Clinical trials are the means by which researchers test new treatments and diagnostic tools to make sure they are safe and effective. But clinical trials are expensive – and not without risk for those taking part.
Setting up a clinical trial also raises a whole raft of scientific and ethical questions for those designing it. For example, how many patients need to have tested a new drug or medical device to make sure it works and to ensure any important side effects are picked up? Do those involved in the trial represent those who will actually use the product later? And which patients should be excluded?
Now, advances in computer modelling promise to help improve the clinical trials process by giving researchers the chance to run their tests on virtual patients first. Whole cohorts of virtual patients – who might be models linked to real-world patients, synthetic (artificial) patients or a mixture of both – could be enlisted to take part in virtual or ‘in silico’ clinical trials. At The Alan Turing Institute, we’re involved in developing models that could be used in virtual trials, as well as guidelines to ensure their outputs can be trusted.
Working with virtual patients could give researchers the freedom to explore different trial designs in order to figure out how they could optimise trials in human patients to reduce costs as well as unnecessary risks. The results of virtual trials could also help researchers to make better predictions about the outcomes of their treatments in groups that are under-represented in real trials. Women, for example, are often less likely to participate in clinical trials of treatments for many different diseases.
In virtual clinical trials, even ‘impossible’ trial designs – such as simultaneous testing of both a drug and a placebo on the same group of patients – could become possible.
Researchers are already developing virtual testing approaches for medical devices including scanners for detecting breast cancer and stents for treating brain aneurysms. Whilst these approaches are not yet widely used to guide decision-making about future trials in humans, early results suggest virtual trials can make accurate predictions that compare to those of real trials.
Testing medical devices in a virtual setting might mean bringing together multiple models. As well as models of patients, virtual trials could incorporate models of the devices they’re designed to test, models that convert predictions made about patients into clinical outcomes relevant to the trial, and even models that simulate the behaviour of clinicians.
At the Turing, researchers within my team at the Turing Research and Innovation Cluster in Digital Twins (TRIC-DT) are building ‘patient-specific’ models (also known as ‘digital twins’) of human hearts. Our models account for differences in the shape, size and other aspects of patients’ hearts, which can affect how they respond to treatments for certain heart conditions. Similar patient models – representing a variety of different hearts – could be used in virtual trials, in combination with models of, say, pacemakers or implantable defibrillators, to predict how individual patients might respond to treatment with these devices. Such predictions could support the design of real trials by, for example, informing the settings that should be tested for pacemakers or highlighting the need to include different groups of people based on heart size (particularly as women’s hearts are, on average, smaller).
However, before the makers of drugs and medical devices start carrying out virtual clinical trials, they need to know that they can rely on the results. So how can they be sure that real patients will respond like virtual ones?
To try to answer this question, the Turing has been collaborating with the Food and Drug Administration (FDA), which approves the use of new medical devices in the US, and medical device developers at Medtronic. In our new paper, we outline processes for assessing the credibility of virtual clinical trials designed to test medical devices and show, in an example workflow, how these assurance processes could be organised.
Assurance processes for computational modelling involve a combination of verification and validation procedures. Verification refers to checking for errors in computer code and inaccuracies in calculations, whilst validation is concerned with comparing the results from models to those obtained in the real world – in this case, the results of real-world trials. Researchers also need to be able to quantify the level of uncertainty associated with any predictions made by their models.
Our suggested workflow organises these key components into a system where researchers must assess not just how their device models, patient models and other models perform in isolation, but also in combination, in the context of a virtual trial setting.
This work can help regulatory bodies responsible for reviewing applications from the developers of medical devices, which may increasingly be tested using virtual means. In the US, the FDA recently published guidance for medical device submissions that involve modelling and simulation. Our new work supports this guidance by laying the foundations for consistent credibility assessment processes.
In the future, researchers could be running virtual clinical trials ahead of, or alongside, most real-world trials in humans – and animals too. Robust credibility assessment processes for the models underlying these virtual trials will be needed to ensure that we can trust their results, and harness their benefits to test medical treatments in safer, fairer and more affordable ways.
Read the paper:
Credibility assessment of in silico clinical trials for medical devices
Top image: momius
