Ce texte est une introduction à la modélisation mathématique en cancérologie, et en particulier à la modélisation des gliomes, qui sont une forme particulière de tumeurs cérébrales. L'objectif est d'illustrer comment une modélisation mathématique peut contribuer à répondre à des questions médicales et à mieux comprendre comment traiter des tumeurs cancéreuses.

The PASADENA study is an ongoing Phase II, multicenter, randomized, double-blind, placebo-controlled trial evaluating the safety and efficacy of intravenous prasinezumab, administered every 4 weeks, in early-stage Parkinson’s disease (PD). During the double-blind study period, prasinezumab-treated individuals showed less progression of motor signs (Movement Disorder Society-sponsored revision of the Unified Parkinson's Disease Rating Scale [MDS-UPDRS] Part III). We evaluated here whether the effect of prasinezumab on motor progression, assessed as change in MDS-UPDRS Part III in OFF- and ON-state, and MDS-UPDRS Part II scores was sustained over 4 years from the start of the trial. We compared participants enrolled in the PASADENA open-label extension (OLE) to an external comparator arm derived from the Parkinson's Progression Markers Initiative (PPMI) observational study. Both PASADENA delayed- (n = 94) and early-start (n = 177) groups showed a slower decline (less increase in score) on MDS-UPDRS Part III in OFF- (-51% for the delayed-start group and − 65% for the early-start group) and ON-state (-94% for the delayed-start group and − 118% for the early-start group), and on MDS-UPDRS Part II (-48% for the delayed-start group and − 40% for the early-start group), compared with the PPMI external comparator (n = 303). This exploratory analysis, which requires confirmation in future studies, suggests that the effect of prasinezumab in slowing motor progression in PD may be sustained long-term.

BR, GP, and AS are employees of F. Hoffmann-La Roche Ltd. All other authors declared no competing interests for this work. NK and NI are employees of PumasAI who are involved in the development of DeepNLME-based software.

Background: Objectively measuring Parkinson’s disease (PD) signs and symptoms over time is critical for the successful development of treatments aimed at halting the disease progression of people with PD. Objective: To create a clinical trial simulation tool that characterizes the natural history of PD progression and enables a data-driven design of randomized controlled studies testing potential disease-modifying treatments (DMT) in early-stage PD. Methods: Data from the Parkinson’s Progression Markers Initiative (PPMI) were analyzed with nonlinear mixed-effect modeling techniques to characterize the progression of MDS-UPDRS part I (non-motor aspects of experiences of daily living), part II (motor aspects of experiences of daily living), and part III (motor signs). A clinical trial simulation tool was built from these disease models and used to predict probability of success as a function of trial design. Results: MDS-UPDRS part III progresses approximately 3 times faster than MDS-UPDRS part II and I, with an increase of 3 versus 1 points/year. Higher amounts of symptomatic therapy is associated with slower progression of MDS-UPDRS part II and III. The modeling framework predicts that a DMT effect on MDS-UPDRS part III could precede effect on part II by approximately 2 to 3 years. Conclusions: Our clinical trial simulation tool predicted that in a two-year randomized controlled trial, MDS-UPDRS part III could be used to evaluate a potential novel DMT, while part II would require longer trials of a minimum duration of 3 to 5 years underscoring the need for innovative trial design approaches including novel patient-centric measures.

Legends for supplementary Figures S1, S2 and S3

In this perspective, we briefly review the stateof-the-art covering the interface between quantitative systems pharmacology (QSP) and artificial intelligence (AI) and machine learning (ML); in particular, how AI/ML are conceived and used as many attempts to address methodological pain points of QSP.In a second part, we invite the reader to step out from this discipline-centric view and discuss a paradigm shift consisting of repurposing AI into QSP.In 2022, the Journal of Pharmacokinetic and Pharmacodynamic published a special issue edited by Cho, Zhang, and Bonate with 10 scientific articles illustrating ways for coupling quantitative systems pharmacology (QSP) and artificial intelligence (AI).The review from Zang et al. summarizes the current state-of-the-art 1 and explains that AI/machine learning (ML) are currently used for four main applications related to QSP: parameter estimation, model structure, complexity reduction, and virtual population generations (see Figure 1).What may strike the reader is that -although there are good reasons to think that improving on such technical domains will increase the impact of QSP -these four domains speak toward technical dimensions of QSP as a discipline.Estimation of parameters, writing down optimal model structures, reducing model complexity, or properly generating virtual population are terms which will talk to modeling experts but probably only to them.Parameter estimation is known to be an issue for QSP because the models are often too large with respect

