September 20, 2024 Brainstorm Overview

# Digital Twins of Study Subpopulations (DTS^2^s)
Optimizing Design Decisions During Multiphase Development of Adaptive Digital Interventions

## Goals for the Brainstorm
My primary goal for us is to revisit our work on simulation test beds and evaluate how well the DTS^2^ framing fits and is a compelling instantiation of the [digital twin](https://www.ncbi.nlm.nih.gov/books/NBK605499/) concept. My secondary goals focus specifically on the paper we are putting together on the topic, targeting a November submission. I want your help understanding what you would want to learn from a framework paper like this and what you would need to see to be convinced that this is a worthwhile contribution.

## Questions to Think About
- Do we need this DTS^2^ framework? Or is it already obvious how digital twins could be applied to the pre-trial optimization of JITAIs?
- Does this DTS^2^ framework help? Do you think others would find DTS^2^s a compelling way to optimize adaptive digital interventions for trial settings?
- What can we learn from digital twins to improve our own modeling, simulation, and decision-making? (e.g., transfer learning from one test bed to the next)
- What content would you find more or less important to a paper on DTS^2^s?
- How would you improve the title, outline, or overview figure of the paper?

## Overview
### Abstract
Just-in-time adaptive interventions (JITAIs) are nascent precision medicine systems that require numerous non-trivial design decisions (e.g., hyperparameters for a reinforcement learning algorithm). These design decisions must be made prior to clinical trial evaluation and cannot be changed during a trial to maintain intervention fidelity. Clinical trials are costly to conduct, and negative results can spell the end of an intervention’s development. How should a research team strategically make such high-stakes design decisions for a JITAI prior to its clinical trial evaluation? This paper introduces the framework digital twin of a trial population (DTS^2^) to address this question. DTS^2^s are “fit-for-purpose” (term emphasized in recent [Consensus Report on digital twins from the National Academies](https://nap.nationalacademies.org/catalog/26894/foundational-research-gaps-and-future-directions-for-digital-twins)) instantiations of the digital twin concept. Trial settings impose unique constraints on the bidirectional feedback between physical and virtual twins. Data collected during a trial can only be used to inform design decisions after the trial. Design decisions for an upcoming trial can only be informed using data from previous trials. DTS^2^s are thus fit for the purpose of simulating an upcoming trial population’s behavior, using models informed by previous trials’ data. In this paper, we detail the design process of DTS^2^s for JITAIs. We share lessons learned over the course of three multiphase JITAI deployments and outline guiding principles to follow. We also highlight the recurring challenge of mismatching variables and study characteristics from the data in one trial to the subsequent trial. We term these mismatches *data impoverishment* and provide solutions to data impoverishment when designing and updating DTS^2^s.

### Overview Figure

### Proposed Contributions for the Paper
1. Formalizes DTS^2^ framework to instantiate digital twins for JITAI development
2. Elucidates the design process of DTS^2^s, with accompanying best practices
3. Synthesizes solutions to data impoverishment, with examples from prior work