Kevin A. Hallgren, Ph.D. - US grants

Individual drinking shows a consistent, positive correlation with social network drinking (Beattie, 2001;Longabaugh, Wirtz, Zywiak, &O'Malley, 2010;Project MATCH Research Group, 1997, 1998). This association is thought to be caused by two simultaneous interacting processes, where a heavy drinking social network influences an individual's drinking (i.e., social influence), while the individual's drinking influences his or hr selection of heavy drinking network members (i.e., social selection;Krull, Sher, &Jackson, 2007;Schulenberg 1999). These simultaneous, reciprocal processes create a positive feedback loop that in a network of social relationships leads to non-linear dynamic effects of drinking behavior. Such effects can be modestly understood when only the individual components of the system are sampled and analyzed;however, the full interacting network must be studied in its entirety to account for many the complexities observed. Gathering full social network data can be difficult and expensive, and including multiple observations over time adds further complications. Because of the sampling difficulties and presence of non-linear dynamic effects, computer simulations of social networks are often used to understand these systems. Simulations of the spread of HIV, infectious diseases, and obesity have provided useful strategies for targeting prevention and treatment, and have yielded additional, specific hypotheses that can be explored in future simulated or real- world networks (Bahr, Browning, Wyatt, &Hill, 2009;KosiDski &Grabowski, 2007;Kretzschmar &Weissing, 1998). One study has conducted preliminary simulations of alcohol dependence in social networks (Braun, Wilson, Pelesko, Buchanan, &Gleeson, 2006), and found that treating 8% of the alcohol-dependent individuals at random created an exponential decay in alcohol dependence rates in the system, but treating 4% or 6% did not. However, this research did not address other relevant hypotheses that may be guided by simulation studies, and several methodological factors limit the generalizations of this study's findings. The present study will generate computer simulations of drinking in social networks for the purpose of understanding how drinking spreads within a social network. Computer simulations will generate various types of stochastic actor-based networks (Snijders, van de Bunt, &Steglich, 2010;Watts, 1999) and will simultaneously model changes in individual drinking over time as a function of social network drinking (i.e., social influence), and changes in social network composition as a function of individual drinking (i.e., social selection). Individual- and system-level covariates will be inclued in the model, such as gender, individual- level susceptibility to developing an alcohol problem, and system-level efforts that increase or decrease alcohol consumption. Simulated networks will be manipulated to test which components, when targeted for treatment or prevention, create maximal effects in reducing alcohol problems for the larger network. Results will also inform hypotheses for future research studies that may use real-world observations of social networks.

Individual drinking shows a consistent, positive correlation with social network drinking (Beattie, 2001; Longabaugh, Wirtz, Zywiak, & O'Malley, 2010; Project MATCH Research Group, 1997, 1998). This association is thought to be caused by two simultaneous interacting processes, where a heavy drinking social network influences an individual's drinking (i.e., social influence), while the individual's drinking influences his or hr selection of heavy drinking network members (i.e., social selection; Krull, Sher, & Jackson, 2007; Schulenberg 1999). These simultaneous, reciprocal processes create a positive feedback loop that in a network of social relationships leads to non-linear dynamic effects of drinking behavior. Such effects can be modestly understood when only the individual components of the system are sampled and analyzed; however, the full interacting network must be studied in its entirety to account for many the complexities observed. Gathering full social network data can be difficult and expensive, and including multiple observations over time adds further complications. Because of the sampling difficulties and presence of non-linear dynamic effects, computer simulations of social networks are often used to understand these systems. Simulations of the spread of HIV, infectious diseases, and obesity have provided useful strategies for targeting prevention and treatment, and have yielded additional, specific hypotheses that can be explored in future simulated or real- world networks (Bahr, Browning, Wyatt, & Hill, 2009; KosiDski & Grabowski, 2007; Kretzschmar & Weissing, 1998). One study has conducted preliminary simulations of alcohol dependence in social networks (Braun, Wilson, Pelesko, Buchanan, & Gleeson, 2006), and found that treating 8% of the alcohol-dependent individuals at random created an exponential decay in alcohol dependence rates in the system, but treating 4% or 6% did not. However, this research did not address other relevant hypotheses that may be guided by simulation studies, and several methodological factors limit the generalizations of this study's findings. The present study will generate computer simulations of drinking in social networks for the purpose of understanding how drinking spreads within a social network. Computer simulations will generate various types of stochastic actor-based networks (Snijders, van de Bunt, & Steglich, 2010; Watts, 1999) and will simultaneously model changes in individual drinking over time as a function of social network drinking (i.e., social influence), and changes in social network composition as a function of individual drinking (i.e., social selection). Individual- and system-level covariates will be inclued in the model, such as gender, individual- level susceptibility to developing an alcohol problem, and system-level efforts that increase or decrease alcohol consumption. Simulated networks will be manipulated to test which components, when targeted for treatment or prevention, create maximal effects in reducing alcohol problems for the larger network. Results will also inform hypotheses for future research studies that may use real-world observations of social networks.

