Table of Contents

1. Definition

In silico trials are the use of individualised computer simulation in the development or regulatory evaluation of a medicinal product, medical device, or medical intervention¹. In place of a physically enrolled human cohort, an in silico trial enrols a population of virtual patients generated from mechanistic, multiscale models of human physiology and pathophysiology, and the resulting simulated evidence is treated as a complement or substitute for experimental evidence^{1 2}.

The biological basis of an in silico trial is the set of physiological and pathophysiological mechanisms encoded in the underlying model. Depending on the context of use, these mechanisms operate at molecular, cellular, tissue, organ, or whole-body scale. The Archimedes model of diabetes, for example, is a system of differential equations covering the pertinent organ systems with more than fifty continuously interacting variables that represent anatomy, pathophysiology, diagnostic tests, treatments, and clinical outcomes³. The University of Virginia/Padova type 1 diabetes simulator encodes glucose–insulin dynamics for one hundred in silico adults, one hundred adolescents, and one hundred children⁴. Cardiac safety simulations encode ventricular cardiomyocyte electrophysiology, including the ion-channel currents that govern repolarisation⁵, whereas stroke models combine patient-specific vascular anatomy with haemodynamic and thrombus mechanics^{6 7}. Population-level variability is introduced by sampling model parameters from distributions estimated on clinical and epidemiological data, which produces virtual cohorts that reproduce the heterogeneity of real patient populations².

2. Purpose

The primary value proposition of in silico trials is that they permit the safety, efficacy, and performance of a biomedical product to be investigated exhaustively, rapidly, and at low marginal cost, without exposing human subjects or animals to risk^{1 2}. They allow investigators to interrogate scenarios that would be impractical, unethical, or prohibitively expensive in a physical trial, including extreme dosing regimens, rare subpopulations, and device configurations that would never receive ethical approval for first-in-human testing².

In silico trials improve upon the legacy model of preclinical evaluation, which relies heavily on animal experimentation. Genomic responses in mouse models of burns, sepsis, and trauma have been shown to correlate poorly with the corresponding human inflammatory conditions, which undermines the translational value of animal data for human disease⁸. Because in silico trials are parameterised directly on human physiological, clinical, and epidemiological data, their predictions refer to human biology by construction rather than by cross-species extrapolation^{2 3}. This human relevance has regulatory precedent: in 2008 the United States Food and Drug Administration accepted the UVA/Padova type 1 diabetes simulator as a substitute for preclinical animal trials in the evaluation of closed-loop insulin delivery systems^{9 4}, and the Comprehensive in Vitro Proarrhythmia Assay initiative uses in silico cardiac electrophysiology models within a paradigm intended to eliminate the need for the clinical thorough-QT study⁵.

3. Methodology

The construction of an in silico trial begins with the definition of a context of use, namely a precise statement of the regulatory or development question the simulated evidence will answer, followed by a risk analysis that determines the credibility the model must achieve for that question¹⁰. The model itself is then built using a formalism appropriate to the biology being represented: systems of ordinary and partial differential equations for physiological dynamics³, multiscale computational models that couple vascular anatomy, haemodynamics, and thrombus behaviour for stroke treatment⁶, agent-based models for immune-system processes¹¹, and surrogate models distilled from high-fidelity finite-element simulations where repeated evaluation of a detailed model would be computationally prohibitive⁷.

The models are parameterised and calibrated on experimental and clinical data, including cohort studies, clinical trial results, molecular and electrophysiological assays, and medical imaging^{6 3}. A virtual patient population is generated by sampling the calibrated parameter distributions, and the trial protocol, including arms, randomisation, inclusion criteria, and endpoints, is implemented computationally so that the simulated trial replicates the conduct of the physical trial it is designed to inform^{4 7}. Credibility is established through a verification, validation, and uncertainty quantification process, comprising code verification, solution verification, validation against independent experimental or clinical data, and quantification of the uncertainty in model predictions relative to predefined acceptability thresholds¹⁰. For medical devices, the ASME V&V 40 standard provides a risk-based framework for assessing the relevance and adequacy of these verification and validation activities in proportion to the consequence of the decision the model supports¹².

