Researchers within the United Kingdom tested an AI tool called Foresight that creates digital twins of patients to predict future health and treatment outcomes.

The idea of ​​creating digital twins across different industries allows engineers to check systems in a simulation before deploying them within the physical world. AI tools like Foresight now make this possible for healthcare professionals.

Every time a patient sees a health care provider, information is added to their electronic health record (EHR). Some of this data is structured (age, gender, ethnicity), but most of it’s unstructured, comparable to test results or a health care provider’s notes.

Foresight uses a GPT-based model to convert this data right into a model or digital twin of the patient. Because Foresight is trained on large amounts of other patients’ EHR data, it’s then capable of predict health outcomes comparable to the forms of illnesses a patient is more likely to develop or their response to a specific form of treatment.

James Teo, professor at King’s College Hospital and co-author of the study, explained the importance of this fact. Teo said on

You could take a patient’s EHR after which simulate multiple versions of the patient to predict their health trajectory. Traditionally, a health care provider would should read a patient’s electronic medical record, settle on a treatment option, after which evaluate the outcomes after a while to watch the effectiveness of the treatment.

With Foresight, a health care provider could simulate multiple potential treatments, with the model predicting the short- and long-term outcomes of every treatment. This is a far cheaper approach and saves the patient the “let’s do this” approach that many physicians should resort to.

CogStack retrieves EHR data, MedCAT annotates it, Foresight Core is the deep learning model for modeling biomedical concepts, and the Foresight web application enables interaction with the trained model. Source: The Lancet Digital Health


The study, published in The Lancet Digital Healthexplained how the researchers trained three different foresight models using hospital datasets from two UK hospitals and a publicly available dataset within the US for a complete of 811,336 patients.

Foresight’s task was to pick out the disorder a patient was most certainly to develop from an inventory of 10 possible disorders. Using the 2 UK data sets, the following disruption could possibly be accurately predicted 68% and 76% of the time, respectively, and using US data, 88% of the time.

When tasked with predicting the following latest biomedical “concept,” which could possibly be a disorder, symptom, relapse, or drug, Foresight achieved 80%, 81% accuracy using the UK and US datasets or 91%.

The differences in performance show how dependent AI tools are on high-quality data.

As exciting as this application of AI is, the researchers discover several challenges that should be overcome. Finding ways to make the model properly weigh latest treatments and interventions, or properly assess the importance of probability versus urgency and impact are only two examples.

The researchers are working on developing Foresight 2, which they are saying will likely be a more accurate model.

With the invention of recent medicines and ideas comparable to patient modeling, simulation and prognosis, AI may have a big impact on the standard of our healthcare.

This article was originally published at