About

Walter Dempsey is an Associate Professor of Biostatistics and an Assistant Research Professor at the Institute for Social Research. His research focuses on developing statistical methods for digital and mobile health, as well as traditional biomedical research involving survival analysis and temporal process regression. His work involves creating models for complex longitudinal and survival data, as well as relational structures such as interaction networks, with the aim of informing decision making in health through intervention evaluation and development. Prior to his current roles, he was a postdoctoral fellow in the Department of Statistics at Harvard University, working within the Statistical Reinforcement Learning Lab under Susan Murphy. He earned his PhD in Statistics from the University of Chicago, where he also completed his B.Sc. in Mathematics, Statistics, and Economics. His research interests include statistical methods and theory for digital health, complex survival data, and social network modeling, with a focus on designing models that reflect empirical behavior and satisfy invariance principles. His work aims to bridge theory and applied science to improve understanding, prediction, and treatment of health phenomena.

Research topics

• Computer Science
• Medicine
• Medical emergency
• Applied psychology
• Political Science
• Psychiatry
• Psychology
• Risk analysis (engineering)
• Engineering
• Engineering ethics
• Emergency medicine

Selected publications

• Evaluation of the HeartSteps Online Sampling Algorithm

  arXiv (Cornell University) · 2025-01-03

  preprintOpen access

  Micro-randomized trials (MRTs), which sequentially randomize participants at multiple decision times, have gained prominence in digital intervention development. These sequential randomizations are often subject to certain constraints. In the MRT called HeartSteps V2V3, where an intervention is designed to interrupt sedentary behavior, two core design constraints need to be managed: an average of 1.5 interventions across days and the uniform delivery of interventions across decision times. Meeting both constraints, especially when the times allowed for randomization are not determined beforehand, is challenging. An online algorithm was implemented to meet these constraints in the HeartSteps V2V3 MRT. We present a case study using data from the HeartSteps V2V3 MRT, where we select appropriate metrics, discuss issues in making an accurate evaluation, and assess the algorithm's performance. Our evaluation shows that the algorithm performed well in meeting the two constraints. Furthermore, we identify areas for improvement and provide recommendations for designers of MRTs that need to satisfy these core design constraints.

  PublisherOA PDFDOI

• A Continuous-Time Dynamic Factor Model for Intensive Longitudinal Data Arising from Mobile Health Studies

  Psychometrika · 2025-06-15 · 1 citations

  articleOpen access

  Intensive longitudinal data (ILD) collected in mobile health (mHealth) studies contain rich information on the dynamics of multiple outcomes measured frequently over time. Motivated by an mHealth study in which participants self-report the intensity of many emotions multiple times per day, we describe a dynamic factor model that summarizes ILD as a low-dimensional, interpretable latent process. This model consists of (i) a measurement submodel-a factor model-that summarizes the multivariate longitudinal outcome as lower-dimensional latent variables and (ii) a structural submodel-an Ornstein-Uhlenbeck (OU) stochastic process-that captures the dynamics of the multivariate latent process in continuous time. We derive a closed-form likelihood for the marginal distribution of the outcome and the computationally-simpler sparse precision matrix for the OU process. We propose a block coordinate descent algorithm for estimation and use simulation studies to show that it has good statistical properties with ILD. Then, we use our method to analyze data from the mHealth study. We summarize the dynamics of 18 emotions using models with one, two, and three time-varying latent factors, which correspond to different behavioral science theories of emotions. We demonstrate how results can be interpreted to help improve behavioral science theories of momentary emotions, latent psychological states, and their dynamics.

  PublisherOA PDFDOI

• A just-in-time adaptive mobile application intervention to reduce sodium intake and blood pressure in patients with hypertension: Rationale and design of the LowSalt4Life 2 trial

  American Heart Journal · 2025-10-27 · 1 citations

  articleOpen access

  BACKGROUND: Excess dietary sodium intake is a major contributor to hypertension (HTN) and cardiovascular disease in the U.S., yet most Americans exceed recommended sodium intake levels. Despite the established benefits of sodium reduction, behavioral adherence remains a challenge. Just-in-time adaptive interventions (JITAIs) provide a novel mobile health (mHealth) strategy to support low-sodium choices in real-time at restaurants and grocery stores. METHODS: LowSalt4Life 2 is a 6-month randomized controlled trial evaluating a mobile application-based sodium-focused JITAI among adults with HTN. In phase one, participants were randomized 1:1 to either the LowSalt4Life app with JITAI (App+JITAI) or the app alone. At 2 months, the App+JITAI group was re-randomized to either continue with the standard JITAI or receive a personalized JITAI (pJITAI) informed by reinforcement learning based on prior engagement. The primary outcome was change in systolic blood pressure (SBP) at 2 months. Secondary outcomes included changes in BP medication, dietary sodium intake, and engagement metrics. RESULTS: As of October 2025, 410 participants had been enrolled and completed follow-up. The trial's final results are anticipated in Spring 2025. Primary analysis focuses on the change in SBP between the App+JITAI and App-only groups, as well as the added value of personalization in the second study phase. CONCLUSIONS: LowSalt4Life 2 tests a scalable, context-aware digital intervention designed to reduce dietary sodium and improve BP management. By personalizing engagement based on user behavior, this study seeks to advance mHealth strategies for HTN self-management and inform future interventions targeting dietary behaviors in real-world settings.

