What is an in silico trial? The use of individualised computer simulation in the development or regulatory evaluation of a medicinal product, medical device, or medical intervention. Why are they needed? To save resources – money, experimental animals, time – in pre-clinical and clinical studies.
Computer simulations are used in other fields such as aviation, motor racing, and meteorology. But can they predict human cardiovascular health? They already are. Biophysical models turn concepts from physics and physiology into mathematical equations that predict the outcome of therapies and describe patient phenotypes. Ideally, they would be fed diagnostic information from an individual patient in the clinic and then turned into a personalised heart model (digital twin), potentially revealing underlying disease substrates not obvious from individual measurements.
In silico trials exist in many domains relevant to the ESC community. For example, the Virtual-heart Arrhythmia Risk Predictor (VARP) study focussed on arrhythmia risk stratification of patients after myocardial infarction using personalised heart models. Furthermore, the comprehensive in vitro proarrhythmic assay (CiPA) initiative aims to engineer an assay for assessment of the proarrhythmic potential of newly developed pharmacological compounds through testing drug effects in a combination of in silico, in vitro, and human stem-cell derived myocyte models.
The need for interoperability of different models
- Interoperability is linked to the research question: international platforms exist for pharmacokinetic simulation; models of more specific and complex subdomains are less suitable for standardisation.
- The American Society of Mechanical Engineers (ASME) is working on standardisation protocols to compare models of the heart and vessels.
- Comparison is difficult because most models are designed for a specific purpose. And they vary in scale, from macroscopic (haemodynamics, mechanics) down to transcription of proteins, ion channels and cellular electrophysiology.
- A step towards a more comprehensible set of models is international funding: partners are teaming up using European Commission Horizon 2020 grants.
- What should be done if several models give different answers to the same question?
Should we aim for a virtual human?
- The concept of a virtual physiological human that aims to describe everything should not be a goal per se.
- The more complex the model, the more difficult it is to validate, due to growing uncertainty.
- Personalisation poses challenges because the more complex a model is, the more difficult it is to make it relevant to a specific patient population.
- Complex models can be very useful for specific research questions and specific contexts.
- Complexity depends on the question you want to address and the system you want to simulate. E.g. comparing the haemodynamic effects of different pacing protocols does not require detailed subcellular models.
- How can patient variability be factored into models – e.g. some never experience arrhythmia despite having the substrate. It is impossible to predict when and under what circumstances an arrhythmia will occur, but a model could identify patients that are more susceptible.
- Could models identify features that predict ventricular tachycardia and who should receive an implantable cardioverter defibrillator (ICD)? They may add value in high-risk groups.
- Models often describe a steady situation, while most diseases progress. More work is needed in this area, especially to predict long-term outcomes after drug or device therapy.
How should performance of models be assessed?
- Animal experiments are still needed to test reliability of models.
- Predictive ability is one way to assess quality. This should be standardised – for example testing models against a common real-world dataset to judge reliability.
- The ESC could set challenges whereby groups show the performance of their model against a specific question and dataset with a known result.
- It is rare to find comprehensive datasets that are able to completely validate a model. This needs to be addressed in a more systematic way.
- Quality of data to input into the model is a big issue. What might look like a good image dataset clinically is often insufficient for modelling.
- You learn most from wrong models: the biggest steps in understanding diseases are when the model is unable to reproduce the phenotypic features of patients. Collaboration with physiologists and clinicians leads to new hypotheses.
- Validation of single models is crucial, but we also need rules for in silico trials.
- The ESC could bring together stakeholders in Europe to produce a consensus document on how to assess models and conduct in silico trials in cardiology.
Can in silico trials replace clinical trials?
- There will never be a complete in silico trial; testing in humans will always be needed. They can replace some early research to reduce risk, speed up development, and reduce cost.
- In silico trials can test the validity of a hypothesis before testing in the clinic. If the clinical and in silico trial results differ, you can ask: 1) is your model incorrect 2) is the clinical trial result a chance finding.
- In silico clinical trials provide an early opportunity to get the human warranty on tools coming from artificial intelligence that will be used for clinical decision support systems.
Can in silico trials provide supplementary data for regulatory approval?
- The US Food and Drug Administration (FDA) is looking at alternative methods to evaluate devices including computer modelling and in silico trials. In Europe, groups could advocate for an expert laboratory for in silico studies
- Animal testing of devices developed for humans is not logical. Could modelling data be used to supplement regulatory submissions?
- The new EU medical device regulation, effective in May 2020, includes software as a medical device. It is unclear how this will affect in silico data.
Current applications of in silico data encompass virtual control arms and simulation of drug effects. Standards are needed on how to assess models and conduct in silico trials. More discussion is required on how in silico trials could provide supplementary information for regulatory approval.