A new artificial intelligence tool that can create virtual representations of patients and predict individual health trajectories has been hailed a potential gamechanger for the clinical trial sector.

Researchers from the University of Melbourne used three datasets containing thousands of electronic patient health records to train an existing large language model (LLM). The AI model called DT-GPT, analyzed medical data of patients with either Alzheimer’s disease or non-small cell lung cancer, as well as patients admitted to intensive care units.

The model created digital twins of these patients and forecasted how their health was likely to change over time under treatment, helping to predict the course of their disease. It was able to make accurate predictions by utilizing its pre-existing knowledge of medical literature and evaluating the patient’s medical histories including laboratory results, diagnoses, and treatments.

The model wasn’t given information on the health outcomes of the patients, allowing researchers to validate its predictions. The paper is published in the journal npj Digital Medicine.

Digital twins improve predictions

Lead researcher, Associate Professor Michael Menden, said, “For each patient, we created a virtual replica by initializing the model with their individual clinical profile. For example, we created virtual twins of 35,131 intensive care unit (ICU) patients and accurately predicted what would happen to their magnesium levels, oxygen saturation and their respiratory rate over a 24 hour period, based on their laboratory results from the previous day.”

Overall, the DT-GPT model outperformed 14 other state-of-the-art machine learning models in predictive accuracy.

Researchers say their model could be used to simulate clinical trial outcomes, potentially making drug development faster, cheaper, and more efficient. “This technology paves the way for a shift from reactive to predictive and personalized medicine,” Associate Professor Menden said. “It could enable doctors to anticipate if their patient’s health will deteriorate so they can intervene earlier.

“It could also be used to predict negative side effects of medications, allowing doctors to tailor treatment plans to suit each patient’s unique characteristics and medical history, ultimately increasing the chances of a positive health outcome.”

Potential for personalized medicine and drug trials

The model has the ability to quickly interpret dense and messy data and has a conversational interface where users can interact like a chatbot to understand the reasoning behind its predictions. As DT-GPT harnesses generative AI, it can also make “zero-shot predictions,” which are educated guesses about laboratory values the model hasn’t been trained on.

“To use an analogy, it’s like asking the model to predict how tall someone will grow without providing the person’s height records and only giving their previous weight and shoe sizes,” Associate Professor Menden said. “Our model accurately predicted how lactate dehydrogenase (LDH) levels changed in non-small cell lung cancer patients 13 weeks after they started therapy, despite not training the model for this purpose.

“We compared it to traditional machine learning models, which were specifically trained for 69 clinical variables, including LDH, which we in comparison only educated guessed. Very surprisingly, the DT-GPT’s zero-shot predictions, its untrained guesses, were more accurate in 18% of cases.”