|In Silico Clinical Trials:
How Computer Simulation Will Transform The Biomedical Industry
An international research and development roadmap for an industry-driven initiative
Edited by the Avicenna Consortium:
Marco Viceconti – Insigneo Institute, Sheffield (UK)
Edwin Morley-Fletcher – Lynkeus, Rome (IT)
Adriano Henney – Obsidian Biomedical Consulting, Manchester (UK)
Martina Contin – VPH Institute, Leuven (BE)
Karen El-Arifi – Insigneo Institute, Sheffield (UK)
Callum McGregor – Lynkeus, Rome (IT)
Anders Karlström – Obsidian Biomedical Consulting, Manchester (UK)
Emma Wilkinson – Insigneo Institute, Sheffield (UK)
“Avicenna – A Strategy for in silico Clinical Trials” is a Coordination and Support Action funded by the European Commission as part of the Seventh Framework Program for Research and Technological Development (FP7), under the Information Communication Technologies Programme (Contract Number 611819).
This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).
Chapter I.In silico clinical trials: a layperson’s introduction 4
Chapter II.Avicenna roadmap: motivation and process 8
II.1.Engineering a new industry 8
II.2.The Avicenna consensus process 11
Chapter III.The industrial need for in silico clinical trials 23
III.1.Pharma and devices: development pipelines 23
III.2.Modelling and simulation in the current industrial practice 24
III.3.Identify the 'issues' 26
III.4.Drivers and barriers for ISCT 28
Chapter IV.The socioeconomic need for in silico clinical trials 29
IV.1.The cultural resistances 30
IV.2.Socio-economic issues 33
IV.3.Ethical issues 51
Chapter V.In silico clinical trials use cases for medical devices 52
V.1.Modernising the development of medical devices 52
V.2.In silico clinical trials: Current practice 58
V.3.In silico clinical trials: Best practice 58
V.4.Use of In silico clinical trials for medical devices 59
Chapter VI.In silico clinical trials use cases for pharmaceuticals 62
VI.1.Modernising the development of pharmaceuticals 62
VI.2.In silico clinical trials: Current practice 66
VI.3.In silico clinical trials: Best practice 67
Chapter VII.In silico clinical trials: horizontal challenges and emerging technologies 73
VII.1.Horizontal research challenges 73
VII.2.The bigger picture: horizontal challenges 75
VII.3.Annex VII-1: RTD challenges as defined during Avicenna event four 79
Chapter VIII.In silico clinical trials: research challenges related to medical devices and combined products 80
IX.1.Beyond validation: model credibility 80
IX.2.In silico design and pre-clinical assessment of wearable or implantable devices 81
IX.3.Automate ISCT for medical devices 83
IX.4.Visual analytics to explore high-throughput simulation results 83
IX.5.The physiological envelope, the deployment envelope 84
IX.6.Reducing, refining, and partially replacing clinical trials 84
IX.7.Annex VIII-1: Device RTD challenges defined during event four 85
Chapter X.In silico clinical trials: research challenges related to pharmaceuticals and biotech products 88
X.2.Annex IX-1: pharma RTD challenges defined during event four 96
Chapter XI.The Avicenna Alliance 98
XI.1.Establishing a pre-competitive alliance 98
Chapter XII.Conclusions 100
References – a roadmap bibliography 103
Annex 1: experts involved in the Avicenna consensus process 115
As it evolved, the Avicenna Research and Technological Development Roadmap became a very large document, which was intended to serve multiple purposes, and inform multiple categories of stakeholders. To facilitate the reading, it was decided to divide it in 12 independent chapters, each a stand-alone document, but at the same time part of multiple reading trajectories:
Chapter I. In silico clinical trials: a layperson’s introduction
Chapter II. Avicenna roadmap: motivation and process
Chapter III. The industrial need for in silico clinical trials
Chapter IV. The socioeconomic need for in silico clinical trials
Chapter V. In silico clinical trials: use cases for medical devices
Chapter VI. In silico clinical trials: use cases for pharmaceuticals
Chapter VII. In silico clinical trials: horizontal challenges and emerging technologies
Chapter VIII. In silico clinical trials: research challenges related to medical devices and combined products
Chapter IX. In silico clinical trials: research challenges related to pharmaceuticals and biotech products
Chapter X. The Avicenna Alliance
Chapter XI. Conclusions
Each reader is welcome to “compose” his/her roadmap at will; here are some recommended reading lists, for some families of stakeholders:
- EC reviewers, other organisations interested in similar roadmapping exercises: I-XI.
