TL; DR

1. Currently, MIDD ideally remains centred on in vivo clinical trials, which are the ultimate proof. In the future, however, it is plausible that in vivo clinical trials will move to the role of a model validation instrument, while the essential evidence supporting scientists’, regulators’ and doctors’ decisions will come from simulations obtained with validated models.

2. With the use of in silico, R&D's time and cost will be reduced and less patients have to be recruited for RCTs.

3. To accelerate adoption by the scientific community, a handful of challenges need to be addressed. Besides technical challenges that are more or less fixed, regulators need to formalize their thoughts on in silico clinical trials and accept that limitation in knowledge is a major obstacle in using models.

Reduced time and cost

The in silico clinical trial paradigm, grounded in the EM methodology, is a viable alternative to address the issues raised by the decreasing efficiency of new therapeutic R&D. Multi-scale mechanistic models of selected diseases, applied to VPs of interest, combine knowledge and data to account for the complexity of the biology of life and illnesses.

Running in silico clinical trials is a time- and cost-effective way of exploring the entirety of potential targets and target combinations, such that those candidates eventually transitioned into clinical development have a higher likelihood of demonstrating efficacy on clearly characterized responders subgroups. Back-of-the-envelope calculations make a strong case in favour of the in silico paradigm. Based on Bloomberg’s estimation of $1.3bil of spending on 78 clinical development programs by the four largest drugmakers in immuno-oncology (Bristol-Myers Squibb, Merck & Co, Roche, AstraZeneca) and a total number of more than five thousand combination pairs, the total amount necessary to explore all of these options would represent a staggering $100bil, which is equivalent to circa 0.5% of the US national debt. In comparison, the total cost of this systematic exploration in silico would amount to $20mil, covering the cost of developing the necessary models of cancer types. While this comparison is partly an underestimation (real trials would still need to be financed to confirm in silico findings), cost savings would nevertheless be substantial and the capital deployed more efficiently on the most beneficial candidates. If a cost-effective solution exists to explore the entirety of potential immunotherapeutic combinations, we owe it to patients to adopt it. These remarks apply as well to the Covid-19, with the vaccine issues added to therapies.

Furthermore, virtual patients can ameliorate the problem of patient recruitment, i.e. in larger trials, rare diseases or ethically challenging trial designs. The careful characterization of responders and size of efficacy will work in favour of the reduction of the number of subjects (or statistical units) to be recruited for the same statistical significance.

The cost of implementing this new paradigm of in silico R&D is low in comparison to the cost of late stage failure in phase 2 or 3 for a single project[1] [2]. Scannell and Bosley have demonstrated by simulation that a mere 0.1 absolute increase in the correlation value between in vitro or in vivo test output and clinical outcomes in man would result in at least a tenfold improvement in R&D efficiency [3]. The in silico paradigm’s cost/benefit ratio alone is a compelling reason for widespread adoption.

Perspectives

The approach described above has established in silico clinical trials as an attractive and viable solution to advance clinical research in therapeutics, including the ever increasing needs to combine therapies. Proofs of concept were drawn from a variety of therapeutic areas and in a number of different stages of the typical R&D process. Early success points to a widespread acceptance in the near future. To accelerate adoption by the scientific community, a handful of challenges need to be addressed. First and foremost, regulators need to formalize their thoughts on in silico clinical trials.

Stakeholder engagement

The application of mathematical modelling approaches to pharmaceutical R&D has been advocated since the early 2000s, yet industry has been slow to adopt them. Hood and Perlmutter concluded that these approaches would revolutionize the way new medicines are developed in a more personalized framework, but questioned who would lead this change and how industry would explore it[4]. Examples demonstrating that computer modelling can have an impact on the drug R&D pipeline were the discovery of EGFR inhibitors’ mode of action in cancer and the application of predictive biosimulation in planning clinical studies[5] [6]. Building on this emerging evidence of utility, one of a series of reports by PricewaterhouseCoopers analysed the drug R&D process and advocated the need to make research more predictive, with “virtualisation” using modelling and simulation systems supporting the “seismic shift” that was considered necessary to address the prevailing challenges[7]. More recently, modelling and simulation was employed to study efficacy, safety and dosing schedules for rivaroxaban, a novel anticoagulant therapy[8] and to find the best regimen for novel classes of anticancer drugs [9].

