Back in 2021 the PHG Foundation team looked into an overview of an enthralling new approach in the relatively obscure world of modelling. Not the front cover of Vogue type modelling, but 'the scaled down representation of physical structures used to test hypotheses in research' type modelling.
In engineering, a generation of digital twins has been busy saving the lives of their physical siblings. But, in relation to human health, preliminary PHG Foundation research has found the concept has yet to take off. A recent article, How my digital twin helped me survive cancer, has suggested otherwise - have digital twins gone from zero to health hero in less than a year? Could I, perhaps, get myself an avatar to provide highly personalised healthcare advice that could stave off another knee injury? Or should that troublesome joint require surgery, could AI identify how likely I (specifically) would be to suffer deep-vein thrombosis as a result?
Well no, not quite. The story of Anna's digital twin was a vision piece from Siemens Healthineers - a look into the future of what might be possible one day. So where on that journey are we today?
Just another 3D model?
Even in engineering, not everyone agrees on what, exactly, a digital twin is. In the type of controversy beloved of experts in any field, researchers dispute: is it a technology, a methodology, or a solution? For most of us the definition from the Frederick National Laboratory for Cancer Research: ‘A digital twin is a virtual model used to make predictions and run simulations without disrupting or harming the real-world object’ is a starting point.
A digital twin is a computer simulation (or virtual replica) of a physical entity such as an engine, or a bridge; or of a process such as a manufacturing pipeline. In other words - a model. Modelling is a common technique in clinical and public health decision-making that uses data captured from traditional records such as family history and medical records.
Of course, this provides only a snapshot of a person’s health at the time it was recorded. But by updating the records with new information from the patient, and perhaps assessing it against that of similar patients, the progression of the disorder can be monitored and predictions can be made about the likely impact of different interventions.
Just a handful of years ago, 3D computer modelling was taking the medical world by storm – fuelled by the arrival of 3D printing. Patient-specific 3D computer and printed models are most commonly generated from imaging data such as a patient’s CT or MRI scans.
As with 3D modelling, the digital twin can be used for analysing, monitoring, or making predictions about the physical entity and the impact of interventions. So where's the added value?
"Medicine lacks a key component of scientific practice because you can't perform controlled experiments at the level of individuals. Digital twins solve this problem by allowing virtual experiments and controls in which we evaluate what will happen to a specific individual given a specific treatment." James Glazier
'There is a two-way interaction between the physical entity and the digital twin, which isn’t necessarily the case with a static computerised model.' says Dr Sobia Raza, author of the PHG Foundation explainer on Digital Twins. This ability to interact with evolving datasets allows the digital twin to mirror changes taking place in its physical twin, potentially in real time. This could be wear and tear on an organ or joint in the human body, or the health of the individual in their entirety, shaped by time, genetics, the environment or nutrition to name just a few factors. And there are many factors.
As Prof James Glazier, who is exploring how digital twins might be used to personalise responses to pandemics notes "Medicine lacks a key component of scientific practice because you can't perform controlled experiments at the level of individuals. Digital twins solve this problem by allowing virtual experiments and controls in which we evaluate what will happen to a specific individual given a specific treatment."
In other words, humans comprise and are subject to a multitude of variables that keep, well, varying. And while each of us shares a basic common blueprint, we each too have our own biological and physical quirks, the results of our genomic and multi'omic profile as well as constantly changing external forces. Humans simply cannot control all the other variables to prevent them from changing too, and therefore influencing the study.
The four core elements of the digital twin - the physical being, virtual model, the flow of data between them, and crucially, the AI that analyses that data - promise not only a biologically identical version of the patient but one that responds to constantly changing internal and external influences, just as that patient would.
They can be used to pinpoint what is happening within a patient now - changes that would not otherwise be apparent until symptoms have developed and the patient becomes ill (prevention); or to test what is likely to happen to a specific patient given a specific course of treatment (prediction). At its most ambitious a digital twin could be used understand, predict and diagnose every miniscule change within an individual, using a continuous stream of real time data.
Using digital twins in healthcare
Smart environments: From building safer bridges to creating smart hospitals is more a leap of scale than a whole new direction. One initiative is the GE Healthcare Command Centers where the digital twin of a hospital can be used to test changes in operational strategy, capacity, staffing, and care delivery as part of the decision-making process.
Planning medical interventions: Digital twins could be used for optimising drug dosage or therapeutic choice by simulating the impact of treatments to evaluate them before use in the human twin. This offers the prospect of significant improvements in patient safety when testing the efficacy of a therapy. Studies using digital twins to plan and optimise heart surgeries are showing promising results.
