A decade ago, the idea of digitally replicating a human being sounded more fiction than fact. Fast forward several years and the concept of creating a digital replica of a living human being that changes as conditions change, has generated both heightened interest and ongoing support. These virtual copies of living organisms reveal insights into the important physiologic processes necessary for everyday living, and makes better sense of the data from biological systems than ever before. Etiometry proudly pioneered this concept for health care in 2010, being the first to create a medical device software that can interpret patient data based on in-silico replicas of human physiology.
When we started nearly 10 years ago there wasn’t much support nor enthusiasm for the idea of combining differential equations to model autonomic regulation, hemodynamics, and metabolic interactions, all of which could help monitor the fundamental risks in critical care, such as the risk for inadequate oxygen delivery. The original detailed mathematical physiologic models developed by Guyton in the 1970s were confined only in academia and basic research. The concept of applying these models to real world monitoring may have been ahead of its time. The small field of real-time data analytics was focused on simple feature extraction such as heart rate variability, or applying well-studied techniques such as logistic regression to predict critical outcomes.
Our extensive experience in aerospace engineering contributed to making model-based digital human twins a reality. An airplane is not flown regulated by a black box model, but instead the specific laws of physics that enable an airplane to fly are embedded in computers that effectively create a digital twin of the airplane. The computer continually receives and interprets hundreds of data streams from various sensors and compares it to digital copies of the airplane in different states to infer the plane’s current status and trajectory. (As an example, the two recent Boeing 737 MAX crashes were attributed to the fact that the airplanes’ digital twins critically diverged from the current state of the actual planes.) We knew from the very beginning that although the modeling approach would be much more complex, it would lead to more meaningful and scalable interpretations of patient data. When a black box approach malfunctions, it is hard to diagnose or fix. When a model approach malfunctions this provides another datapoint and opportunity that will help improve the models. This additional data provides the digital copy with more detail and deeper context, information which can be used to advance our understanding of Physiology.
When “Novel Risk-Based Monitoring Solution to the Data Overload in Intensive Care Medicine” was published in 2013, and subsequently “Next Generation Patient Monitor Powered by In-Silico Physiology” published in 2015, we identified this as the appropriate solution to a fundamental problem with ineffective patient monitoring in the ICU, when clinicians are presented with an overwhelming amount of sensor data from connected patient monitoring devices. From that time, our algorithms have earned two FDA clearances, and we continue to update our models in an effort to ensure our digital twins closely resemble their real-world siblings.