Abstract
In mobile health interventions aimed at behavior change and maintenance, treatments are provided in real time to manage current or impending high-risk situations or promote healthy behaviors in near real time. Currently there is great scientific interest in developing data analysis approaches to guide the development of mobile interventions. In particular data from mobile health studies might be used to examine effect moderators—individual characteristics, time-varying context, or past treatment response that moderate the effect of current treatment on a subsequent response. This article introduces a formal definition for moderated effects in terms of potential outcomes, a definition that is particularly suited to mobile interventions, where treatment occasions are numerous, individuals are not always available for treatment, and potential moderators might be influenced by past treatment. Methods for estimating moderated effects are developed and compared. The proposed approach is illustrated using BASICS-Mobile, a smartphone-based intervention designed to curb heavy drinking and smoking among college students.
Introduction
Mobile health (mHealth) interventions, such as smartphone apps that deliver "just-in-time" prompts for mindfulness or smoking cessation, offer a transformative way to support behavior change. A critical question for researchers is "what works for whom and when?" Answering this requires identifying moderators: factors like a user’s current stress level, location, or previous response to treatment that influence the effectiveness of a current intervention.
The Challenge of Time-Varying Feedback
Analyzing mHealth data presents unique statistical challenges. Unlike traditional longitudinal studies, mHealth interventions involve frequent, sequential treatments where past actions can influence future states.
As shown in the figure above, the state of a user at time $t$ ($X_t$) and the subsequent outcome ($Y_{t+1}$) are influenced by previous treatments ($A_{t-1}$). This creates complex feedback loops where past treatments influence future moderators. Furthermore, users are not always "available" for treatment; for instance, they may be driving or may not have completed a necessary self-report. Traditional methods like Generalized Estimating Equations (GEE) often fail in these settings, producing biased estimates because they do not properly account for these dynamic causal relationships.
Methodology: Centered and Weighted Least Squares (CWLS)
To address these issues, the authors propose a new formal definition of moderated effects using a potential outcomes framework. This framework distinguishes between proximal effects (immediate impact) and lagged effects (impact over several time steps), while explicitly accounting for time-varying availability.
The core of the proposed estimation method is Centered and Weighted Least Squares (CWLS):
- Centering: The treatment indicator is centered by subtracting the probability of receiving treatment. This "orthogonality" ensures that the treatment effect estimate is not biased by errors in the "nuisance" part of the model (the part describing general outcome trends).
- Weighting: Weights are applied to ensure estimates represent the average effect across the population of interest, even when individual availability varies.
This approach is robust; treatment effect estimates remain consistent even if the researcher incorrectly models the relationship between the outcome and other background variables.
Evaluation and Results
The CWLS method was evaluated through both simulation studies and a real-world application.
Simulation Studies
The authors created synthetic datasets with known causal effects to compare CWLS against standard GEE. In scenarios where moderators were omitted or correlation structures were misspecified, CWLS maintained a 95% confidence interval coverage. In contrast, standard GEE methods saw coverage drop as low as 57% to 78%, demonstrating significant bias in the presence of time-varying feedback.
BASICS-Mobile Application
The method was applied to the BASICS-Mobile study, a 28-day pilot involving 28 college students. The intervention used a smartphone app to deliver "urge surfing" mindfulness messages to reduce heavy drinking and smoking.
Key findings included:
- Moderation by Self-Regulation: Mindfulness messages successfully reduced smoking rates, but only when the user’s need to self-regulate thoughts was stable or decreasing (95% CI: -5.45 to -0.15 cigarettes per day). When the need to self-regulate was increasing, the treatment showed no significant effect.
- Proximal Utility: The study found no evidence of "lag-2" effects, suggesting that these specific mindfulness prompts provide immediate, short-lived utility rather than long-term delayed impact.
Conclusion and Future Directions
This research provides a rigorous foundation for analyzing moderated effects in mHealth. By formalizing the role of availability and addressing time-varying feedback, the CWLS method offers a robust tool for intervention optimization. Future work will look to extend this framework to categorical or count-based outcomes and incorporate high-dimensional data from wearable sensors to automatically select the most important moderators from a large pool of candidates.
Summary updated