What they are, How they’re used, and Challenges to be addressed
Key Takeaways
Digital twins provide more dynamic models than conventional simulations and will provide valuable tools for precision medicine. Conventional models and simulations provide a snapshot of a system rather than a real-time, dynamic model powered by sensor data and AI. Digital twins can couple statistical and mechanistic modeling fueled by real-time data flows from sensors that can respond to changes in behavior, therapies and overall patient status.
Digital twins have similar bias and ethical issues as AI in general. The same issues that have emerged with AI models in clinical decision support or bias from sensors that are less accurate on darker skins tones can impact digital twins and will need to be addressed with adequate safeguards and patient safety checks.
The FDA is actively engaged in utilizing digital twins in the regulatory science for medical devices as well as developing the standards that will guide regulation. Dassault Systems has a collaboration utilizing their Living Heart Project to model medical device applications. A number of collaborations with industry groups are under way that will inform the underlying regulatory science and standards for risk informed credibility standards for computational models.
Introduction
We are hearing more talk of the use of “digital twins” across biopharma, epidemiology, hospitals and even in pandemic planning for cities or smart, healthy cities. In this post I’ll be unpacking what digital twins are and how they differ from conventional modeling exercises and simulations. The broad digital twin market across industries is currently estimated to be over $3B and expected to grow to $48.2B by 2026.
Digital twins provide a more dynamic model or simulation that responds to shifts in data in real-time that can accurately reflect responses to treatment compared to [conventional simulations].
Digital twins are a virtual representation of a system, an organ or body, or a city, for example that utilizes AI, IoT/wearables, the cloud and data from multiple sources (sensors) in real-time to enable analyses for better decisions. Engineers have been utilizing digital twins for design, optimization, process control, virtual testing, and predictive maintenance.
The bleeding edge of digital twins research explores new Use and Maintenance-as-a-Service business models for some products (eg. GE’s maintenance for turbine engines), augmented and virtual reality applications, and novel ways to use AI for identification of new innovations. A digital twin of a Tesla, for example, provides real-time monitoring of a vehicle and can troubleshoot for problems with the real automobile and push a software upgrade to a vehicle to address a problem.
From our review of the literature on digital twins we found a great deal of blurring of boundaries at times between conventional simulations or modeling vs digital twins. Both use virtual environments but digital twins can use real-time continuous data flows where a conventional simulation model represents a snapshot based on a given dataset. Digital twins, therefore, provide a more dynamic model or simulation that responds to shifts in data in real-time that can accurately reflect responses to treatment compared to a static model.
Use Cases in Health for Digital Twins
The impact of the pandemic on urban life has prompted exploration of how digital twins of cities can play a role in preparedness efforts and forecasting different epidemiological models…The models can integrate environmental data, epidemiological data and economic data to help city planners develop preparedness efforts for the next pandemic.
One of the earliest digital twin use cases in healthcare involved Dassault’s Living Heart Project (developed in 2014) which was the first virtual model/digital twin of a living organ. This was created to model individualized therapies, surgeries and treatments via computational modeling and simulation that could also be applied to medical devices and trials for treatments involving the heart. The FDA has now adopted the Living Heart Project for simulating clinical trials, or in silico trials, with less reliance on animal models and for improving the speed and safety of trials.
| Company | Use Case | Notes |
|---|---|---|
| Philips | Heart Model, Dynamic Heart Model | Echocardiography based twin for medical device modeling |
| Hewlett Packard Enterprise | Blue Brain Project | Mouse brain connectome |
| OnScale | Project BreatheEasy | Lung Model for Covid-19 |
| Health Catalyst | Population Health, Public Health | Digital twin approach modeled on Formula One race cars |
Since the Dassault Systems application we have seen a number of other in silico organ system models such as Philips’ HeartModel and DynamicHeartModel that work with echocardiography to model heart function. Philips’ clients include a number of medical device makers including Medtronic and Boston Scientific. Hewlett-Packard Enterprise and the Ecole Polytechnique Federale de Lausanne have developed a similar model of the brain, Blue Brain Project, that is a model of a mouse brain that has reconstructed the connectome of the mouse brain as well as algorithms for modeling the circuitry. OnScale developed “Project BreatheEasy” or digital twins of lungs with the goal of optimizing outcomes of COVID-19 patients and utilization of ventilator resources. Researchers in the area of digital twins predict that someday they will be able to integrate genomic (and other -omic) data to reproduce the entire body. Health Catalyst has been borrowing from the experience of Formula One race cars to apply digital twins to population health management and public health challenges such as the COVID-19 pandemic.
