Wrap-Up and Next Steps

Where to go from here with R for HTA

R for HTA (Basics) — Workshop 2026

RRC-HTA, AIIMS Bhopal | HTAIn, DHR

What You Accomplished in 3 Days

Day 1: Foundations + Decision Trees - R fundamentals (syntax, data types, visualization) - GDM diagnostic decision tree - TES therapeutic decision tree

Day 2: Markov Models + PSA + GenAI - CKD Markov cohort model (4 states) - Probabilistic sensitivity analysis - How to use AI tools to help write code

Day 3: PSM + Shiny Apps - Breast cancer partitioned survival model - PSA for PSM (parameter distributions, CEAC, CE plane) - Interactive Shiny app templates

Recap the journey. In three days, participants went from R basics to building complete HTA models with interactive apps.

Four Model Types You’ve Mastered

library(DiagrammeR)

grViz("
digraph models {
  graph [rankdir=TB, bgcolor='transparent', fontname='Helvetica']
  node [fontname='Helvetica', fontsize=10, style='filled,rounded', shape=box, width=2.5]

  dt_diag [label='Diagnostic\nDecision Tree\n(GDM)', fillcolor='#59a14f', fontcolor='white']
  dt_ther [label='Therapeutic\nDecision Tree\n(Stents)', fillcolor='#f28e2b', fontcolor='white']
  markov [label='Markov\nCohort Model\n(CKD)', fillcolor='#e15759', fontcolor='white']
  psm [label='Partitioned\nSurvival Model\n(Breast Ca)', fillcolor='#76b7b2', fontcolor='white']

  workshop [label='This Workshop', shape=ellipse, fillcolor='#4e79a7', fontcolor='white', width=2]

  workshop -> dt_diag
  workshop -> dt_ther
  workshop -> markov
  workshop -> psm
}
")

Figure 1: Four model structures for different HTA questions

You can now: - Identify which model fits a given decision problem - Build and validate each type - Run sensitivity analyses and PSA - Communicate results to stakeholders

Summarize the scope. Four distinct model types, each useful for different clinical questions.

Key R Packages for HTA

Package	Purpose
heemod	Markov models (easy syntax, automatic discounting)
dampack	Decision trees, sensitivity analyses (tornado plots, frontiers)
BCEA	Bayesian cost-effectiveness (CEAC, EVPI, credible regions)
hesim	Microsimulation & DES (complex state-transition models)
flexsurv	Parametric survival fitting (Weibull, log-logistic, etc.)
survHE	Survival analysis for HE (fit, compare, integrate)

You’ve already used the core concepts in this workshop. These packages automate and streamline them.

Introduce the major packages in the R-HTA ecosystem. Participants have done the manual version; now they know what these packages do.

The DARTH Group: Your Community

DARTH = Decision Analysis in R for Technologies in Health

They publish: - Tutorial papers (Lancet, MDM, other journals) - GitHub repositories with fully worked examples - Coding standards for reproducible HTA

Find them: GitHub, DARTH website, MDM journal

R-HTA Consortium: Connect with other analysts, webinars, code sharing

Point to existing communities and resources. Participants aren’t alone; there’s a growing community.

Your Learning Path Forward

grViz("
digraph path {
  graph [rankdir=TB, bgcolor='transparent', fontname='Helvetica']
  node [fontname='Helvetica', fontsize=10, style='filled,rounded', shape=box, width=3]

  now [label='This Workshop\n(Read, modify, run R code)', fillcolor='#4e79a7', fontcolor='white']
  next1 [label='Next 1-2 weeks\n(Adapt templates for your problem)', fillcolor='#59a14f', fontcolor='white']
  next2 [label='Weeks 3-4\n(Add features: PSA, new outcomes)', fillcolor='#f28e2b', fontcolor='white']
  next3 [label='Months 2-3\n(Share Shiny app, deploy)', fillcolor='#e15759', fontcolor='white']
  community [label='Ongoing\n(DARTH, R-HTA Consortium,\npublish methods)', fillcolor='#76b7b2', fontcolor='white']

  now -> next1 -> next2 -> next3 -> community
}
")

Figure 2: Progression from this workshop to mastery

Give concrete milestones. Learning is a journey; outline realistic next steps.

Phase 1: Consolidate (Next 1-2 Weeks)

1. Take one workshop template
2. Modify it for a problem you care about
3. Change parameter values, time horizon, strategies
4. Run the model several times to understand assumptions
5. Get feedback from colleagues

Goal: Build confidence with the templates. Understand how each assumption drives results.

Actionable first steps. Keep it concrete and achievable.

Phase 2: Expand (Weeks 3-4)

1. Identify a feature you want to add:

   • PSA (if starting from a deterministic model)
   • New outcome measure
   • Sensitivity analysis tornado plot

2. Use AI tools (ChatGPT, Claude) to help:

   • Show them the workshop code
   • Ask for modifications
   • They generate scaffolding; you refine

3. Read the relevant package vignette

4. Test and validate

Practical guidance on using AI. Participants shouldn’t be intimidated by code they don’t yet understand.