- Policy makers, research funding agencies, charities: I, II, VII-X
- Industry executives: Executive Summary, I, IV, X
- Pharma producers, research hospitals, CRO, consultants, regulators: I, VI, IX, X
- Device producers, research hospitals, CROs, consultants, regulators: I, V, VIII, X
- Patients’ organisations: I, II, IV, X
- Providers: I, V-X.
The term In Silico Clinical Trials refers to “The use of individualised computer simulation in the development or regulatory evaluation of a medicinal product, medical device or medical intervention”.
While computer simulation is widely used for the development and de-risking of a number of “mission-critical” products such as civil aircraft, nuclear power plants, etc., biomedical product development and assessment is still predominantly founded on experimental rather than computer simulated approaches. The need for long and complex experiments in vitro, on animals, and then on patients during clinical trials pushes development costs to unsustainable levels, stifling innovation, and driving the cost of healthcare provision to unprecedented levels.
The Avicenna action, funded by the European Commission, has engaged 525 experts from 35 countries, including 22 of the 28 members of the European Union, in an 18 month consensus process, which produced this research and technological development roadmap.
This document provides an overview of how biomedical products are developed today, where In Silico Clinical Trials technologies are already used, and where else they could be used. From the identification of the barriers that prevent wider adoption, we derived a detailed list of research and technological challenges that require pre-competitive funding to be overcome.
We recommend that the European Commission, and all other international and national research funding agencies, include these research targets among their priorities, allocating significant resources to support approaches that could have huge socioeconomic benefit.
We also recommend industrial and academic stakeholders explore the formation of a pre-competitive alliance to coordinate and implement public and private funded research on this topic.
Last, but not least, we recommend that the regulatory bodies across the world avoid becoming the bottleneck for innovation and, in collaboration with academic and industrial experts, develop the framework of standards, protocols and shared resources required to evaluate the safety and the efficacy of biomedical products using In Silico Clinical Trials technologies.
Chapter I.In silico clinical trials: a layperson’s introduction
Authors: Marco Viceconti, James Kennedy, Adriano Henney, Markus Reiterer, Sebastian Polak, Markus Reiterer, Dirk Colaert, Jean-Pierre Boissel, Martina Contin, Claudia Mazzà, Annamaria Carusi, Enrico Dall’Ara, Matthew, Iwona Zwierzak, Karen El-Arifi, Massimo Cella, Dirk Colaert, Boissel, Giuseppe Assogna, Robert Hester, Filipe Helder Mota.
Summary: chapter II provide an introductory description of the ISCT technologies, and of the problems that they are expected to solve.
Any biomedical product1 to be distributed commercially must undergo a development and assessment process before being placed on the market. The appropriate level of scrutiny and rigorous testing before commercialisation is of paramount importance, due to the risk of potential harm. In most cases the producing company must demonstrate the efficacy of the product in healing or alleviating the effects of a disease or disability, as well as an acceptable safety profile, before any widespread use.
The only conclusive way to ensure the safety and efficacy of a biomedical product is to test it on humans. This is done through clinical assessment, which is usually carried out in three phases prior to the product reaching the market as well as during post-marketing surveillance:
Phase I. The product is tested on a small group of patients or healthy volunteers under strictly controlled conditions, in order to ensure that it can be used safely without any unexpected side effects.
