bioRxiv

5 of 14

Relationship to Science opinion piece

  • This paper:
    • No discussion of any particular disease or biological process
    • This paper focused on
      • defining terms (“personal medicine” vs. “digital twin”)
      • challenges
      • opportunities
    • This paper is the gestalt understanding of more than 50 presentations to the VP WG viewed from a DT point of view.

​

​

6 of 14

Major Components of this Paper

  • Define terms:
    • Medicine (medical practice outside the two domains listed below)
    • Precision Medicine (aka Personalized Medicine) 🡨 Includes Omics data
    • Digital Twin 🡨 Includes high frequency longitudinal data on the individual

The key differentia

All of these have the goal of making a prediction based on data. The prediction may be statistical, inferential or mechanism based. They all intend to help a patient by designing the optimal medical response. THOSE CHARACTERISTICS ARE NOT differentia!

7 of 14

Major Components of this Paper

  • Define terms:
    • Medicine
    • Precision Medicine (aka Personalized Medicine) 🡨 Includes Omics data
    • Digital Twin 🡨 Includes high frequency longitudinal data on the individual
  • Major challenges in Digital Twins
    • Data ownership, sharing, storage, aggregation and dissemination
    • Regulatory requirements

8 of 14

Major Components of this Paper

  • Define terms:
    • Medicine
    • Precision Medicine (aka Personalized Medicine) 🡨 Includes Omics data
    • Digital Twin 🡨 Includes high frequency longitudinal data on the individual
  • Major challenges in Digital Twins
    • Data ownership, sharing, storage, aggregation and dissemination
    • Regulatory requirements
  • Hardware levels
    • For high intensity environments such as in an intensive care unit.
      • Medical device regulatory level
    • Disease specific DTs, such as glucose monitors and implanted artificial pancreas.
      • Medical device regulatory level
    • Fitbit / Smart Watch / Smart Phone
      • Level of activity, sleep, heart rate, blood oxygen, EKG, location, contact tracing, …
      • Regulatory requirements?

9 of 14

Major Components of this Paper

  • Define terms:
    • Medicine
    • Precision Medicine (aka Personalized Medicine) 🡨 Includes Omics data
    • Digital Twin 🡨 Includes high frequency longitudinal data on the individual
  • Major challenges in Digital Twins
    • Data ownership, sharing, storage, aggregation and dissemination
    • Regulatory requirements
  • Hardware levels
    • For high intensity environments such as in an intensive care unit.
      • Medical device regulatory level
    • Disease specific DTs, such as glucose monitors and implanted artificial pancreas.
      • Medical device regulatory level
    • Fitbit / Smart Watch / Smart Phone
      • Level of activity, sleep, heart rate, blood oxygen, EKG, location, contact tracing, …
      • Regulatory requirements?

Who owns and acquires the device and software?

10 of 14

Major Components of this Paper

  • Define terms:
    • Medicine
    • Precision Medicine (aka Personalized Medicine) 🡨 Includes Omics data
    • Digital Twin 🡨 Includes high frequency longitudinal data on the individual
  • Major challenges in Digital Twins
    • Data ownership, sharing, storage, aggregation and dissemination
    • Regulatory requirements
  • Hardware levels
    • High intensity environments such as in an intensive care unit.
      • Medical device regulatory level
    • More complex, disease specific DTs such as glucose monitors and implanted artificial pancreas.
      • Medical device regulatory level
    • Fitbit / Smart Watch / Smart Phone
      • Level of activity, sleep, heart rate, blood oxygen, EKG, location, contact tracing, …
      • Regulatory requirements?
  • Development Communities (Large corporation for some devices, might be able to crowdsource at the Fitbit level?)
  • Potential Benefits

11 of 14

Possible low cost (well, relatively low) developmental pathway

  • Fitbit / Smart Watch / Smart Phone
    • Not necessarily registered as a medical device.
      • The basic Fitbit is not registered with the FDA and is unregulated
      • Device with more complex capabilities, such as an EKG, are licensed as class II devices that need to show they reproduce the behavior of existing devices.
  • Funding opportunities, for example, Bridge2AI
    • “10,000” steps, could make a game out of it
    • Funding from medical insurance companies (like car insurance companies reducing rates if they can monitor your driving habits)
    • Bridge2AI
      • Collection of Fitbit or Smart watch type data over a large cohort would be a suitable Bridge2AI project
      • Many parameters that a Fitbit type device can monitor have relatively little value as a single measurement, but high frequency longitudinal measurements of multiple parameters across a large population may greatly enhance the value of the individual measurements.

