Models4PT: Inspired by DAGitty

Models4PT will be (is) a dynamic tool for creating and analyzing graphical causal models, inspired by the foundational work of DAGitty. While preserving and building upon DAGitty’s robust algorithms and user interface, Models4PT is being developed to meet the unique needs of physical therapy (PT) clinicians and researchers.

This repository expands DAGitty's core functionality to integrate into the broader Clinical Inquiry Network ecosystem, supporting advanced clinical reasoning, causal modeling, and evidence-informed practice in physical therapy based on a critical realist approach to clinical inquiry.

---

About Models4PT

Models4PT will be (is) a dynamic tool for creating and analyzing graphical causal models tailored to the needs of physical therapy clinicians and researchers. It integrates into the Clinical Inquiry Network, which seeks to enhance clinical reasoning, causal modeling, and evidence-informed practice in physical therapy. This initiative includes:

1. Stats4PT:
   A statistical education and guidance platform for physical therapy.
   Learn more here.

2. The Clinical Inquiry Fellowship:
   A program for advancing clinical reasoning and critical inquiry in PT practice.
   Learn more here.

3. Collaborative Tools:
   A suite of resources to model complex patient cases, integrate research findings, and explore clinical reasoning frameworks.

---

Enhancements in Progress

Transition to a PT-Focused Tool

• Adapting the platform to address physical therapy-specific use cases.
• Adding customizable templates and libraries for common PT causal models.

Integration of Bhaskar’s Critical Realist Framework

• Explicit modeling of the empirical, actual, and real domains in causal modeling.
• Support for latent variables and deeper structural analysis (including hierarchical models - see below).

Expanded Inference Methodologies

To support comprehensive clinical reasoning, Models4PT will integrate options for probabilistic causal inference models, including:

1. Induction (Statistical Inference):
   • Analyzing patterns and generalizations from data to support clinical decisions.
2. Deduction (Probability Logic):
   • Exploring deterministic and probabilistic relationships between variables.
3. Abduction (Bayesian Inference):
   • Generating hypotheses and evaluating their likelihood using Bayesian methods.

Improved User Experience

• Enhanced saving/loading functionality using local storage and, eventually, backend storage for collaborative workflows.
• Streamlined interfaces for creating, modifying, and exporting models.

Planned Extensions

1. Hosting the platform at models4pt.com to provide a publicly accessible, PT-centric tool.
2. Adding features to:
   • Identify and visualize confounders relevant to PT interventions and research.
   • Automate suggestions for evidence-based adjustments to causal models.
3. Integration into broader clinical reasoning education and consulting services.

Hierarchical Model Support

Models4PT is preparing to support hierarchical causal models. This functionality will allow nodes to reference submodels, enabling multilevel causal reasoning and integration of critical realist perspectives.

Key Objectives:

• Extend variable metadata to include submodel references.
• Prepare the backend database for hierarchical relationships.
• Ensure algorithms like adjustment set calculation and d-separation are hierarchy-ready.
• Add placeholders in the interface for hierarchical metadata.

Future Goals: This feature will support advanced use cases, including nested causal systems and multilevel analyses, while preserving compatibility with current core functionality.

---

Running the GUI Locally

To test or use the current version of Models4PT locally:

1. Clone the repository.
2. Open gui/dags.html in a modern web browser.
3. Follow the instructions in the GUI for saving/loading models.

---

Acknowledgments

We extend our gratitude to DAGitty.net and its creator, Johannes Textor, for their pioneering work in causal modeling. Models4PT builds upon DAGitty’s robust foundation to create a tailored tool for physical therapy professionals and researchers.

For more information on DAGitty, visit:

• Website: dagitty.net
• Publications:
  1. Textor, J., et al. (2017). Robust causal inference using directed acyclic graphs. International Journal of Epidemiology.
  2. Ankan, A., et al. (2021). Testing Graphical Causal Models Using the R Package “dagitty”. Current Protocols.

---

Contact

For questions about this project or collaboration opportunities, please contact:

• Sean Collins, PT, ScD
  Peripatetic PT
  Stats4PT
  Clinical Inquiry Fellowship

  Updated as of January 21, 2025