  PublisherDOI

• Improving prediction in M-estimation by integrating external information from heterogeneous populations

  ArXiv.org · 2025-09-04

  preprintOpen access1st authorCorresponding

  A novel approach to improve prediction and inference in M-estimation by integrating external information from heterogeneous populations is proposed. Our method leverages joint asymptotics to combine estimates from external and internal datasets, where the external dataset provides auxiliary information about a subset of parameters of interest. We introduce a shrinkage estimator that combines internal and external estimates under a general class of transformations that ensure consistency across populations.

  PublisherOA PDFDOI

• Improving Affective Associations With Physical Activity via a Message-Based mHealth Intervention (WalkToJoy): Proof-of-Concept Study

  Journal of Medical Internet Research · 2025-06-20 · 3 citations

  articleOpen access

  BACKGROUND: Traditional mobile health interventions for physical activity (PA) primarily rely on reflective self-regulatory processes, often neglecting the role of affective associations in sustaining long-term engagement. The WalkToJoy intervention addresses this gap by applying the affective-reflective theory to enhance intrinsic motivation for PA among adults aged ≥40 years through affective message framing, evaluative conditioning, and belief updating. OBJECTIVE: This proof-of-concept study evaluated the feasibility of the message-based WalkToJoy intervention package and examined the impact of its 3 components-walking suggestion prompts, salience messages, and planning prompts-on affective and behavioral outcomes related to walking. METHODS: We conducted a fully remote, 6-week full factorial experiment with an embedded microrandomized trial (MRT) involving 49 adults aged ≥40 years. Statistical analyses, including paired t tests and generalized estimating equations, assessed pretest-posttest changes and the effects of smile-inducing walking suggestion prompts with short animated images (GIF images), salience messages, and planning prompts on weekly affective measures and daily step counts. In addition, MRT analyses evaluated the proximal effects of these components. Poststudy interviews were thematically analyzed to contextualize participants' experiences and engagement with the intervention. RESULTS: Significant pretest-posttest improvements were observed across affective outcomes on a 7-point Likert scale-affective attitudes improved by 0.547 points (P<.001), affective valuations improved by 0.718 points (P<.001), affective reflection improved by 0.692 points (P<.001), and anticipated affect improved by 0.692 points (P<.001). While the average daily steps showed a nonsignificant pretest-posttest increase of 80 steps (P=.79), further analysis revealed an increase of 506 steps (P=.07) when comparing baseline to the average of weeks 4 to 6. Among the intervention components, GIF prompts significantly increased anticipated affect by 0.345 points (P=.046) and average daily step count by 1834 steps (P=.05) compared to identical text-only prompts. However, MRT analysis found no significant increase in 4-hour step counts following the walking suggestion prompts (P=.55), which was explained by qualitative findings suggesting that participants interpreted messages as flexible day-long reminders rather than immediate calls to action. Salience and planning prompts did not yield substantial quantitative effects but were positively received by participants for promoting mindfulness and personalized engagement. CONCLUSIONS: The WalkToJoy intervention is a feasible and promising approach for improving affective associations with walking. Walking suggestion prompts were particularly effective in boosting engagement and mitigating message fatigue, highlighting the potential of affect-driven interventions to enhance PA motivation and adherence.

  PublisherDOI

• Improving Affective Associations With Physical Activity via a Message-Based mHealth Intervention (WalkToJoy): Proof-of-Concept Study (Preprint)