12 of 14

Interested in Contributing?

  • Currently have a 9-page draft/outline
  • Would like to submit by mid-September.
  • Could brand as coming from Viral Pandemics WG
    • Could acknowledge VP WG instead of branding it as VP WG work product.
  • J Sluka will be lead author;
    • Contact JSluka@iu.edu if you are interested
  • Co-authorship will require substantive contributions, minimally critical reading with suggestions for improvements, fleshing out citations, figure creation etc.

13 of 14

ZOOM chat

  • From Me to Everyone: 03:06 PM
  • All of our previous conferences are listed at https://www.imagwiki.nibib.nih.gov/content/msm-viral-pandemics-meetings
  • Rusty's talk is at https://youtu.be/j1D14B6C9qc and slides https://drive.google.com/file/d/1CwPKbcurgmnSAwREztCF5UBZY5P35N2q/view?usp=sharing
  • From John Rice to Me: (Direct Message) 03:29 PM
  • good ideas
  • From glazier@iu.edu to Everyone: 03:45 PM
  • I'd add discussion of 1) Equity, 2) Digital Twin real-time vs off-line. 3) Agency: who makes decisions based on the twin observations---notify, doctor, patient, machine makes decision.
  • How did "Precision medicine' get hijacked by the 'Omics people?
  • There are a lot of "Wellness" digital twins, which monitor activity or stress and suggest exercise, relaxation,.... Grrthm, and others are selling devices of this kind
  • What would we need to measure to deliver Gary An's digital twin? Would we need implanted sensors? Could it be done transdermally? There are attempts to do at least some bloood chemistry transdermally
  • What about monitoring urine, exhalates,..... Lower frequency but non-invasive and give chemistry
  • Usual hierarchy of application: diagnosis, prognosis, treatment optimization
  • From Gary An to Everyone: 03:46 PM
  • My perpetual interest is in getting better at designing better therapeutics... Prognosis and Diagnosis are not sufficient. Digital Twins are only relevant in terms of being able to parse responders from non responders to a particular therapy
  • From glazier@iu.edu to Everyone: 03:47 PM
  • Self improving models

14 of 14

ZOOM chat

  • From Chantal Darquenne to Everyone: 03:48 PM
  • Unless your digital twin can alert and suggest behavior changes BEFORE therapeutics are needed.
  • From Gary An to Everyone: 03:50 PM
  • Preventive medicine can only go so far. For the foreseeable future people are going to get sick
  • From glazier@iu.edu to Everyone: 03:52 PM
  • Mechanistic vs AI "models"
  • AI can't extrapolate very well, it is good at interpolating
  • From John Rice to Everyone: 03:54 PM
  • A distinction between making predictions about me based on some average for the population, vs making it based on a long history about ME. If my BP goes up every morning and has for the last 20 years, is is very different form son=me one whos bp has never done that and now started to.
  • From Gary An to Everyone: 03:57 PM
  • Have to go to the OR, will catch up with you all
  • From Me to Everyone: 03:57 PM
  • For the foreseeable future most medical issues are largely preventable.
  • From glazier@iu.edu to Everyone: 03:59 PM
  • We definitely are NOT saying a digital twin is AI
  • again, dynamics is critical
  • From Mitchel Colebank to Everyone: 04:02 PM
  • Machine learning would be forecasting outcomes using data, whereas digital twin-mechanistic modeling would be forecasting outcomes based on data-informed surrogates of a physical process? So the advantage over machine learning would be simulating the mechanisms at play
  • From Russ Taylor to Everyone: 04:02 PM
  • rht@jhu.edu
  • May be a good resource: http://www.nature.com/articles/s43588-021-00072-5
  • Need to no say that fitbit is the Dt.