Samer Mouksassi 1
1 UCSF (San Francisco, USA)
Objectives
Pharmacokinetic/pharmacodynamic (PK/PD) modeling outputs, including goodness-of-fit (GOF) plots, visual predictive checks (VPCs), forest plots, and exposure–response analyses, are traditionally delivered as static figures generated through scripted workflows. While reproducible, static outputs limit interactive interrogation of model behavior, dynamic stratification of subpopulations, and exploratory data analysis by non-modeling stakeholders. Recent advances in web-based visualization frameworks and AI-assisted coding tools have substantially reduced the technical barriers to developing interactive applications. This work reviews the application of dynamic visualization tools across the PK/PD workflow and evaluates their potential to enhance exploratory analysis, model diagnostics, and regulatory communication.
Methods
A structured review of open-source interactive visualization frameworks was conducted, drawing on real project experience across three workflow domains:
(1) exploratory data analysis (EDA) and data quality review;
(2) structural and covariate model diagnostics, including VPCs and covariate effect visualization;
(3) reporting and communication to support dose selection and regulatory submissions.
Tools were evaluated based on rendering performance, degree of linked interactivity, integration with R and nonlinear mixed-effects platforms (NONMEM, Phoenix, Monolix, nlmixr2), reproducibility considerations (version control and data provenance), and feasibility for regulatory-facing deliverables.
Results
Modern interactive visualization ecosystems are predominantly JavaScript-based. Key libraries include D3.js, which provides granular control over graphical elements, animation, zooming, brushing, and responsive layouts, alongside higher-level libraries such as Plotly.js, Chart.js, and DataTables. React is increasingly adopted for component-based interface management, while D3 handles data transformation and rendering logic.
In EDA, linked brushing, zooming, and dynamic group splitting enable real-time detection of data anomalies, illogical observations, and outliers across concentration–time, covariate, and dose panels simultaneously, reducing iterative cycles of static plot regeneration. While R packages such as ggquickeda, esquisse, and Observable Plot offer accessible entry points, many pharmacometric workflows remain template-driven and static. Tools including R Shiny (Chang et al., 2015), mrgsolve (Baron, 2022), tidyvpc, gPKPDviz, and ModVizPop have lowered the barrier to live data exploration beyond pre-specified plotting recipes.
For model diagnostics, interactive GOF plots enable instantaneous stratification by study or multiple covariates without script re-execution, supporting rapid hypothesis generation during covariate model development. Dynamic covariate correlation networks, missingness heatmaps, and linked scatter plots with brushing accelerate exploratory covariate assessment by allowing analysts to identify complete covariate profiles of outlier subjects on demand. The tidyvpc package, with dplyr-compatible syntax and Plotly/ggplot2 backends, supports real-time VPC restratification by subpopulation (e.g., pediatric vs. adult, renal impairment groups), bin adjustment, and toggling between prediction-corrected and standard VPCs without re-running simulations when precomputed summaries are available. Interactive forest plots via coveffectsplot and PMXForest further enhance interpretation through parameter filtering and adjustable reference values.
From a regulatory perspective, HTML-based outputs have been accepted since 2014, and the R Consortium Pilot 4 demonstrated validation and submission of a full R Shiny application, establishing the feasibility of validated interactive electronic submissions. These precedents suggest that interactive reports could allow reviewers to explore diagnostics and simulation scenarios dynamically, complementing traditional static tables and figures. Examples from real submissions will be presented, and code will be shared publicly.
Conclusions
Interactive visualization tools offer a complementary paradigm to static PK/PD reporting by enabling real-time data exploration, dynamic diagnostics, and improved communication with interdisciplinary stakeholders. Their integration into pharmacometric workflows is increasingly feasible given mature open-source frameworks and AI-assisted development. Broader adoption will depend on standardized implementation practices, transparent validation strategies, and clearer regulatory guidance regarding interactive submission content. Dynamic visualization is poised to become an integral component of modern pharmacometric analysis rather than an auxiliary presentation layer.
References:
Bostock M et al. D3: Data-Driven Documents. IEEE Trans Vis Comput Graph. 2011.
Chang W et al. Shiny: Web Application Framework for R. R package. 2015.
Baron KT. mrgsolve: Simulation from ODE-Based PK/PD Models. J Pharmacokinet Pharmacodyn. 2022.
R Consortium. Pilot 4 Report (2024).
Reference: PAGE 34 (2026) Abstr 11873 [www.page-meeting.org/?abstract=11873]
Poster: Methodology - Other topics