PDF file, 258K, Figure S 1: Stepwise procedure applied to build the model structure (left panel); non-specific and specific evaluation criteria for model selection (right panel). Figure S 2: Results of the sensitivity analysis for the model on the 21 patients treated with PCV. The analysis consists of repeating the estimations, leaving out one patient's data at a time. Parameter estimates are represented with histograms indicating the number of patients (y-axis) for each value of the parameter (x-axis). It appears that no single patient substantially influenced the estimation. Figure S 3: Comparison of parameter estimates in the 21 patients treated with PCV when the variability of KDE is fixed to 0 or fixed to 70%. Figure S 4: Comparison of parameter estimates between the PCV dataset (n = 21) and the pooled dataset (n = 40, comprising patients from the PCV and radiotherapy datasets). The probability density functions calculated using the standard errors on the estimates are represented in continuous and dashed lines for the PCV and pooled datasets, respectively. They show that with the exception of two treatment-specific parameters (the efficacy parameter γ and the transfer rate from Qp to P, ), the parameters are homogeneous. Figure S 5: MTD observations (symbols) and individual predictions (solid line) for six individuals sampled from the PCV dataset. Included is the 90% confidence interval around the individual predictions obtained by simulations using the standard errors of the empirical Bayes estimates.

To support further development of model-informed drug development approaches leveraging circulating tumor DNA (ctDNA), we performed an exploratory analysis of the relationships between treatment-induced changes to ctDNA levels, clinical response and tumor size dynamics in patients with cancer treated with checkpoint inhibitors and targeted therapies. This analysis highlights opportunities for pharmacometrics approaches such as for optimizing sampling design strategies. It also highlights challenges related to the nature of the data and associated variability overall emphasizing the importance of mechanistic modeling studies of the underlying biology of ctDNA processes such as shedding, release and clearance and their relationships with tumor size dynamic and treatment effects.

Model-based approaches are instrumental for successful drug development and use. Anchored within pharmacological principles, through mathematical modeling they contribute to the quantification of drug response variability and enables precision dosing. Reinforcement learning (RL)-a set of computational methods addressing optimization problems as a continuous learning process-shows relevance for precision dosing with high flexibility for dosing rule adaptation and for coping with high dimensional efficacy and/or safety markers, constituting a relevant approach to take advantage of data from digital health technologies. RL can also support contributions to the successful development of digital health applications, recognized as key players of the future healthcare systems, in particular for reducing the burden of non-communicable diseases to society. RL is also pivotal in computational psychiatry-a way to characterize mental dysfunctions in terms of aberrant brain computations-and represents an innovative modeling approach forpsychiatric indications such as depression or substance abuse disorders for which digital therapeutics are foreseen as promising modalities.

LDH (upper panels) and NSE (lower panels) individual log-transformed concentration-time profiles from training dataset (left panels) and external dataset (right panels). Thick lines are loess smooth of the data and dashed gray lines represent the threshold normal values of each biomarker.

The promise of transforming digital technologies into treatments is what drives the development of digital therapeutics (DTx), generally known as software applications embedded within accessible technologies-such as smartphones-to treat, manage, or prevent a pathological condition. Whereas DTx solutions that successfully demonstrate effectiveness and safety could drastically improve the life of patients in multiple therapeutic areas, there is a general consensus that generating therapeutic evidence for DTx presents challenges and open questions. We believe there are three main areas where the application of clinical pharmacology principles from the drug development field could benefit DTx development: the characterization of the mechanism of action, the optimization of the intervention, and, finally, its dosing. We reviewed DTx studies to explore how the field is approaching these topics and to better characterize the challenges associated with them. This leads us to emphasize the role that the application of clinical pharmacology principles could play in the development of DTx and to advocate for a development approach that merges such principles from development of traditional therapeutics with important considerations from the highly attractive and fast-paced world of digital solutions.

Prediction accuracy at the individual level of RECIST model: Predicted probabilities of progression (sorted in ascending order) during treatment (SCAN 1 and SCAN 2, in green and orange) and at the first two follow-up CT scans (in blue and red) for patients in internal dataset (left) and external dataset (right) obtained from all data available (i.e. biomarker and CT scans outcome). Empty rhombuses represent patients that did not have disease progression in the CT scan and filled squares depict patients with observed disease progression at each CT scan. Dotted and solid lines depict the 25th and 75th quantiles corresponding to the individual posterior distribution.

Supplementary Figure S3: Comparison of QW versus Q2W CEA-IL2v dosing schedule. Top panel: Predicted pharmacokinetic population profile from cycle 1 to 4 at 20 mg QW (left); Predicted corresponding tumor uptake (right). Bottom panel: same two graphics for Q2W schedule (pharmacokinetic - left, and tumor uptake - right).