Phase II. The product is tested on a larger group of patients, in order to verify whether it is effective, and produces the expected effects (through direct indicators of efficacy, or simple proxy measures) in those patients.
Phase III. The product is distributed to a much larger group of patients, in multiple hospitals and possibly in multiple countries, to evaluate its efficacy on clinical outcomes in a much larger community, ideally reflecting the wider population, and to identify any less frequent, unexpected safety or efficacy problems.
Post-marketing studies. If efficacy and lack of frequent unexpected effects are supported by phase III trial findings, and, consequently, the product has been accepted for use, a number of issues remain that require further clinical studies. These include efficiency and effectiveness in real world and different populations from those involved in phase III trials (a transposability problem due to the limited representativeness of patients included in phase II/III trials) and pricing which often needs further data to be fixed, for example calculating the population benefit compared to competitors. In some countries, regulators and/or payers request periodical re-assessment of effectiveness and efficiency.
By the time a clinical trial for a new product starts, the company will have already completed extensive testing using a series of laboratory experiments in what is called the pre-clinical evaluation period. Depending on the type of product, these tests can be done on a laboratory bench or in a mechanical testing frame, in vitro (literally meaning inside the glass), which may include looking at how a small culture of cells responds to the product; ex vivo (meaning out of the living organism, and used to indicate studies done on tissues or organs extracted from a body), for example inserting a medical device into a cadaver to verify that it can be safely implanted; or in vivo (meaning in the living) using animal models designed to mimic the human condition that the product is intended to treat.
The preclinical testing process represents an essential step in the development of any potential biomedical product. It is the means by which the fundamental basis for why a product might work is evaluated, and, hopefully, confirmed. However, due to the hugely complex nature of human diseases, the significant differences between individuals, and the inevitable variability in how a treatment is administered, it is not unusual for a product to perform exceptionally well in tightly controlled laboratory tests, but show some serious problems during clinical trials. According to the Tufts Center for the Study of Drug Development2 the development of a new pharmaceutical product, and its introduction into the market, is estimated to exceed US$2.5 billion, nearly 75% of which is spent in the various phases of clinical development. Every time a product fails late in the process, for example at the end of phase II or even phase III, the company suffers a huge loss.
Whilst clinical trials may tell us that a product is unsafe or ineffective, they rarely tell us why, or suggest how to improve it. As such, a product that fails during clinical trials may simply be abandoned, even if a small modification would solve the problem. This results in an ‘all-or-nothing’ mind-set in the biomedical industry, where the scope of the R&D investment virtually requires that a biomedical company focuses on reducing the risk of a potential product. This paradigm stifles innovation, decreasing the number of truly original biomedical products presented to the market every year, and at the same time increases the cost of development (which, paradoxically, further increases the risk). As a result, it is also becoming increasingly difficult for companies to undertake projects on rare diseases, since the associated costs cannot be justified against the limited return on investment.
The biomedical industry is not the only technology sector that deals with highly complex and potentially critical systems. In other sectors, such as aerospace, computer/chip design and nuclear industries, computer modelling and simulation is used extensively during both product development and assessment to overcome similar problems with mission-critical products. Can the same approach be used for biomedical products? In addition to traditional in vitro and in vivo studies, might we adopt a third way for developing and testing biomedical products by making use of this ‘in silico’ technology? In silico is an allusion to the Latin phrases in vitro or in situ, and stands for computations carried out on a silicon computer chip.
Computer modelling and simulation is already being used in the development of biomedical products. Pharmaceutical companies use computer models to estimate the pharmacokinetics (the movement of a drug into, through, and out of the body) and the pharmacodynamics (the biochemical and physiological effects on the body) of a new compound. Medical device companies use computational fluid dynamics to predict how blood or other bodily fluids move inside and around the device being tested, or structural finite element analysis to make sure that the forces exchanged between the body and the device will not cause any harm.
