Medical digital twins: Modeling healthcare from the patient to the ward

• A digital twin focuses on real-time data and continuous feedback, different from predictive analytics, which are used solely for forecasting
• Digital twins can be practically implemented on multiple levels within healthcare, from the individual organ of a patient to a hospital workflow
• Digital twins promise to improve operational efficiencies and provide more tailored patient care

Despite living longer, people are  experiencing higher rates of acute and chronic illnesses, increasing the burden on healthcare systems.¹ Shortages of hospital beds and staffing, along with limitations in manual patient monitoring, can lead to operational inefficiencies and increased costs.²

To address these challenges, healthcare organizations are using data to dramatically shift patient care. At the patient level, electronic health records (EHRs),  information from disease registries, “-omics” data (such as genomics, biomics, etc.), and demographic information can be combined to create statistical models of a patient’s real-time and future health.³ This is what is referred to as a medical digital twin.

But medical digital twins don’t have to be limited to human bodies, they can also work as virtual representations of systems or processes, such as radiology workflows or a patient ward. Digital twins allow health systems to use real-time data to simulate and predict how different factors will impact not only individual patient responses, but also overall healthcare operations and entire hospital care units.³

At Roche Experience Days (RED) 2024, Dr. Gurpreet Singh, Founder and CEO of Respiree, provided insights into how digital twins are impacting healthcare and allowing physicians to make better decisions for patients.

Though predictive analytics and digital twins are sometimes used interchangeably, there are several distinct differences between predictive analytics and digital twins, according to Dr. Singh. First, a digital twin places more emphasis on real-time data, which is critical to training  artificial intelligence-machine learning (AI-ML)  models. “The model needs to get feedback about what’s happening next,” said Dr. Singh. By contrast, predictive analytics can use asynchronous data, for example in forecasting. According to Dr. Singh, a digital twin AI-ML model is characterized by a need for more real-time, continuous feedback. This data loop then allows the digital twin to accurately predict how the patient will respond.

Another difference is that a digital twin relies on the interconnected communication between channels or systems, such as data from wearables, clinical monitoring systems, EHRs, and medical imaging.³ Effective integration and interaction between these systems is crucial for seamless real-time data exchange.

And lastly, a digital twin includes both predictive analytics and prescriptive analytics to simulate responses to stimuli, such as mechanical or physiological stimuli. “You allow not only forecasting information, but really providing a recommendation, and then monitoring that recommendation,” said Dr. Singh. For a patient, this enables the regulation of therapeutic interventions and determines how the patient is responding. At the systems level, a digital twin identifies inefficiencies, provides recommendations to optimize staffing or equipment use, and monitors the impact of changes.³

Dr. Singh shared three key factors about digital twins:

1. Bidirectional data flow: Input data, such as mechanistic, genomics, proteomics, and physio-molecular, is fed into the models with subsequent data output. “It needs to drive patient care. It needs to either improve through some form of intervention, which then again goes back into the model to let the model learn—so effectively a bidirectional data flow,” commented Dr. Singh.
2. Digital models: The models are developed with AI and ML, and may include large language models (LLM). These foundational models provide deeper, richer insight into patient outcomes.
3. Standardized and harmonized data inputs: Continuous variable data comes into the models at the resolution of milliseconds, seconds, and minutes, says Dr. Singh. The data needs to be standardized and harmonized into diverse datasets that provide relevant information to the healthcare system or payers.

As discussed previously, digital twins can be practically implemented on multiple levels within healthcare, according to Dr. Singh. A first level can be of an organ, such as a digital twin of a pancreas, which can simulate responses to the automated delivery of insulin. At another level, a digital twin can model the operational workflow of a radiology unit, for example, to help identify improvement potential, such as shorter wait times, faster patient turnaround, increased equipment utilization, and lower staffing costs.

At an even higher level there can be a digital twin of a patient care unit or hospital ward. At this level, digital twins can help optimize bed allocation or support early prediction of disease progression, without relying on manpower or manual resources. “Beyond just the model development, it is also important to close the workflow. Essentially it needs to be able to allow for proactive management,” remarked Dr. Singh. “How can you allow the nurses and the healthcare systems to identify patient deterioration way before it happens, but not when it happens, so that you allow them to optimize their workflow?” said Dr. Singh.

As costs and healthcare demands continue to rise, digital twins can provide significant support in improving operational efficiency and tailored care. “There is potentially a usefulness in terms of using digital twins and data to really augment resourcing, staffing, and to support personalized treatments,” commented Dr. Singh.

Digital twins will be especially important as new technologies hit consumers. Dr. Singh added, “A lot of consumer wearable devices are also coming to the market. There’s more vested interest in building connected IoT (Internet of Things) systems.”

Compared to classical predictive AI, digital twins offer a more connected and integrated system, concludes  Dr. Singh. While predictive AI focuses on individual predictions, digital twins use  real-time data processing, and incorporate continuous feedback.^3,4 This technology is changing  healthcare, creating a more dynamic system that not only enhances medical treatment, but also the experiences of patients and providers in the healthcare setting.

To watch the full video of Dr. Singh’s presentation at RED 2024 click here.