The patient’s multi-omic, image, and pathological data collected can be used to build a digital twin to simulate various clinical scenarios, including drug combinations, doses, side effects, and durations for treatment plans and preferences.
Drug discovery: Drugs can be twinned too. In a global health emergency, such as the COVID-19 pandemic, the imperative is to get novel therapies into the hands of healthcare providers as soon as possible. This can, as we have seen, lead to public anxiety about the long term safety of, for example, vaccinations. An AI-driven digital vaccine twin could self-analyse its responses to changing data, to evaluate longer term safety and efficacy. The data could be generated from cohorts of virtual patients, created to match actual populations rather than only those that take part in research studies - offering opportunities to reduce racial and socio-economic disparities in data that compound existing health inequalities.
Nevertheless, digital twins don’t have to necessarily be linked and mirror the physical device exactly. They can also be predictive wherein a range of potential future states can be created and tested independently on the digital twin.
The Nth degree of personalisation
An interesting controversy is around the degree of twin-ness a model needs to be considered a twin. A generic artificial pancreas that uses real-time data from a specific patient to adjust drug dosage is one end of the spectrum. At the other end, and some might consider these the only true twin, is the 'biological, physiological, anatomical, demographic, environmental match'. Something like the CloudDTH model propsed by a team of researchers from Beijing and Sweden.
Cloud digital twin healthcare (CloudDTH) aims to address challenges around care for vulnerable people in the community. These include limited direct contact, poor compliance with treatment plans and responding to emergencies.
Just as the engineers in Norway attached sensors to bridges, wearables and sensors located in and around the home could monitor changes in the environment (e.g. wind speed and temperature) and the body (e.g. blood pressure, blood oxygen, body temperature) in a continuous cycle of real-time monitoring, digital consultation and assessment. This fully connected, AI-driven system goes beyond modelling to provide early warning alerts and medication reminders. It could build a picture of the staff and resources available to help optimise treatment for both patient and healthcare services.
Based on the digital twin model, medical experts or doctors do not need to see patients in person. They can determine the symptoms and prescribe treatment plans based on various real-time dynamic data from DT models, and optimize treatment options through model iteration on a virtual platform.
For those already concerned about alienation under an automated health system, this may be sounding early warning bells of its own. But there are those who anticipate improved interactions between patient and clinician as a result of better data.
The outcome will then be presented in a human-understandable format to the patient and clinician. CPDTs (Cancer Patient Digital Twins) will continuously account for the evolving cancer state, thereby improving outcomes and patient-clinician interaction.
Time for a policy review?
In the UK, at least, pressures on health and care services - and of course the COVID-19 pandemic – have boosted interest in public health initiatives and their emphasis on prevention in preference to treatment. However, while there are initiatives examining digital twins for the built environment, such as work by The Open Data Institute and The Alan Turing Institute, policy is quiet on the subject of the digital twin in biomedicine and healthcare. Perhaps this reflects the immaturity of the concept of creating digital twins for patients or other elements of healthcare.
This relative silence could be about to change. Features and journal articles on the topic have exploded onto the internet in the last 12 months. Frontiersin.org has recently issued a call for papers on the subject. Forbes has declared digital twins one of the five major healthcare tech trends for 2022 - just ahead of personalised medicine and genomics. And, as noted, Siemens Healthineers, one of the tech companies working on digital twin technology, have been ambitious enough to project a vision for a fully personalised digital healthcare assistant.
However, even a generous definition of digital twin identifies only a handful of applications in, or close to, real world implementation. These include the ‘artificial pancreas’, for type-1 diabetes patients, a device that monitors blood glucose and automatically adjusts the amount of insulin administered through a pump. And despite the promising research results, the graduation of digital twins into routine cardiology could still be a decade away.
Debate as to what constitutes a digital twin might seem like semantics, but is actually vital for effective regulation. We are not merely another structure: Norway's Stavå Bridge is not going to object to a potential invasion of privacy or question how the decision to replace its central girders was arrived at. As so many people have concerns as to how personal health data could and should be used for research, how much more intrusive could having an exact digital replica of our whole biological identity be?
There are many questions which are yet to be answered for even the more routine digitisation already underway in healthcare - where and how data is captured and from whom; who is allowed to do what with it; interoperability of all the disparate technologies; is it affordable - for everybody?
Credits:
Created with images by mandritoiu - "Bayonne Bridge, New Jersey at dusk" • Rawpixel.com - "Daughter holding her mothers hand in the hospital" • Hunor Kristo - "elderly woman"