shopify analytics

In Silico Clinical Trials Service (ISCTS)

Virtual Patients and Virtual Doctors for in silico clinical trials

Overview

In Silico Clinical Trials Service (ISCTS)

Supporting Medical Scientists to Reduce Time, Costs and Risk of In Vivo Experiments

ISCTS is a suite of software services supporting medical scientists in carrying out clinical trials for new treatments completely in silico, that is by means of computer simulations over a population of virtual patients.

In silico clinical trials allow medical scientists to reduce and postpone invasive, risky, costly and time-consuming in vivo experiments on animals and humans to much later stages of the process, when a deeper knowledge of the effectiveness and side-effects of the new treatment has been acquired via simulations.

Virtual Doctors Treating Virtual Patients

In silico clinical trials are a disruptive key innovation and a highly supported research field in medicine, with a strong potential for improvement of clinical and therapeutical approaches as well as cost reduction in the development of new treatments.

By putting together a VPH model (and its associated population of Virtual Patients) and the model of a new treatment (Virtual Doctor), Paeon ISCTS allows medical scientists to evaluate how the treatment performs on the input population of Virtual Patients and to search for treatment variations optimised for a particular class of Virtual Patients.

In Silico Treatment Verification

ISCTS can be used to verify how a given treatment protocol (modelled as a Virtual Doctor) performs on a population of Virtual Patients.

ISCTS simulates the correct administration of the treatment protocol to all virtual patients and computes suitable statistics.

Actual fertility treatments currently in use at the Division of Reproductive Endocrinology of University Hospital Zurich have been successfully defined as Virtual Doctors and verified through ISCTS on the entire population of virtual patients associated to the Paeon VPH Model.

In Silico Treatment Optimisation

ISCTS can also be used to optimise the performance of a given treatment (modelled as a Virtual Doctor) on a certain class of Virtual Patients. To this end, ISCTS searches for variations of the input treatment model which maximise performance on a given set of Key Performance Indicators.

Variations with respect to the input treatment may differ in terms of, for example, the timings and administered doses in the various circumstances and/or for the patient classification criteria. Users can easily annotate the Virtual Doctor modelling the reference treatment to define which variations ISCTS will consider.

Beyond Fertility Treatments

Like all software services developed within Paeon, ISCTS is not restricted to fertility treatments, as it is ready to be used with other VPH models and other treatment protocol models.

Main Features

From In Vivo to In Silico Clinical Trials

During a classical in vivo clinical trial, medical scientists administer the treatment under test to a set of patients (volunteers). For each patient, an iterative process is devised, where the doctor, on the basis of some clinical measurements on the patient, decides the next clinical actions to perform, as defined by the treatment protocol.

At the end of the clinical trial, performance of the treatment protocol under test are analysed and suitable statistics are computed.

ISCTS allows medical scientists to shift this workflow into an in silico environment, by allowing them to automatically administer a model of the treatment protocol (a Virtual Doctor) on a set of Virtual Patients.

ISCTS supports medical scientists in performing in silico clinical trials with two main software services allowing, respectively, in silico verification and in silico optimisation of treatment protcols.

ISCTS services are computationally very intensive. To this end, suitable algorithms have been devised to take benefit from high-performance parallel computational infrastructures in the cloud.

ISCTS on UZH Fertility Treatments

During the project, three fertility treatment protocols used by University Hospital Zurich (UZH) in daily clinical practice have been modelled as Virtual Doctors.

Such treatment protocols have been verified and optimised in silico through ISCTS on the population of Virtual Patients associated to the Paeon VPH Model.

In Silico Verification of Treatment Protocols

When verifying a new treatment protocol in silico via ISCTS, the treatment protocol model (Virtual Doctor) is automatically administered to all virtual patient in the input population, also evaluating patient inclusion/exclusion criteria.

Beyond suitable statistics, ISCTS also provides the medical doctors with means to access the log of the (in silico) treatment administration for each Virtual Patient.

In Silico Verification of UZH Fertility Treatments

A Success Case

The figure shows an example of treatment (in silico) administration log for the UZH Long Treatment Protocol on a Virtual Patient satisfying the treatment inclusion criteria and for whom the treatment succeeds.

