Announcements

March 2023 training institute to support use of novel experimental designs in education sciences—now open to applications

This four-day training at the d3center in Ann Arbor will provide scholars in the education sciences with a rigorous foundation in the design, funding, conduct, and analysis of novel experimental design methods for optimizing adaptive interventions, including SMART designs. It will also promote ongoing professional career development by offering scholars guidance and mentorship before, during, and after the in-person training institute to support research plans pertaining to the construction of adaptive interventions in educational settings. The training will take place March 14-17, 2023.

Welcome to our new team members!

Partner Updates

Spotlight on Hybrid Experimental Designs

Digital interventions have quickly become an essential tool for the treatment of a wide variety of disorders, including HIV and Substance Abuse Disorders (SUDs). d3center researchers recognize that increasing engagement with these interventions is critical to unlocking the transformative potential of mobile health. Alongside collaborators, they are working to develop novel experimental designs with the construction of engagement-focused digital interventions in mind. Earlier this month, an article introducing the Hybrid Experimental Design (HED) and a software package that simulates and analyzes data from a hypothetical HED were published by Drs. Inbal Nahum-Shani, John J. Dziak, Maureen A. Walton, and Walter Dempsey.

Publication

“Hybrid Experimental Designs for Intervention Development: What, Why, and How”
Advances in Methods and Practices in Psychological Science

Human delivery of interventions—by clinical staff, for example—can be more engaging than digital intervention alone but potentially more expensive and burdensome. Hence, the integration of digital and human-delivered components is critical to building effective and scalable psychological interventions. Existing experimental designs are not equipped to inform joint sequencing and adaptation of digital and human-delivered components. The Hybrid Experimental Design (HED) is a new experimental approach that can be used to answer scientific questions about building psychological interventions in which digital and human-delivered components are integrated and adapted at multiple timescales.

Software Release

Our entire software collection is downloadable and free to use.

Simulating and analyzing data from a Hybrid Experimental Design (HED)

Researchers can use this companion R code to simulate data and walk through the analysis of data from a Hybrid Experimental Design that combines elements of a SMART with elements of a Micro-Randomized Trial (MRT). The code can also be used to plan the sample size required for a hybrid experiment under given assumptions.

‘Tree-Based’ Publications

When we teach people about adaptive interventions, we draw them as a branching structure that grows from left to right. If you were to turn any one of these schematics 90° to the right, you would see a “tree.” In a pair of manuscripts, d3center collaborators Lu Wang, Yebin Tao, Nina Zhou, and Daniel Almirall develop a “tree-based” reinforcement learning” algorithm for estimating an easily-interpretable (tree-based) adaptive intervention from existing observational study data. Yebin et al. (2018) introduce the method. Zhou et al. (2022) extend the method to address an issue unique to observational study data: sometimes the treatment sequences observed in the data are no longer viable in clinical practice or not consistent with the type of adaptive interventions the analyst wants to develop.

“Tree-based reinforcement learning for estimating optimal dynamic treatment regimes”
The Annals of Applied Statistics

“Estimating tree-based dynamic treatment regimes using observational data with restricted treatment sequences”
Biometrics

Both manuscripts illustrate the method of using data from an observational study of adolescents with substance use disorder. Behavioral intervention scientists can use these methods in the preparation phase of their research to generate hypotheses about components of an adaptive intervention. The adaptive intervention can then be optimized in a subsequent optimization trial, such as a SMART.

Crash Course

Quick but thoughtful answers to the questions we hear the most.

Q: Must an adaptive intervention recommend a single intervention option at each decision point? For example, must an adaptive intervention recommend a single intervention option for non-responders?

A: No—an adaptive intervention may recommend that a practitioner choose between multiple intervention options at any given decision point. For example, the two-stage adaptive intervention below recommends the addition of a behavioral intervention or an increase in medication dosage as a second-stage treatment for children who did not respond to the first-stage intervention.

This type of adaptive intervention might result from clinical expertise, prior data, or a randomized trial that suggests there is no evidence of a difference between intensifying behavioral intervention and adding low-dose medication for those who did not respond to the first-stage behavioral intervention.

Learn more about the protocolized but flexible nature of adaptive interventions.

Upcoming Events

Stay in the loop.