4. Performance

The empirical performance of in silico trials is demonstrated most directly by their capacity to reproduce the results of completed clinical trials. The Archimedes diabetes model was validated by simulating randomised controlled trials and comparing the outcomes calculated by the model with the outcomes observed in the trials, and it was concluded that the model is a realistic representation of the anatomy, pathophysiology, treatments, and outcomes pertinent to diabetes and its complications^{3 13}. The 2013 version of the UVA/Padova type 1 diabetes simulator was shown to match the glucose traces of an actual clinical trial of closed-loop insulin delivery⁴. In the in silico thrombectomy framework, a validation trial demonstrated that a surrogate thrombectomy model could reproduce a recanalisation rate similar to that of the corresponding real intervention, and subsequent exploratory trials distinguished the performance of commercially available stent retrievers according to the fibrin composition of the occluding thrombus⁷.

Reproducibility and credibility are governed by the verification, validation, and uncertainty quantification methodology and, for devices, by the risk-based credibility assessment of ASME V&V 40, which ties the required rigour of validation evidence to the regulatory risk of the decision being made^{10 12}. Current technical boundaries remain substantial. The predictive value of a simulated trial is bounded by the completeness of the underlying biological knowledge and by the quality and availability of the calibration and validation data¹⁰. A recent systematic review of in silico clinical trials for medical devices concluded that regulatory adoption remains fragmented, and that standardised validation frameworks, regulatory standards, interdisciplinary cooperation, and clear guidelines are required before in silico evidence can be accepted routinely rather than case by case¹⁴.

5. Applications

Primary research applications include the preclinical evaluation of closed-loop insulin delivery systems for type 1 diabetes, for which the UVA/Padova simulator has been accepted by the Food and Drug Administration as a substitute for animal trials^{9 4}; cardiac safety pharmacology, in which the Comprehensive in Vitro Proarrhythmia Assay combines ion-channel assays with in silico electrophysiology models to predict proarrhythmic risk and to reduce reliance on the thorough-QT clinical study⁵; and the evaluation of endovascular thrombectomy devices for acute ischaemic stroke, in which in silico trials are used to compare device performance across virtual patient subpopulations and to improve patient stratification for subsequent clinical trials^{6 7}.

Industrial and translational applications extend to vaccine evaluation, in which agent-based immune-system simulators have been used to test candidate COVID-19 vaccines in virtual populations¹¹, and to the simulation of comparator arms, in which the placebo arm of a clinical trial for osteoporosis treatments has been reproduced in silico to support the assessment of treatments for hip fracture prevention¹⁵. The standardised contexts of use are defined through the regulatory frameworks for computational modelling and simulation, notably the ASME V&V 40 standard for medical devices¹² and the verification, validation, and uncertainty quantification methodology for biomedical products¹⁰. Deployment remains at an early and expanding stage: in silico trials currently supplement rather than replace conventional clinical evidence in most indications, and regulatory acceptance is granted case by case, although precedents now exist in diabetes technology, cardiac safety, and medical device evaluation^{14 1}.

6. Variations

• Model-Informed Drug Development Trials: Pharmacometric and disease-progression simulations used to select doses, design trial protocols, and predict clinical outcomes during drug development; the Archimedes diabetes model and the cardiac safety models of the CiPA initiative are representative exemplars^{3 5}.

• In Silico Medical Device Trials: Simulations in which the performance of a medical device is evaluated on a virtual patient population under a risk-based credibility framework; applied to endovascular thrombectomy devices and closed-loop insulin delivery systems^{12 7 9}.

• Agent-Based Immunological Trials: Trials built on agent-based models of the immune system, in which virtual patients and simulated immune processes are used to evaluate vaccines and immunotherapies, as demonstrated by the in silico trial of candidate COVID-19 vaccines on the UISS platform¹¹.

• Hybrid In Vitro–In Silico Trials: Paradigms that combine in vitro assays with in silico models so that experimentally measured inputs, such as ion-channel blockade, drive model-based prediction of clinical risk; the Comprehensive in Vitro Proarrhythmia Assay is the principal exemplar⁵.

• Synthetic Control Arm Trials: Trials in which the comparator or placebo arm is generated in silico from validated mechanistic models and historical patient data rather than by enrolling control subjects, as demonstrated by the simulation of the placebo arm for osteoporosis treatments to prevent hip fractures¹⁵.