While these technologies are of great value, current in silico technologies struggle to help address a number of very difficult questions, including: Why do some patients react adversely to a drug, while others are fine? Another such problem would be: Why is it that blood clots form around the device in a few patients, while in most they do not? In short, what is missing is the ability to assess how potential biomedical products affect individual patients, who may have multiple variable factors that lead to the questions posed above. Some examples of how computer modelling and simulation can attempt to address this individual variability include:
Using a computer model of the patient to take account of factors such as his/her particular physiology, the individual manifestation of the disease being treated, lifestyle, and the presence of other unrelated diseases.
Using a computer model of the treatment to account for the consequences of compliance, or lack thereof, on expected outcomes in taking the drug at the times and dose prescribed. Or, in the case of a surgically implanted device, to account for the variability in surgeons’ experience and technique, as well as the particular anatomy and activity level of the patient.
If we could develop reliable computer models of the treatment (effect of the drug or device on the organism) and its deployment (administration of the drug or surgical procedure), together with reliable computer models of the patient’s characteristics, we could perform exploratory trials within the computer: in silico clinical trials (ISCT). This would enable the simulation of a number of elements affected by the administration of the candidate biomedical product. In such a scenario, ‘virtual’ patients would be given a ‘virtual’ treatment, enabling us to observe through a computer simulation how the product performs and whether it produces the intended effect, without inducing adverse effects that might be potentially dangerous for the patient. We believe that such ISCT could help to reduce, refine, and partially replace real clinical trials by:
Reducing the size and the duration of clinical trials through better design, for example, by identifying characteristics to determine which patients might be at greater risk of complications or providing earlier confirmation that the product is working as expected. ISCTs might also be used to ‘leverage’ a smaller clinical trial population, by adding simulated patients that might fill gaps in the individual variability seen in ‘real’ patients.
Refining clinical trials through clearer, more detailed information on potential outcomes and greater explanatory power in interpreting any adverse effects that might emerge, as well as better understanding how the tested product interacts with the individual patient anatomy and physiology, and predicting long-term or rare effects that clinical trials are unlikely to reveal.
Partially replacing clinical trials in those situations where ISCT can generate scientifically robust evidence. We already have examples where the regulators have accepted the replacement of animal models with in silico models under appropriate conditions. While real clinical trials will remain essential in most cases, there are specific situations where a reliable predictive model could conceivably replace a routine clinical assessment.
Complementing clinical trials by offering the ability to test experimental scenarios, which would normally be less probable in real patient cohorts. For example: What if the patient has the disease under investigation, but also diabetes and a heart rhythm disorder?
ISCT will involve the generation of computer models that will be applied to each patient enrolled in a trial simulating his/her disease and the treatment being tested. These models will predict the outcome and will be used alongside, or as part of, an existing clinical trial. The predictive accuracy of the models can be tested against the observations produced by the parallel clinical trial. Once this process is repeated for a sufficiently large number of patients, this data can be used with other available information (for example, the distribution of genotypes that are known to be relevant to the course of the disease for product mode of action but which are not regularly recorded in clinical trials) to design ‘virtual populations’. Altogether, this will produce a virtual library of data that can be used to test other in silico treatments, either for a different product or a refinement of the existing one. These simulations can first be used to develop a new product, and then to complement and refine the real clinical trial.
On this basis, we have defined ISCT as:
The use of individualised computer simulation in the development or regulatory evaluation of a medicinal product, medical device or medical intervention. It is a subdomain of 'in silico medicine', the discipline that encompasses the use of individualised computer simulations in all aspects of the prevention, diagnosis, prognostic assessment and treatment of disease.
Ultimately, ISCT can be used to obtain a quick and informed answer to questions such as: What if the effect is 20% less than expected?; What if the body weight is twice the one observed in our population?; What if the patient has a 10% increase in creatinine clearance? This opens the door to a whole new concept of medicine, based on the ability to predict reliably. The rest of this report will investigate in detail the issues with the current methods, and the factors that still prevent a wider adoption of ISCT technologies. From these reflections we set out the roadmap for research and technological development in the area of ISCT.