Phase 3: Share (Months 2-3)

1. Deploy a Shiny app to shinyapps.io (free tier)
2. Share your Quarto report with stakeholders
3. Invite colleagues to explore and give feedback
4. Iterate on the model

This is where impact happens. Your models inform real decisions.

Sharing is crucial. The model in a drawer helps no one.

Phase 4: Contribute (Ongoing)

• Join R-HTA Consortium or local R user group
• Attend webinars and workshops
• Help others by sharing your code
• Consider publishing methods in open-access journal

The virtuous cycle. Learn, build, share, teach.

Four Model Types: A Decision Guide

Your decision problem:…

→ Is it about diagnostic accuracy? → Use Diagnostic Decision Tree (sensitivity, specificity, prevalence)

→ Is it about a one-time treatment choice? → Use Therapeutic Decision Tree (event probabilities, immediate costs/benefits)

→ Does the disease progress over time? → Use Markov Cohort Model (states, transition probabilities, cycles)

→ Do you have survival curves from a trial? → Use Partitioned Survival Model (OS and PFS, extrapolation)

Practical decision tree for choosing model types. Participants should reference this.

Key Principles to Remember

Transparency first. Every assumption visible and adjustable. Avoid hiding logic.

Start simple. Simplest model that answers the question. Add complexity later.

Validate early. Check that your model reproduces expected results before publishing.

Document as you go. Clear comments explaining each modeling choice.

Test your code. A bug in a transition probability propagates through the analysis.

Share your work. HTA community improves when we share methods and code.

Core principles that apply regardless of model type or package.

Publishing with R: Quarto for Reproducible Reports

Quarto weaves R code and narrative into a single reproducible document.

A typical HTA report: - Executive summary - Model structure explanation - Embedded code chunks (transparent) - Plots and tables generated directly from R - Sensitivity analyses - Conclusion

Output: HTML, PDF, Word, or website — write once, export everywhere.

Benefit: Reviewers can see every line of code that produced every number.

Quarto is powerful for reproducibility. Many journals now encourage or require it.

You Don’t Need to Be a Programmer

You need to be an HTA analyst who has learned to use R as a tool.

Just as you might use Excel, but better.

Your models will be: - Reproducible (everyone can run the code) - Transparent (all assumptions visible) - Shareable (interactive apps, GitHub)

Important

You’ve invested three days in learning these skills. That investment pays dividends.

Final encouragement. Participants shouldn’t feel like they need to be programmers. They’re analysts who happen to use R.

The R Community is Welcoming

When you have questions: Ask them.

When you get stuck: Others have been stuck there too.

When you solve a problem: Share the solution.

Resources: - Stack Overflow (tag: r, health-economics) - R-HTA Consortium forums - GitHub Discussions - DARTH tutorials - Local R user groups

Community support is real. Reduce isolation. Participants are joining a helpful ecosystem.

Resources to Dive Deeper

Books: - R for Data Science by Hadley Wickham - Mastering Shiny by Hadley Wickham - Decision Modelling for HE by Briggs, Claxton, Sculpher

Online: - DARTH tutorials (GitHub) - R-HTA Consortium webinars - Shiny official documentation

Packages to explore: - heemod vignette - dampack vignette - BCEA vignette

Provide a reading list for self-directed learning.

ParCC: Your Tool from RRC-HTA

[Information about ParCC — the platform or tool your organization offers]

Check the workshop GitHub repository for: - All slide decks (reveal.js) - All R code and Quarto documents - All Shiny app templates - Sample data and results

Reference any institutional tools or resources. Direct participants back to the workshop repository.

Your Next Steps (Concrete)

This week: 1. Run one of the workshop apps locally 2. Modify a parameter and observe the change 3. Understand how the code works (read comments, not every line)

Next week: 1. Adapt a template for a problem at your organization 2. Change parameter values to match your data 3. Share results with a colleague

Next month: 1. Add a feature (plot, sensitivity analysis, PSA) 2. Deploy a Shiny app 3. Get feedback

Concrete, achievable milestones. Reduce overwhelm. One step at a time.

Thank You!

You’ve worked hard over three days. You now have the skills to: - Build health technology assessment models in R - Validate and test them rigorously - Communicate results interactively to stakeholders - Contribute to reproducible, transparent decision-making in health

Go build something great.

Questions? Contact: - RRC-HTA, AIIMS Bhopal [email/contact] - HTAIn, Department of Health Research [email/contact]

Final thanks and inspiration. End on a high note.

Workshop Materials

All materials (slides, code, templates, data) available at:

GitHub: [repository URL]

Local path: /day3/

How to use the repository: 1. Fork it (create your own copy) 2. Clone to your computer 3. Explore and modify the templates 4. Share improvements via pull request

Happy coding!

Direct participants to the resource repository. Make it easy to access and use materials after the workshop.