  2025-04-10

  preprintOpen access

  <sec> <title>BACKGROUND</title> Traditional mobile health interventions for physical activity (PA) primarily rely on reflective self-regulatory processes, often neglecting the role of affective associations in sustaining long-term engagement. The WalkToJoy intervention addresses this gap by applying the affective-reflective theory to enhance intrinsic motivation for PA among adults aged ≥40 years through affective message framing, evaluative conditioning, and belief updating. </sec> <sec> <title>OBJECTIVE</title> This proof-of-concept study evaluated the feasibility of the message-based WalkToJoy intervention package and examined the impact of its 3 components—walking suggestion prompts, salience messages, and planning prompts—on affective and behavioral outcomes related to walking. </sec> <sec> <title>METHODS</title> We conducted a fully remote, 6-week full factorial experiment with an embedded microrandomized trial (MRT) involving 49 adults aged ≥40 years. Statistical analyses, including paired &lt;i&gt;t&lt;/i&gt; tests and generalized estimating equations, assessed pretest-posttest changes and the effects of smile-inducing walking suggestion prompts with short animated images (GIF images), salience messages, and planning prompts on weekly affective measures and daily step counts. In addition, MRT analyses evaluated the proximal effects of these components. Poststudy interviews were thematically analyzed to contextualize participants’ experiences and engagement with the intervention. </sec> <sec> <title>RESULTS</title> Significant pretest-posttest improvements were observed across affective outcomes on a 7-point Likert scale—affective attitudes improved by 0.547 points (&lt;i&gt;P&lt;/i&gt;&amp;lt;.001), affective valuations improved by 0.718 points (&lt;i&gt;P&lt;/i&gt;&amp;lt;.001), affective reflection improved by 0.692 points (&lt;i&gt;P&lt;/i&gt;&amp;lt;.001), and anticipated affect improved by 0.692 points (&lt;i&gt;P&lt;/i&gt;&amp;lt;.001). While the average daily steps showed a nonsignificant pretest-posttest increase of 80 steps (&lt;i&gt;P&lt;/i&gt;=.79), further analysis revealed an increase of 506 steps (&lt;i&gt;P&lt;/i&gt;=.07) when comparing baseline to the average of weeks 4 to 6. Among the intervention components, GIF prompts significantly increased anticipated affect by 0.345 points (&lt;i&gt;P&lt;/i&gt;=.046) and average daily step count by 1834 steps (&lt;i&gt;P&lt;/i&gt;=.05) compared to identical text-only prompts. However, MRT analysis found no significant increase in 4-hour step counts following the walking suggestion prompts (&lt;i&gt;P&lt;/i&gt;=.55), which was explained by qualitative findings suggesting that participants interpreted messages as flexible day-long reminders rather than immediate calls to action. Salience and planning prompts did not yield substantial quantitative effects but were positively received by participants for promoting mindfulness and personalized engagement. </sec> <sec> <title>CONCLUSIONS</title> The WalkToJoy intervention is a feasible and promising approach for improving affective associations with walking. Walking suggestion prompts were particularly effective in boosting engagement and mitigating message fatigue, highlighting the potential of affect-driven interventions to enhance PA motivation and adherence. </sec> <sec> <title>CLINICALTRIAL</title> <p/> </sec>

  PublisherDOI

• Characteristics of Tailored Text Messages that Maximize Physical Activity amongst Cardiac Rehabilitation Enrollees: Micro-randomized Trial Moderator Analysis (Preprint)

  2025-06-29

  preprint

  <sec> <title>BACKGROUND</title> Emerging data suggest text message-based mHealth interventions may enhance physical activity levels in patients with cardiovascular disease enrolled in cardiac rehabilitation. Optimal characteristics of text messages that lead to maximal patient engagement and drive meaningful behavioral change are not well understood. </sec> <sec> <title>OBJECTIVE</title> To understand the impact of text message-level and participant-level characteristics that are hypothesized to lead to increased physical activity after delivery. </sec> <sec> <title>METHODS</title> The VALENTINE (Virtual Application-supported ENvironment To INcrease Exercise) study was a randomized controlled trial to evaluate a mobile health intervention delivered to low and moderate-risk adults enrolled in cardiac rehabilitation. Embedded within this study was a micro-randomized trial focused on the effect of text messages on physical activity levels amongst intervention participants. Participants in the intervention group received text messages through a smartwatch (Apple Watch or Fitbit Versa) that were tailored to the time of day, day of the week (weekday versus weekend), weather, and time since enrollment in cardiac rehabilitation. Text messages also differed in content type (walking vs. anti-sedentary) and level of personalization (inclusion of participant name or not). Delivery was randomized at four user-selected time points daily, with participants having a 25% probability of receiving a notification at any time point. The primary outcome was step count 60 minutes after a decision point. The current analysis focuses on the text message and participant-level factors that moderate the intervention’s effect on the primary outcome. Given potential measurement differences determined a priori, analyses were stratified by device type and phase of cardiac rehabilitation and adjusted for age, sex, and baseline activity status using a generalization of regression analysis. </sec> <sec> <title>RESULTS</title> Over 70,552 randomizations occurred in 108 participants (mean age 59.5 [SD 10.7] years; 32.4% female; 17.6% non-White; 63.0% Apple Watch users) over six months. Overall, no text message characteristics, (including personalization with participant name) or participant characteristics (including baseline physical activity) consistently impacted text message responsiveness for either device type. Although not consistently significant between device types and across phases of the trial, there was a trend towards increased responsiveness to text messages that promoted walking (as compared to anti-sedentary messages) and when delivered to younger (age &lt; 65 years) and male participants. </sec> <sec> <title>CONCLUSIONS</title> In this randomized clinical trial, we found that tailored text messages improve physical activity levels amongst cardiac rehabilitation enrollees in the initiation phase, but this effect was not explained by text message or participant-level moderators. Additional work is needed to explore the impact of tailoring based on an extended set of personal and environmental factors to optimize delivery and efficacy of text message-based interventions. </sec> <sec> <title>CLINICALTRIAL</title> Unique identifier: NCT04587882 </sec> <sec> <title>INTERNATIONAL REGISTERED REPORT</title> RR2-10.1016/j.ahj.2022.02.012 </sec>