Chapter II.Avicenna roadmap: motivation and process
Authors: Marco Viceconti, Anders Karlström, Martina Contin, Jean-Pierre Boissel.
Summary: chapter III provides a general motivation for the roadmap, and a description of the consensus process, including AO, events, collaborative editing, etc. It also includes an annex with the name of all those who advised the Avicenna consensus process.
II.1.Engineering a new industry
In 1955 Solomon and Gold published a three compartments model of potassium transport in human erythrocytes (Solomon and Gold, 1955). This appears to be the first paper indexed by Index Medicus (now PubMed) with the keywords ‘physiology’ and ‘computer’. From that first study until the late 1980s, most computer models aimed to capture the basic mechanisms underlying physiological or pathological processes in mathematical form, without intending to make quantitatively accurate predictions. In the 1990s, the development of stochastic modelling and increased computational powers enabled the development of population-specific models that aimed to predict the average value of specific quantities over a population ((Eberl et al., 1997; Chabaud et al., 2002; Duval et al., 2002; Clermont et al., 2004; Kansal and Trimmer, 2005; Bouxsein et al., 2006; Ribba et al., 2006; Vande Geest et al., 2006; Rostami-Hodjegan and Tucker, 2007). In the early 2000s, the computational ecology community started to debate the virtues of individual-based models for population ecology (Lomnicki, 2001). Soon after in silico medicine research also began to use the first patient-specific models (Chabanas et al., 2003; Viceconti et al., 2004; Fernandez and Hunter, 2005; Wolters et al., 2005; Li et al., 2008; O'Rourke and McCullough, 2008). Some analysts started to suggest that such approaches could be useful in the development of new medical products (PricewaterhouseCoopers, 2008).
In 2007, a group of experts published Seeding the EuroPhysiome: A Roadmap to the Virtual Physiological Human3. They presented a scenario where imaging and sensing technologies were used to generate quantitative information about the biology, physiology, and pathology of a patient at different scales of space and time. This information would then be used as the input for multiscale computer models encapsulating all the knowledge available for a given disease process, in order to produce patient-specific predictions for diagnosis, prognosis, and treatment planning.
Since then, dozens of single groups and consortia around the world have developed a whole set of new technologies and methods, initiated with a similar perspective to that original research roadmap. While the vision of the Virtual Physiological Human (VPH) is not yet entirely realised, VPH technologies are being assessed clinically in a number of practical applications, and preliminary results suggest important improvements over current standards of care.
In some of these projects it has been necessary to simulate the treatment in addition to the pathophysiology in order to predict how a patient would respond to a particular treatment option.
In the RT3S project4, the deployment and the fatigue cycling of peripheral vascular stenting was modelled. The VPHOP project5 included a model of the effect of bisphosphonates on the metabolism of bone tissue. Some other projects have gone even further, for example, the PreDICT study6 which used VPH models to assess the cardio-toxicity of new drugs. Another project used an in silico acute stroke model to explore why hundreds of compounds that have been shown efficacious in rodent models failed in phase II or III clinical trials. The ratio of astrocytes over neurons, which is quite different in human brains and in rodents, was suggested as the cause (Dronne et al., 2007). One of the essential traits of the VPH approach is the recognition that there is no preferential scale, and each problem should be tackled starting from the space-time scale where the process is observed (middle-out approach).
