Solving health problems may be an educated guess by digital twins
A new AI tool that predicts your health by creating your digital twin is the latest way technology is transforming health. See how it works.
An artificial intelligence tool that can predict a patient’s health has been created by Victorian researchers, introducing a novel approach to personalised medicine.
Called DT-GPT, it creates a “digital twin” to make educated guesses about how a patient may respond to a drug or therapy, even the progress of their disease.
The University of Melbourne researchers describe the deep learning tool as a game-changer for those with complex medical conditions including endometriosis, asthma, lung cancer and even Alzheimer’s disease.
The tool reads a patient’s medical history which includes lab results, diagnoses and treatments, and generates their digital twin or a virtual human.
This digital twin provides a detailed, personalised forecast that predicts how their key health indicators will change in days, weeks, or even months.
“We can give clinicians a tool in the hands, so that they can use it as a sparring partner to create a virtual copy of a patient and then do simulations with it,” lead researcher Michael Menden said. “They can ask ‘what if’ to get a meaningful prediction on an outcome.”
Associate Professor Menden said it may also help accelerate drug discovery and preclinical trials as the tool can predict potential side effects.
He said DT-GPT had been “trained’ using data from tens of thousands of electronic patient health records where consent had been given for access.
“Pandora’s box is open and there’s no going back,” A/Prof Menden said. “AI is here to stay (in precision medicine) and it will make our lives better, but we have to do it right.”
He said this meant making sure creating digital twins for patients was done ethically from the start.
“We can do amazing predictions with this technology, but it is important to have trust, and the community sees the value of it and that it’s not stigmatised so it becomes a useful tool,” Assoc. Prof Menden said.
In proof-of-concept research published in NPJ Digital Medicine, the team outlined how they had created virtual twins of 35,131 ICU patients. These accurately predicted what would happen to their magnesium levels, oxygen saturation and respiratory rate over the next 24 hours based on their laboratory results from the day before.
They also used the tool to predict cognitive decline in a group of Alzheimer’s patients over 24 months.
He said getting access to large quantities of data to train the model was key.
The researchers have recently partnered with the Royal Women’s Hospital to further develop DT-GPT AI software to build digital twins for endometriosis patients.
Assoc. Prof Menden said The Women’s had a unique database of endometriosis patients.
“We bring the technology to the table. But we are always looking for good opportunities to work with clinicians, and with exciting data sets and biomedical research questions,” he said.
The AI model also has the ability to interpret dense and “messy” data, where there are errors or inconsistent formatting, as well as a conversational ability like a Chabot so users can understand the reasoning behind its predictions.
Assoc. Prof Menden said it opens the door for a shift from reactive to predictive and personalised medicine.
“These deep learning models can look at hundreds of thousands of medical records, so they see more data than any clinician in their lifetime, and they can do educated guesses,” he said.
“Within five years there will be a lot of virtual humans and digital twins. This is going to change the biomedical landscape.
“I’m convinced it will change clinical practice, clinical trial planning and drug discovery and development.”