These developments are encouraging, with an increased interest in model-based drug development and quantitative systems pharmacology, both of which extend beyond the classical PKPD modelling that is well established in the industry by adopting more mechanistic elements. This has generated growing interest from regulatory agencies[10] [11] [12]. The US Food and Drug Administration (FDA) is moving toward accepting simulation-based evidence [13] and the European Medicine Agency (EMA) is expected to follow suit. Pan-European initiatives are increasingly visible. After the successful completion of the European Commission-funded project “Avicenna: A Strategy for In Silico Clinical Trials'' and the subsequent publication of the Roadmap that emerged, discussions with the Commission have led to the recent birth of the Avicenna Alliance for Predictive Medicine[14]. This initiative brings together expertise from academia, in the form of the Virtual Physiological human Institute (VPHi[15]), and a wide representation of industries, to build on the outcome from the project. At the European Commission level, interest in in silico clinical trials is evidenced by the publication of a specific call for projects (Horizon 2020 Health PM16 “In-silico trials for developing and assessing biomedical products“). Further, it is also recognized that there is a need for expert views in this area to inform the development of policy frameworks where currently none exists, as these technologies increasingly are brought to bear in the marketplace. This is a key function of the Avicenna Alliance, working closely with the European Commission and building connections with similar activities in the USA and Asia Pacific region to ensure as far as is possible the harmonization of policy across territories ab initio. It is anticipated that these initiatives will eventually result in the publication of formal guidelines by the FDA, as well as the European Commission and the EMA, which will help to build confidence and accelerate broader adoption.

On the payer front, the EM has been endorsed by the European Network for Health Technology Assessment (EunetHTA) as the only reliable methodology to explore the relation between baseline risk and treatment effect, to select patients to be treated and enable the translation of clinical efficacy data into real world outcomes[16].

Where are we and a look at the future

The approach described in this document proposes M&S as a viable solution to address the current challenges of drug R&D. Proofs of concept exist that M&S can make a difference in a variety of therapeutic areas and in a number of different stages of the typical R&D process[17]. However, it remains to demonstrate that full MIDD can do better than the traditional approach, decrease the number of failures, save money and reduce ethical losses. To accelerate adoption by the scientific community, a handful of challenges need to be addressed. Beside technical challenges (see Table 6) that actually are more or less fixed (solutions exist for all of those listed in Table 6, the details of which are beyond the scope of this document, however they can be improved) intellectual barriers need to be addressed. First, the assumption that limitation in knowledge is a major obstacle in using models. Second, regulators need to formalize their thoughts in in silico clinical trials.

Currently, MIDD ideally remains centred on in vivo clinical trials, which are the ultimate proof. In the future, however, it is plausible that in vivo clinical trials will move to the role of a model validation instrument, while the essential evidence supporting scientists’, regulators’ and doctors’ decisions will come from simulations obtained with validated models.

Table 6: Technical Hurdles

• Adjusting the scope of the knowledge to be captured in the model in order to best address the problem of interest
• Choosing the appropriate level of granularity
• Identifying and gathering the relevant documentation for validation (knowledge and data)
• Mixing data from various origin/quality in order to built VP
• Constructing the joint distribution of the VP
• Establishing links between VP descriptors derived from model parameters and real recorded patients’ characteristics
• Seeking for a compromise between level of details in the computational model and available mathematical methods and computational facilities while keeping the research objective reachable
• Finding a solution for the sloppiness issue that would not impair the accuracy of the prediction
• Establishing validation scales/scores that enable reliable conclusion according to the modelling objective(s)
• Accounting for the variability of knowledge pieces integrated in the model as it is captured in the VP in the prediction marking of uncertainty
• Accounting for the strength of knowledge pieces integrated in the model pieces of knowledge strength of evidence in the prediction marking of uncertainty
• Selecting simulation outputs that help in interpreting the prediction(s)
• Ensuring transparency throughout the whole process, from prediction back to knowledge
• Accounting for and choosing the time scaling for submodel integration
• Choosing "time step" for solving equation