By Dattaraj Rao
It’s perhaps one of the greatest on-screen pairings of the last 40 years. Arnold Schwarzenegger in his 80s glory opposite all four feet and ten inches of Danny DeVito. The catch? They’re twins. I’m talking, of course, about the 1988 movie not so creatively named Twins that takes the joke of comparing a former Mr. Olympia to the short and squat DeVito — no, not that type of squat — and turns it into an entire movie.
Despite the ridiculous premise of the movie, it captures a poignant image for the future of digital healthcare and beyond. A hot topic in the industry today is creating a digital twin for healthcare patients. In other words, this is the idea of digitally replicating a dynamic copy of a patient complete with all their historical and future healthcare data. Schwarzenegger and DeVito represent two very different outcomes for digital twins, and they highlight a central challenge that must be overcome if they are to power healthcare decision making into the future.
What is a Digital Twin?
Before diving into this challenge, we need to better outline the specifics of a digital twin. Consider the “San Junipero” episode from the Netflix series, “Black Mirror.” It follows two elderly women who upload their consciousness to the cloud and fall in love with each other via their “digital twins.” This is a far cry from what is possible today, of course, but the idea of creating a dynamic replica of important healthcare data is very much a reality.
Today, we have more healthcare data than we know what to do with. To name a few, we have EMR notes, test results, medical procedures, hospital telemetry, social media activity, wearable devices, smartphone data on personal activity, and social and environmental exposures such as job stress or a big move.
A digital twin captures this data on an ongoing basis and helps clinicians really know a patient and predict when and where to intervene. The core of the digital twin is the algorithms that aggregate and compare millions of data points across the life of a patient. They give sense and purpose to the data. Think of it as a computing platform that curates data and automates actions for doctors and patients alike.
For instance, the conventional method of monitoring a diabetic patient is to regularly track blood sugar levels and medication consumption. Though advances in continuous blood sugar monitoring are making this process easier, digital twins take this data collection one step further by also analyzing a patient’s diet to see if they have high levels of refined sugars. It can then analyze this against genetic, lifestyle, and historical drug response data to map out a plan of action.
Furthermore, a huge advantage of a digital twin is in what-if scenarios. You could use the digital twin for virtual analysis to test how the body will respond to certain treatments and drugs. If you have a really effective digital twin, you could almost be able to do a virtual clinical trial without a human participant at all.
Not all Data Diets Are Alike
Yet, there is a central challenge to ensuring that a digital twin can do this type of heavy lifting. On one hand, the type of data that is being fed into the system is huge and unstructured, and making sense of it in real-time is difficult. On the other hand, the algorithms are data-hungry.
Just as Schwarzenegger needs a constant high-protein diet and a regular workout regimen, a quality digital twin needs a comprehensive diet of data and training that ensures that no biases or gaps in parameters lead to misevaluation. A faulty twin — the DeVito — is an imperfect copy of the patient that is starved of critical data that continues the healthcare status quo. This means that patients will be misdiagnosed or diagnosed too late, leading to treatable conditions snowballing into larger and more expensive issues.
In the case of the diabetic patient, if their digital twin is the DeVito rather than the Schwarzenegger, it means that something as crucial as the patient’s diet might not be under consideration. Blood sugar levels could still be monitored, but they won’t give a comprehensive view of the patient that could lead to holistic treatment and lifestyle change.
Going to the Weight Room
Healthcare is traditionally an area of siloes, but comprehensive treatment will demand data interoperability that allows all possible inputs to speak to each other. Creating a digital twin requires starting with addressing a specific clinical or economic need, but it must be able to expand to the full range of care delivery or misdiagnosis and waste will result. As we capture and utilize more and more unstructured data via wearable technology or smartphones, there really isn’t any excuse for not leveraging that data to save lives.
The digital twin is a model that puts computing and data at the center of care delivery. The future of digital twins will require creative solutions that organize, move, and store data while also focusing on the design, deployment, and management of algorithms for a variety of patient types. In the end, when looking at your digital twin, you’ll want to make sure that it captures an accurate picture of your life without missing a factor that could risk it. By training our models and feeding them a balanced diet of data, we can all end up with a Schwarzenegger rather than a DeVito.
Of course, Schwarzenegger and DeVito are merely colorful examples. Ultimately, we don’t want an overpowered digital twin any more than we want an underpowered one. They should be exact digital copies of patients, complete with all their ailments and imperfections. This true picture of a patient’s health is only one of strength in that it has been trained and fed the very best data available, and it will revolutionize our ability to test and predict healthcare outcomes.
Dattaraj Rao, Innovation and R&D Architect at Persistent Systems, is the author of the book “Keras to Kubernetes: The Journey of a Machine Learning Model to Production.” At Persistent Systems, Dattaraj leads the AI Research Lab that explores state-of-the-art algorithms in Computer Vision, Natural Language Understanding, Probabilistic programming, Reinforcement Learning, Explainable AI, etc. and demonstrates applicability in Healthcare, Banking and Industrial domains. Dattaraj has 11 patents in Machine Learning and Computer Vision.