Adaptive interventions, particularly just-in-time adaptive interventions (JITAIs), are reshaping personalized treatments in real time. At the Future of Adaptive Intervention Research (FAIR) Summit, early-career researchers gathered to workshop research projects in development and receive feedback from d3center faculty and invited experts in adaptive interventions. The event fostered collaborative discussions, highlighting challenges and opportunities for advancing the science of adaptive interventions.
Reducing Participant Burden
A prominent theme was reducing participant burden in intensive longitudinal studies, where frequent data collection can lead to dropout. Researchers brought forward proposals for optimizing the timing and frequency of assessments. One presenter shared the concept of a dynamic algorithm that could adapt the frequency of pings based on participant behavior, such as reducing requests in the afternoon if participants tend to ignore them at that time. Expert feedback centered on refining these algorithms to better capture high-quality data while minimizing participant fatigue.
A related workshop explored the idea of tailoring data collection methods to individual behaviors. By adjusting assessments to match participants’ habits, researchers hoped to reduce the risk of burnout and ensure more reliable data. This theme of customization was key as the group sought expert guidance on refining their methods.
Tailoring Variables for Actionable Decisions
A rich discussion emerged around the selection of tailoring variables—factors that guide the timing and nature of interventions. Several early-career researchers proposed variables such as mood or stress, but expert feedback emphasized that not all predictive variables lead to actionable decisions. Short-term fluctuations, like daily changes in stress, were viewed as potentially more useful for real-time interventions than longer-term changes. The group deliberated on how to refine their choice of tailoring variables to prioritize those that inform meaningful intervention decisions in the moment.
Integrating Data for Personalization
Workshop discussions also highlighted the importance of integrating both between-subject (across participants) and within-subject (over time) data for personalized interventions. Early-career researchers experimented with ways to pool data across participants to strengthen predictive models, while accounting for individual differences. One presenter explored using hierarchical models to integrate both population-level trends and individual-specific patterns. Experts encouraged further development of this approach, emphasizing its potential to deliver more accurate, real-time interventions.
Reinforcement Learning for Optimization
Reinforcement learning (RL) was a topic of much experimentation at the summit. Participants discussed RL’s potential to optimize intervention strategies based on real-time data. One presenter shared an early model that could guide the timing of interventions in JITAIs, adjusting based on participant stress levels or behavior patterns throughout the day. Experts provided feedback on how RL could be further refined to balance exploration (testing new strategies) and exploitation (using known strategies) in adaptive interventions.
The flexibility of RL, particularly its ability to learn and adapt continuously, was an area of great interest. Researchers saw potential for RL to help interventions evolve in response to new data, and they received valuable input on how to maximize this adaptability for real-world applications.
Balancing Precision and Practicality
Another key challenge was the difficulty of balancing the development of precise predictive models with real-world feasibility. For example, while reinforcement learning holds promise for enhancing prediction accuracy, participants questioned how to keep these models practical for real-world use. Several researchers workshopped approaches to reduce participant burden in ecological momentary assessment (EMA) studies. One idea involved using dynamic algorithms to adjust the number of questions asked at different moments, aiming to keep participant engagement high without sacrificing data quality. Feedback from experts focused on refining these approaches to strike a balance between data accuracy and practicality.
Future Directions: Collaboration and Innovation
The Summit concluded with a forward-looking discussion on the future of adaptive interventions, with participants sharing ideas about integrating new data streams and refining personalization. Experts encouraged cross-disciplinary collaboration as a key to unlocking new possibilities for personalized care.
As the field evolves, the collaboration between early-career researchers and experts will be crucial in developing scalable interventions that reduce participant burden and improve health outcomes. Through open dialogue and shared experimentation, this community is driving the future of adaptive interventions.
LET’S STAY IN TOUCH
Join the d3center Mailing List
Keep up to date with the latest news, events, software releases, learning modules, and resources from the d3center.