QMP/SMP Methodological Seminar is not currently being offered. Please check back for upates.
This seminar is intended to facilitate collaboration among behavioral, social, health, and data scientists interested in the development and application of adaptive approaches to intervention (prevention, treatment or policy), measurement, and data collection. The term ‘adaptation’, which is broadly defined as ‘the ability to change to suit different conditions’, has different meaning across different domains of behavioral, social and health practice and research. The goal of this seminar is to bridge this gap by exploring and discussing how concepts, tools and procedures used to inform or operationalize adaptation in one domain (e.g., intervention science) can be used to inform or operationalize adaptation in other domains (e.g., survey and data scince), and how the various approaches can be used synergistically to advance precision medicine initiatives. For example, how can we improve health by combining ideas from the design of adaptive interventions, which use ongoing information about an individual or context to decide how to modify treatments over time, with ideas from responsive survey design, which uses ongoing information about an individual or context to decide how best to engage an individual in a research survey, and adaptive measurement, which focuses on efficient, low-burden approaches to measuring change in health constructs?
Seminar topics include the design of adaptive interventions (also known as dynamic treatment regimens), adaptive implementation interventions, just-in-time adaptive interventions in mobile health, adaptive measurement, and responsive survey design, as well as the use of novel experimental approaches and associated data analytic methods to inform the development of these adaptive approaches, such as factorial designs, sequential multiple assignment randomized trials (SMART), or micro randomized trials (MRT).
This monthly seminar will follow a “brainstorming session format” whereby (i) a topic leader presents for approximately 20 minutes, beginning with the scientific background needed to facilitate a discussion and ending with a description of the conceptual or methodological challenge(s) yet to be resolved; (ii) this is followed by approximately 1 hour of discussion/brainstorm. Two different graduate students from the Program in Survey and Data Science (PSDS) and QMP each will summarize the topic leader presentation and the discussion. This format is intended to lead to lively, informal, discussion of new ideas related to adaptive intervention, measurement and data collection in precision medicine.
Behavioral, social, health, and data scientists from across the University of Michigan campus are invited to participate. This includes faculty, postdoctoral fellows and graduate students from survey and data science, statistics, economics, political science, education, business, sociology, psychology, medicine and other cognate disciplines. Students may take this seminar as an elective course for one credit hour; this will require continuous attendance and participation throughout the semester for a passing grade. Any questions should be directed to Brady T. West (email@example.com).
Past Discussion Topics
January 25, 2019 - Kevin Tolliver, U.S. Census Bureau
Other Census Attendees: Stephanie Coffey, Carolyn Pickering
Topic: A SMART Approach to using Text Messages to Increase Interviewer Compliance of Case Prioritization
The Survey of Income Program Participation (SIPP) has used case prioritization with the goal of reducing non-response bias since the 2016 data collection. While there is evidence that case prioritization has led to higher data quality, there is no mandate to ensure survey interviewers follow priority instructions. Without full interviewer compliance, the true effects of the prioritization cannot be evaluated. Beginning in the 2019 data collection, SIPP will test text messaging interviewers to see if notifications sent directly to their phone (as opposed to an email) affects how and when they work their cases. We plan to implement a SMART design aimed at finding the optimal messaging strategy to best encourage interviewers to follow priority instructions. This presentation discusses the design to date and seeks to further improve the design.
February 22, 2019 - Yajuan Si, Research Assisstant Professor, Survey Research Center, Institute for Social Research
Topic: Patient-centered glycosylated hemoglobin monitoring and dynamic treatment regime
The American Association of Diabetes recommends individualized glycosylated hemoglobin testing schedules and treatment plans for complex patients with diabetes. To pursue evidence-based research, we develop an adaptive and effective surveillance tool to inform optimal testing and treatment regimens based on electronic health records.
March 22, 2019 - Matthew Schipper, Research Associate Professor of Radiation Oncology and Biostatistics, University of Michigan
Topic: Individualized Optimal Dose Selection Via Statistical Models
The increasing availability of validated biomarkers and statistical models for outcomes in oncology raises possibility of more individualized and adaptive Radiation Therapy (RT) treatment plans. Two general settings are possible. In one, there are competing efficacy (e.g. tumor control) and toxicity outcomes, both associated with treatment dose. The goal is to select an RT dose that maximizes the expected utility defined as a weighted combination of the probabilities of efficacy and toxicity. In other settings, treatment outcomes can be summarized with a single survival type outcome and the goal is to estimate an optimal dosing rule which will maximize expected survival times. The above methods could be used at baseline or at a mid-treatment timepoint using mid-treatment biomarkers or imaging. In standard clinical practice, RT treatment is planned at baseline and given all in one ‘stage’ but Dynamic Treatment Regimes could be utilized in RT treatments, to allow two or more stages of RT in which some dose is given followed by assessment of some intermediate outcome followed by additional RT.
April 19, 2019 - Brady West, Research Associate Professor, Survey Research Center, Institute for Social Research
Topic: Toward the Optimal Design of Modular Surveys
Respondent burden is a critical issue facing survey researchers, and the field sorely needs continued development of novel methods for reducing this burden. While the concept of modular survey design is not new to market researchers, it has only recently made an appearance on the radar of survey methodologists. This type of survey design, which involves splitting a larger survey up into smaller parts that are completed independently over a short period of time at the convenience of the respondent, has the potential to reduce respondent burden. However, the optimal properties of these types of designs from a total survey error perspective remain unknown, making this a ripe area for future research. I will review recent literature in this area, and discuss potential designs of future studies that can take advantage of this type of measurement approach to collect representative samples of data in a way that is convenient for our study participants.