In particular, the figure shows:

  • on the left, the amount of FSH/LH-based drug (Merional) administered during the stimulation phase as dynamically adapted by the Virtual Doctor under verification (modelling the UZH Long protocol) depending on the patient reactions
  • in the centre, the time evolution of the most important hormones during the downregulation and the stimulation phases
  • on the right, the growth of the Virtual Patient's ovarian follicles during the stimulation phase.
As the hormone time evolutions satisfy all safety thresholds modelled within the Virtual Doctor and the number of mature follicles on day 16 satisfy the treatment success criteria (again modelled within the Virtual Doctor), for this Virtual Patient ISCTS declares success.

In Silico Verification of UZH Fertility Treatments

A Failure Case

Unfortunately the in silico verification of the Virtual Doctors modelling the UZH Long protocols did not succeed for all Virtual Patients. This is not surprising, as the state-of-the-art success rates of fertility treatments is far below 100% also in the most renowned fertility clinics as UZH.

The figure shows an example of treatment (in silico) administration log for the UZH Short Treatment Protocol on a Virtual Patient satisfying the treatment inclusion criteria and for whom the treatment fails.

In particular, on this Virtual Patient the treatment protocol (modelled as a Virtual Doctor) fails to make enough follicles become mature within the defined deadline (day 13 of the stimulation phase), although it tries hard (see left part) to push follicle maturation by constantly increasing the dose of the stimulation drug, after having verified that all safety thresholds regarding hormone blood levels are satisfied (see middle part).

By returning to the user such administration logs, ISCTS supports the medical scientists running the in silico clinical trials with effective means to analyse the reasons of the treatment protocol failures. The treatment model can then be improved for a set of Virtual Patients for whom it fails either manually or automatically by ISCTS itself, through the Treatment Optimisation service described next.

In Silico Optimisation of Treatment Protocols

When optimising a treatment protocol in silico via ISCTS, medical scientists start by annotating the reference treatment protocol model (Virtual Doctor) in order to define which variations ISCTS will consider, in terms of, for example, variations in drug doses, timings, patients classification criteria.

When launched on the annotated reference treatment model and on a population of Virtual Patients, ISCTS intelligently searches for variations of the treatment protocol (according to user annotations) and returns the set of Pareto Optimal variations, that is those performing better than others in the output set for at least one of the given Key Performance Indicators.

To enable further analyses, ISCTS also provides the medical doctors with means to access the logs of the (in silico) administration of all treatment variations for each Virtual Patient.

In Silico Optimisation of UZH Fertility Treatments

The Long Protocol

The figure shows the result of running in silico optimisation of the UZH Long Treatment Protocol on the entire population of Virtual Patients associated to the Paeon VPH model and satisfying the treatment inclusion criteria.

ISCTS was run on the Virtual Doctor modelling the reference UZH Long Treatment Protocol, suitably annotated to define the possible variations to be considered, as follows:

  • Uniform change in the administered doses of stimulation drug: ±37.5 IU (that is, ± 1 dose quantum)
  • Uniform change in the thresholds used to classify patients according to their anti-Muellerian hormone (AMH) level: ±3.5 pmol/l
  • Uniform change in the thresholds used to classify patients according to their Antral Follicle Count (AFC): ±3 follicles
  • Uniform change in the thresholds used to classify patients according to their age: ±2 years.
For example, the last item means that, while the reference treatment defines "young patients" has those aged between 35 and 37 years old, variations can redefine this class as 33-35, 34-36, etc., up to 37-39. For patient classes based on AMH level, AFC and age, the treatment protocol has different drug administration strategies.

ISCTS evaluated the considered treatment variations of the UZH Long Protocol using the following set of Key Performance Indicators (KPIs):

  • the ratio of the Virtual Patients (who represent classes of real patients) for whom the treatment variation succeeds
  • the average number of mature oocytes obtained from such Virtual Patients at the end of the stimulation phase
  • the reduction of the amount of the administered stimulation drug with respect to the reference treatment
  • the ratio of the Virtual Patients for whom the treatment variation is applicable (as a treatment variation can redefine patients inclusion/exclusion criteria).

The final set of Pareto Optimal treatment variations computed by ISCTS is shown in the figure. It can be observed how the reference treatment somewhat averages its performance on all KPIs, although there exist interesting treatment variations seeming to have better performance on a subset of KPIs. Some of such variations are currently under investigation at UZH for their possible deployment in clinical practice.