Extending the potential of precision dosing requires evaluating methodologies offering more flexibility and higher degree of personalization. Reinforcement learning (RL) holds promise in its ability to integrate multidimensional data in an adaptive process built toward efficient decision making centered on sustainable value creation. For general anesthesia in intensive care units, RL is applied and automatically adjusts dosing through monitoring of patient's consciousness. We further explore the problem of optimal control of anesthesia with propofol by combining RL with state-of-the-art tools used to inform dosing in drug development. In particular, we used pharmacokinetic-pharmacodynamic (PK-PD) modeling as a simulation engine to generate experience from dosing scenarios, which cannot be tested experimentally. Through simulations, we show that, when learning from retrospective trial data, more than 100 patients are needed to reach an accuracy within the range of what is achieved with a standard dosing solution. However, embedding a model of drug effect within the RL algorithm improves accuracy by reducing errors to target by 90% through learning to take dosing actions maximizing long-term benefit. Data residual variability impacts accuracy while the algorithm efficiently coped with up to 50% interindividual variability in the PK and 25% in the PD model's parameters. We illustrate how extending the state definition of the RL agent with meaningful variables is key to achieve high accuracy of optimal dosing policy. These results suggest that RL constitutes an attractive approach for precision dosing when rich data are available or when complemented with synthetic data from model-based tools used in model-informed drug development.

The availability of multidimensional data together with the development of modern techniques for data analysis represent an exceptional opportunity for clinical pharmacology. Data science—defined in this special issue as the novel approaches to the collection, aggregation, and analysis of data—can significantly contribute to characterize drug‐response variability at the individual level, thus enabling clinical pharmacology to become a critical contributor to personalized healthcare through precision dosing. We propose a minireview of methodologies for achieving precision dosing with a focus on an artificial intelligence technique called reinforcement learning, which is currently used for individualizing dosing regimen in patients with life‐threatening diseases. We highlight the interplay of such techniques with conventional pharmacokinetic/pharmacodynamic approaches and discuss applicability in drug research and early development.

The availability of multidimensional data together with the development of modern techniques for data analysis represent an exceptional opportunity for clinical pharmacology. Data science-defined in this special issue as the novel approaches to the collection, aggregation, and analysis of data-can significantly contribute to characterize drug-response variability at the individual level, thus enabling clinical pharmacology to become a critical contributor to personalized healthcare through precision dosing. We propose a minireview of methodologies for achieving precision dosing with a focus on an artificial intelligence technique called reinforcement learning, which is currently used for individualizing dosing regimen in patients with life-threatening diseases. We highlight the interplay of such techniques with conventional pharmacokinetic/pharmacodynamic approaches and discuss applicability in drug research and early development.

The assessment of a therapeutic window for pharmaceutical compounds requires the integration of multidimensional data. In this perspective, we highlight how the increased volume of medical and healthcare data, and, in particular, real-world data (RWD), can be integrated within physiologically based modeling framework. We illustrate this perspective view with specific case studies related to pharmacokinetic and cardiac safety predictions.

Waterfall plots are used to describe changes in tumor size observed in clinical studies. They are frequently used to illustrate the overall drug response in oncology clinical trials because of its simple representation of results. Unfortunately, this visual display suffers a number of limitations including (1) potential misguidance by masking the time dynamics of tumor size, (2) ambiguous labelling of the y-axis, and (3) low data-to-ink ratio. We offer some alternatives to address these shortcomings and recommend moving away from waterfall plots to the benefit of plots showing the individual time profiles of sum of lesion diameters (according to RECIST). The spider plot presents the individual changes in tumor measurements over time relative to baseline tumor burden. Baseline tumor size is a well-known confounding factor of drug effect which has to be accounted for when analyzing data in early clinical trials. While spider plots are conveniently correct for baseline tumor size, they cannot be presented in isolation. Indeed, percentage change from baseline has suboptimal statistical properties (including skewed distribution) and can be overly optimistic in favor of drug efficacy. We argued that plots of raw data (referred to as spaghetti plots) should always accompany spider plots to provide an equipoised illustration of the drug effect on lesion diameters.

Immunogenicity is a major challenge in drug development and patient care. Currently, most efforts are dedicated to the elimination of the unwanted immune responses through T-cell epitope prediction and protein engineering. However, because it is unlikely that this approach will lead to complete eradication of immunogenicity, we propose that quantitative systems pharmacology models should be developed to predict and manage immunogenicity. The potential impact of such a mechanistic model-based approach is precedented by applications of physiologically-based pharmacokinetics.

Recent advances in machine learning (ML) have led to enthusiasm about its use throughout the biopharmaceutical industry. The ML methods can be applied to a wide range of problems and have the potential to revolutionize aspects of drug development. The incorporation of ML in modeling and simulation (M&S) has been eagerly anticipated, and in this perspective, we highlight examples in which ML and M&S approaches can be integrated as complementary parts of a clinical pharmacology workflow.