  PublisherDOI

• A circadian and app-based personalized lighting intervention for the reduction of cancer-related fatigue

  Cell Reports Medicine · 2025-03-01 · 5 citations

  articleOpen access

  Lighting interventions can mitigate fatigue by promoting circadian rhythmicity. We test whether individualized, wearable-based lighting interventions delivered via a mobile app reduce cancer-related fatigue in a randomized controlled trial with 138 breast cancer, prostate cancer, and hematopoietic stem cell transplant patients. Participants are randomized to tailored lighting intervention or control. The primary endpoint is PROMIS fatigue 4a at trial end, with secondary endpoints including change in daily fatigue, sleep, anxiety, depression, physical function, and overall health. Fatigue T-scores at week 11 do not differ between groups but decrease significantly from week 1 to week 11 (3.07 points, p = 0.001) in the intervention group, with a significant final-week treatment effect (p = 0.014). Daily fatigue, anxiety, sleep disturbance, and physical function improve within intervention. Further studies are needed to see if these results generalize in broader cancer care. The trial is registered at ClinicalTrials.gov (trial registration number: NCT04827446).

  PublisherOA PDFDOI

• Practical considerations when designing an online learning algorithm for an app-based mHealth intervention

  ArXiv.org · 2025-11-11

  preprintOpen accessSenior author

  The ubiquitous nature of mobile health (mHealth) technology has expanded opportunities for the integration of reinforcement learning into traditional clinical trial designs, allowing researchers to learn individualized treatment policies during the study. LowSalt4Life 2 (LS4L2) is a recent trial aimed at reducing sodium intake among hypertensive individuals through an app-based intervention. A reinforcement learning algorithm, which was deployed in one of the trial arms, was designed to send reminder notifications to promote app engagement in contexts where the notification would be effective, i.e., when a participant is likely to open the app in the next 30-minute and not when prior data suggested reduced effectiveness. Such an algorithm can improve app-based mHealth interventions by reducing participant burden and more effectively promoting behavior change. We encountered various challenges during the implementation of the learning algorithm, which we present as a template to solving challenges in future trials that deploy reinforcement learning algorithms. We provide template solutions based on LS4L2 for solving the key challenges of (i) defining a relevant reward, (ii) determining a meaningful timescale for optimization, (iii) specifying a robust statistical model that allows for automation, (iv) balancing model flexibility with computational cost, and (v) addressing missing values in gradually collected data.

  PublisherOA PDFDOI

• Time-Related Considerations for Modeling Event-Based Data Collected via Ecological Momentary Assessment

  2025-09-22

  articleOpen access

  Ecological momentary assessments (EMAs) and wearable devices afford opportunities to collect real-time data on events experienced in daily life. Examples of event-based data in the psychological and behavioral sciences include smoking a cigarette, experiencing a stressor, having a disruption to sleep, experiencing a depressive or manic episode, drinking an alcoholic beverage, or engaging in a bout of exercise. The increasing availability of dense sampling approaches allows for the measurement of such events at relatively fast timescales (e.g., occurring across minutes, hours, days, or weeks), expanding the possibilities for how time can be conceptualized and modeled. Survival analysis is a modeling approach that allows researchers to address scientific questions regarding whether and when events occur in time. Although not often applied to EMA data, there are myriad research questions relevant to psychosocial and behavioral scientists that can be addressed using survival analysis. In this article, we provide an overview of survival analysis, describe several time-based considerations for modeling event-based EMA data using survival analysis, and provide several illustrative examples of the different time-based considerations. Altogether, the goals of this article are to enhance knowledge of the types of research questions that can be examined using survival analysis, illustrate nuances of applying the method to EMA data, and spark ideas for future empirical and methodological research.

  PublisherDOI

Frequent coauthors

• Charles Robbins

  National Fire Protection Association

  196 shared

• Harold H. Smith

  Southern University

  98 shared

• H Vanderpoel

  National Fire Protection Association

  98 shared

• Samuel Scott

  Harvard University

  49 shared

• Rosario A. Gerhardt

  Georgia Institute of Technology

  49 shared

• Ν Carle

  American Association For The Advancement of Science

  49 shared

• I.W. Chubb

  49 shared

• Widener Bldg

  Eli and Edythe Broad Foundation

  49 shared