Of course this is not the only approach that was pursued. Many research teams worldwide adopted a bottom-up process, in an attempt to translate the systems biology approach into clinical practice (Bousquet et al., 2014; Wolkenhauer et al., 2014; Wang et al., 2015). Some envisaged a future model of Predictive, Preventive, Personalized and Participatory medicine (P4) based on the translation of systems biology, or as later referred systems medicine (Hood et al., 2012). While this approach holds the potential for huge impact, especially in relation to the discovery of new pharmaceutical compounds, in many cases there are knowledge gaps that make the clinical application difficult (Noble, 2003). One particularly important limitation is the ability to model the cell-tissue interaction, as was stressed in the 2009 workshop jointly organised by the United States Environmental Protection Agency and the European Commission7. Some authors have tried to bridge this with phenomenological models, such as the Effect Model Law (Boissel et al., 2013; J-P Boissel, 2015).
All these research activities embraced a scenario in which VPH models could be used not to enhance the clinical management of patients affected by particularly difficult pathologies, but rather to design and assess biomedical products. In 2011, the VPH Institute introduced the term in silico clinical trials (ISCT) to describe this type of activity.
In this document we define ISCT as the use of individualised computer simulation in the development or regulatory evaluation of a medical intervention.
The term individualised probably needs some further clarification. In most if not all ISCT applications the goal is to predict how a product will perform across a population, so why insist on the need for individualised models?
Most of the time a model captures one mechanistic theory, and in this sense is generic; however, it is parameterised to mimic each individual patient. In this sense it would be more correct to say that the model is generic and the parameters are patient-specific. But occasionally a complex model can be fully identified with direct measurements taken from individuals; in most cases some parameters are subject-specific while others are population-specific. In this roadmap we will refer to individualised or patient-specific models not in relation to how they are parameterised, but in relation to their predictive intent, ie, how they are validated. There are three possible expectations for such a model:
Over a cohort of N patients, for whom one can measure the quantity to be predicted, we consider a model validated if it returns a prediction within the distribution of measured values; in other words the model captures one generic behaviour considered representative of a member of that population.
Over the same cohort, the model predicts a central value of the distribution of measurements, typically an average value over the population.
The model is parameterised for each patient in the cohort, and its predictions are compared to the measurements for that individual.
Most predictive models available today are somewhere in between a) and c). So what really defines the Avicenna Community of Practice is the tendency toward c), the recognition that when possible a fully mechanistic, quantitative model capable to predict accurately for each individual member of the population would be superior to any other type of model. What we are proposing is an ideal, to which we should aim as a community; of course case by case there will be variation in how close we get to this ideal for a number of practical reasons including lack of measurements, lack of knowledge, computational complexity, etc.
This document aims to define the research and technological development roadmap needed to make this vision a tangible reality, much as the 2007 document did for VPH research. But it also aims to support the case for the creation of a novel industrial sector capable of providing technologies, consulting, and services for ISCT to the biomedical industry.
This new sector will emerge from two existing areas. The first is the clinical trials industry composed of Contract Research Organisations (CRO), research hospitals, and regulatory experts, which serves the biomedical industry in the design, execution, interpretation, and regulation of clinical trials. The second is the virtual prototyping industry, which provides in silico design and assessment for a variety of products in other industrial sectors such as aerospace and nuclear energy. We propose a new industrial sector that is built on expertise from these existing areas of industry with additional capabilities that are specific to the ISCT domain.
The birth of a service industry to support ISCT is vital for the rapid and widespread adoption of this novel approach. This roadmap will chart the ISCT territory not from a purely cultural point of view, but with guidance from a variety of industry experts, by assessing the barriers and challenges that we need to overcome for this industrial sector to thrive (see figure II-1).
Figure II- The new Community of Practice
II.2.The Avicenna consensus process
The process the Avicenna consortium used to develop this roadmap can be summarised in four steps:
Form a community of practice.
Capture the consensus of the experts within this community by repeating four times;
Poll the community using a formal process known as Alignment Optimisation;
Capture the consensus in drafts versions of the roadmap;
Organise small-group meetings to validate this draft, and brainstorm the next step.
Consolidate all the inputs in a final draft version of the roadmap.
Publicly validate the roadmap with all stakeholders, and present it for discussion at Event Five.
II.2.b.The formation of the community of practice