Applications in drug therapies and vaccines
A recent paper in Science made the call for digital twin-based models for integrating immunology and physiology data that could provide clinical decision support tools for clinicians treating COVID-19 cases. Clinicians would be able to integrate EHR data and model immune and viral responses to various therapeutic approaches. In January 2021, DIGIPREDICT (Edge AI-deployed DIGItal Twins for PREDICTing disease progression) was launched, a new 48-month project focused on viral infections that can help explore “the interplay between viral infection, host response, development of (hyper)inflammation and cardiovascular injury in COVID-19.” This is based on individual patient specific pathophysiology and has the ability to predict the progression of infectious viral diseases.
In clinical trial design, Unlearn.ai, has developed a digital twin application that creates synthetic control arms of clinical trial participants. The goal is to accelerate enrollment times, reduce the risk of failure in clinical trials by creating a computationally generated longitudinal clinical record that captures what would have happened if the patient had received a placebo. Novadiscovery is a French company offering a similar digital twin application for clinical trials by coupling mechanistic models with AI to help de-risk clinical research programs.
Epidemiological Modeling and Smart Cities
The impact of the pandemic on urban life has prompted exploration of how digital twins of cities can play a role in preparedness efforts and forecasting different epidemiological models. CityZenith is a digital twin company focused on addressing pollution in cities as well as various other future health crises cities may face through their SmartWorldPro platform. The models developed can integrate environmental data, epidemiological data and economic data to help city planners develop preparedness efforts for the next pandemic. The platform provides 3D/4D modeling and APIs that can pull in data from common data sources across the domains they address.
Hospital Bed Supply and Supply Chains
In the UK the digital twin company, Iotics, describes itself as “software for data interactions” that creates ecosystems of digital twins to optimize the value of data assets that are often under-utilized by enterprises. The NHS utilized their platform over the past year for locating available hospital beds across the UK health system. The Critical Care Resource provides data on the number of beds available and has the ability to integrate data on test results, ventilators and patient administration
Digital twins technology has roots heavily in cross-purpose these technologies in healthcare asset management and supply chains as well. Some vendors of large imaging systems are already offering such digital twin asset management services to their hospital clients. Another area is vaccine distribution. Every shipment can have a digital twin created from sensor and geo-location data to monitor the integrity of the supply chain coupled with data on demand at various vaccination sites, for example.
Issues and Challenges
Social systems where bodies and cities are located are often difficult to model within strict engineering paradigms and designers of these systems can over-promise on this front.
One technical issue that digital twins applications must confront is knowing how much data is sufficient to accurately model a system for trusted results. Overly complex models do not necessarily perform as well as simpler models with just enough of the right data inputs. Not every clinical decision in cardiology requires data from every cell of the heart and more population focused digital twins will need to choose the most important inputs and this is where some of the greatest risks rest in relying too heavily on any single model or type of data.[1]
The point above raises the same concerns we have been covering at Chilmark Research over the past year or so about bias and privacy in AI/ML. The same safeguards for privacy and vigilance to any potential bias in data and applications exist. While digital twins for clinical applications will pull data from an individual’s health record and wearables, for example, having human oversight and checks for patient safety and bias issues is extremely important.
Cybersecurity and storage of data is another critical aspect of digital twin technology and some of the vendors listed in this post, such as Iotic, have more advanced security and storage features. We would expect to see use of federated learning and differential privacy for digital twins to address privacy and security.
One of the final challenges to address is the evolution of digital twins from an engineering paradigm to systems as complex as healthcare. While the aspirations of digital twins technologies to forecasting future states of systems is real, this is also where these types of technologies have fallen short in the past. Complex social systems where bodies and cities are located are often difficult to model within strict engineering paradigms and designers of these systems can over promise on this front. One final emerging ethical issue is the ownership of new digital products based on data from outside the firm. This will undoubtedly become part of the ongoing debate over control and ownership of health data.
The FDA’s Center for Devices and Radiological Health (CDRH) has been leading the way in the development of the underlying regulatory science (risk profiles, uncertainties, digital evidence, standards) for all forms of computational models used in healthcare including digital twins. One of the advances was the publishing of the V&V40 standard in 2018 that provides guidance on risk-informed credibility standards for computational models in specific contexts. There are a number of collaborations between the FDA and IEEE and others through the Medical Device Innovation Consortium that is working to inform the frameworks for the regulatory science and policy. The FDA clearly recognizes the role that digital twins can play in modeling risk of medical devices and therapies as well as ultimately lowering the cost of clinical trials. This is informing the engagement with companies to develop the appropriate regulatory standards.
[1] See Steve Levin from Dassault Systems here https://www.bio-itworld.com/news/2020/07/09/virtual-twins-their-roles-in-healthcare-dru
0 Comments