Key Highlights and Action Items

This workshop was the second in a series of meetings sponsored by the NIH OBSSR and NIH-wide Adherence Research Network which aims to develop a better understanding of health behavior maintenance to promote and sustain positive health outcomes.

The workshop focused on the operationalization of metrics and measurement for behavior maintenance and innovative tools and approaches for monitoring behavior maintenance. Many key health outcomes depend on behavior maintenance, or consistently performing a health behavior over time and across contexts. Maintenance can take many forms, so researchers must determine how to operationalize and measure both maintenance and deviations from maintenance. 

This one-day virtual workshop featured invited speakers with expertise in diverse areas of health behavior research. It was open to the public and more than 300 attendees joined the meeting. The first session in this workshop focused on the definitions of and criteria for maintenance that are utilized across a range of behavioral domains, including smoking cessation, substance use, physical activity, weight loss, and HIV preventive medicine. The second session focused on measurement strategies, including discussion of innovative tools and methodologies to monitor behavior maintenance in the context of daily life, including ecologic momentary assessment, electronic adherence monitoring, and passive sensors. 

Operationalizing Behavior Maintenance and Associated Determinants

The state of the science differs across the various behavioral domains that were examined in this session. The following considerations were discussed as needing further evidence. One cross-cutting theme regards lack of common definitions and terminology for mechanisms of action, targets of interest/outcomes, and time frame of associations between mechanisms of action effects on outcomes. These limitations can also interfere with attribution of effects.

• Behavior maintenance is dynamic—the process of adopting and maintaining behaviors is highly variable and contextually fluid. Maintenance may be defined in many ways, and definitions and thresholds of maintenance may differ across behaviors.
• Maintenance can take many forms. A key consideration for operationalization is whether behavior maintenance involves acts of omission (e.g., smoking cessation), or acts of commission, (e.g., executing a behavior routinely or in the same way). Operationalizing behavior maintenance may present additional challenges when the behavior is one of commission where a threshold for maintenance needs to be explicitly defined.  In either case, criteria need to be specified that indicate what someone is deviating from maintenance (e.g., indicators of slips or lapses).  In some behavioral domains (e.g., smoking cessation) these criteria have been specified, but in other behavioral domains (e.g., physical activity) they have not.
• Standards for operationalizing behavior maintenance are generally lacking across health domains. Clinical guidelines and treatment recommendations are helpful where available. Thresholds for achieving behavior maintenance have generally not been defined across behaviors and disease targets of interest, with some notable exceptions (e.g., smoking cessation). The diversity and variability of health behaviors makes universal criteria for maintenance elusive. 
•  Health behaviors that require intermittent rather than routine use may present an added challenge for measurement. These may require assessment of other concurrent behaviors or contextual factors to evaluate their practice. For example, medication use may require daily or intermittent maintenance, and circumstances under which use is required (e.g., long- vs. short-term) may vary.
• Psychosocial mechanisms of action likely vary between initiation and maintenance processes. These underlying psychosocial mechanisms of action may be just as important for understanding maintenance as outwardly observable behaviors themselves, and thus must be incorporated into assessments. Some psychosocial factors must be assessed through self-reporting, but synergistically, passive, objective methods provide value to the understanding of, reproducibility and validity of measurements. 
• Incorporating measurements of psychosocial processes and contextual factors alongside longitudinal measurement of behaviors may help us to better understand the processes that underlie the maintenance of behaviors. For example, the assessment of mechanisms of action leading to observable behaviors could potentially allow researchers to differentiate between an individual who is maintaining a behavior automatically and an individual whose behavior maintenance requires a high-effort decision-making process. 

Assessment Tools and Considerations

• Methods of monitoring individual behavior that are sufficiently sensitive to detect changes that predict or precede lapses in maintenance are needed. 
• Ecological momentary assessments provide opportunities to collect fine-grained data on behavior and behavioral determinants during daily living and may minimize challenges posed by recall and self-report biases. Understanding these dynamic processes is important for identifying the underlying mechanisms of behavior change and determining effective treatments.
• Technology-based tools can help behavioral maintenance researchers assess dynamic processes and patterns of behavior. Passive digital monitoring of activities through tools such as smartphones, accelerometers, and electronic pillboxes can generate rich data on dynamic variables that are valuable for both monitoring behavior and informing intervention timing and targets. Although these tools have the potential to advance the field, the state of the science regarding the utility of these tools varies across behavioral domains. 
• Wearable sensors can measure physiologic processes, which can be used to estimate stress and emotion. These sensors have the potential to capture information about these states and to examine how they affect and are affected by behavior (both maintenance and deviations from maintenance). 
• Predictive models can be generated by combining multiple sources of data – including data captured through technology-based tools -- through analytic innovations from artificial intelligence, which can then be tested for predictive validity in selected conditions with selected populations. 
• These technology-based tools may facilitate innovation, but researchers must be mindful that they may sustain or increase disparities. Researchers must be mindful of whether and how technology can be used in different settings; in particular, the sociocultural and sociopolitical factors that can influence use of technology, such as bias in overall adoption of technology, trust in health care systems to use data equitably, and data privacy protections. 

Assessment and Intervention are Interconnected.

• Active monitoring and effective interventions can be closely intertwined. Monitoring systems can be used to trigger, develop, and direct interventions; data from behavioral monitoring can also serve as an intervention itself by providing feedback to individuals on their behavior. In addition, monitoring can identify patterns, facilitators, and barriers, and that information can be used to program interventions.
• There may be particular value in the delivery of interventions at key junctures, such as just-in-time interventions that use information gathered from ecological momentary assessments to target the processes associated with lapses. Just-in-time interventions delivered through tools such as mobile apps could be considered “training wheels” to help participants develop maintenance habits at key time points rather than long-term interventions. Providing support to teach people how to use the technology and engage with the intervention is critical to its success. 
• The type and level of motivation may differ across the phases of behavior change.  At the outset of treatment, getting better may be a high priority, but over time may become less important making continuing medication more difficult. This may apply to any health behavior. Feedback from behavioral monitoring and strong habit formation have been shown to maximize medication use as prescribed; Monitoring cues and rewards, as well as temporal consistency, are important for building habits and contingencies for successful behavior execution.

Research Gaps and Future Directions

• Tools and technology evidence generated in highly resourced research settings needs to be evaluated in situ with intended recipients with evidence based contextual resources.
• Data aggregation methods that produce predictive models needs to be validated with intended recipients in intended settings.
• The ethics of use of AI to develop algorithms for intervention approaches needs to be monitored for bias and privacy in research and clinical applications.
• Preconditions for use of technologies needs to be established and implemented to accurately test efficacy of these tools.