The Effect Model - Nova Discovery



From a law of nature to the universal standard

Discovered by Pr. Jean-Pierre BOISSEL, co-founder and Chief Scientific Officer of NOVADISCOVERY, a new standard has emerged from the Effect Model (EM) law, which rationalizes in silico assay results and de-risks decision-making.

Translational in its scope of application, this methodology addresses any question (therapeutic or diagnostic potential of a new target, efficacy alone, compared or combined of a drug or a drug candidate, determination of the optimal dose, identification of a subgroup of responders...) applied to any study object (biomarker, therapy, disease, single patient and whole population).

The EM is the result of a seemingly simple observation: when a group of individuals is affected by a disease, each member of this group is at risk of variable evolution (some of them will soon confront the disease-related events, while in other individuals these events will be delayed or less frequent). Therefore, these individuals can be distributed along an axis from 0 to 1 with, for each of them, a probability of evolution over time. This is true even for deadly diseases whose time of evolution varies from one individual to another. In applying a given therapy to these same individuals intended to prevent or delay the progression of the disease, the probability of its development will thus be altered in a way which is patient-specific.

The EM defines this relationship between the probability of diseases-related events (tumor progression, side effects, death...) without a therapy (Rc, "control" risk) and with a therapy (Rt, the risk modified by treatment t). In turn, it defines for each individual the predicted therapeutic benefit, or Absolute Benefit AB, resulting from the difference between these two probabilities (AB = Rc - Rt). This value AB measures the absolute benefit of the therapy considered for an individual, a group of patients or a population of interest. With the Effect Model, treatment efficacy becomes a quantified and predictable metric which can be benchmarked against competing therapies.

A two step process

In practice, NOVADISCOVERY applies this methodology by combining a Virtual Population of patients (the group or population of interest) with a Formal Model of the disease and proposed therapy (which encapsulates all useful biological and clinical knowledge) or with statistical data (results of clinical trials, data from cohorts and epidemiological studies). The figure below shows the two-step process implemented by NOVADISCOVERY:  

  • The disease model is first applied to the virtual population of patients to determine the distribution of the risks of the clinical event of interest without treatment Rc (x-axis).
  • The disease model is then combined with the sub-model of the therapy under investigation (marketed drug, drug candidate, therapeutic target), and applied to the same virtual population in order to determine the changes in the distribution of the clinical event risk induced by treatment Rt in the population (y-axis). The line bisector in the Rc,Rt plan marks the threshold of efficacy (or treatment neutrality frontier) of a given treatment for each patient in this population

As a result, the application of this methodology yields the two metrics expected from the assay:

  • The Absolute Benefit of the treatment: a positive value (Rt<Rc) demonstrates the beneficial effect of the treatment, a zero value (Rc = Rt) the lack of benefit, and a negative value (Rt> Rc) signals the harmful nature of the treatment.
  • The Number of Prevented Events (NPE): the total number of avoided clinical events thanks to a treatment for a given population of patients, which is simply the sum of all ABs over the population under investigation.
The Effect Model
From the early phases of research down to marketed drugs, the Effect Model enables rational decision-making on the basis of the optimal collective and individual benefit (the number of prevented clinical events). This is why it has been designated by the European Agency for Health Technology Assessment (EUnetHTA) as the reference standard to translate clinical trials results into real-world patient outcomes.


Multi-scale disease models

NOVADISCOVERY designs and maintains a library of multi-scale disease models in selected therapeutic areas (notably in oncology and immuno-oncology). Each model encapsulates knowledge extracted from the scientific literature to describe the overall mechanisms thought to play a role in the pathophysiological process, from genes to tissues, organs and ultimately populations.  

In practice, the disease model takes two distinct forms: the Knowledge Model and the Formal Model. The former is built from the careful curation of information extracted from original articles, assembled on GITHEALTH as a state-of-the-art review with direct links to each primary source in order to ensure full transparency and auditability. The latter is the mathematical and computational translation of the Knowledge Model.

Mechanistic models can be updated over time as new knowledge emerges on the pathophysiology. Furthermore, each piece of knowledge incorporated into our models is traceable back to the primary source (generally an assertion extracted from a scientific article). The rigorous curation of assertions, through the application of a strength of evidence grid, enables the formal assessment of an assertion’s reliability for predictive modeling purposes.


Accounting for within- and between-patient variability

In contrast to alternative modeling and simulation approaches experimented in biomedical sciences, the Effect Model does not make the implicit (and eminently incorrect) assumption of the existence of a “standard” representative patient. While the disease model is deterministic, within- and between-patient genotypic and phenotypic variability is accounted for in our Virtual Population objects.

The Virtual Population represents the population or the group of interest. Each virtual patient of the population is characterized by a vector of descriptors that translates biological and environmental parameters involved in the course of the disease and interactions between the drug or any intervention and the body.

This Virtual Population can be realistic, i.e., derived from a real population. In such a case, some of or all descriptors are built from epidemiological and/or clinical trial data. Virtual Populations can also be designed through the translation of model parameters into virtual patient descriptors. In this case, their distributions are drawn from knowledge extracted from the literature or based on reasonable assumptions.