? DESCRIPTION (provided by applicant): Research on mechanisms of behavior change (MOBCs) has identified several processes that appear to facilitate successful drinking outcomes in alcohol use disorder (AUD) treatments. Often, different evidence- based treatments appear work through similar mechanisms, suggesting that constructs such as drink-refusal self-efficacy, craving, social support for abstinence, and other MOBCs are key variables that should be assessed, targeted in treatment, and monitored for change. Although the research base on MOBCs continues to expand, more work is needed to understand how to utilize MOBCs in real-world practice settings. In non- AUD contexts, measurement-based care research has shown that mental health symptoms improve more quickly and to a greater degree if they are routinely assessed, presented to patients, and discussed with clinicians. However, the impact of a similar approach for MOBCs - i.e., assessing, providing feedback, and discussing MOBCs as routine practice during treatment sessions- has not been tested. Moreover, the specific manner in which assessment, feedback, and discussion of MOBCs could be facilitated in real-world practice has not been explored. MOBC-based assessment tools could be particularly appealing to frontline clinicians given that (1) theoretical approaches often vary between clinicians despite the underlying MOBCs often being similar and (2) movement toward outcome-based performance measures and healthcare reform will require increased measurement and documentation of relevant treatment targets. The proposed K01 career development award provides training to support a research career focused on developing and testing tools that assist frontline clinicians and enhance the quality of their service delivery. Carefully devised training plans will provide training in (1) implementation science in addiction treatment, (2) user centered technology design, and (3) cross-disciplinary collaboration, research dissemination, and grant writing. These goals will be achieved through several means including mentorship from experts, coursework, clinical training, and hands-on experience. The proposed research project complements these training aims by developing a computer-based clinical support tool that is implementation-focused and helps frontline clinicians assess and review patient MOBC data during their treatment sessions. The project will include phases to (1) assess opinions, needs, and desires of stakeholders around developing such a clinical support tool, (2) iteratively test and modify a software prototype to increase the ease of using the tool and understanding its feedback, and (3) test the tool's feasibility, acceptability, and impact on clinicians' within-session behavior. The research will take place at a publicly-funded treatment setting with frontline clinicians and patients. The award will complement the candidate's existing strengths in MOBC research by providing training in new domains involving cross- disciplinary research partnerships, user-centered design, and implementation-focused research.

Project Summary. Primary care (PC) is well-positioned to detect and address unhealthy drinking and alcohol use disorder (AUD), yet most PC settings have notable gaps in providing alcohol-related care. One way to improve alcohol-related care is to use standardized measurements to screen for unhealthy alcohol use and to assess AUD symptoms, which informs clinical care by detecting unhealthy drinking, assessing AUD severity, informing decisions about AUD treatment intensity, and monitoring clinical outcomes over time. Researchers and national agencies are increasingly calling for the use of standardized measures of unhealthy drinking and AUD symptoms in routine PC settings; however, surprisingly little is known about how such measures perform or what they indicate about health risks when they are used in the context of routine PC?i.e., administered in routine appointments and documented in electronic health records (EHRs). We propose to leverage a large and novel EHR dataset from Kaiser Permanente Washington, a large integrated regional health system. The dataset will include: the AUDIT-C alcohol screening measure, completed as part of an annual screen by >250,000 patients (89% of adults who attend PC appointments); a novel patient-reported AUD symptom checklist that is based on the 11 DSM-5 AUD criteria (?DSM-5 checklist?), completed by over 4,000 patients (>70% of those with AUDIT-C scores ? 7); and diverse health outcome measures available from EHR, administrative, and claims data. The 2-year developmental R21 phase includes psychometric analyses that evaluate the DSM-5 checklist as a scaled measure of AUD severity and the consistency of its performance across demographically diverse subgroups; to our awareness, this is the first study to evaluate any standardized patient-reported AUD symptom measure integrated into routine PC. Cross-sectional analyses will examine associations between the AUDIT-C, DSM-5 checklist, and important health outcome measures known to be associated with drinking, including systolic blood pressure, patient-reported depression symptoms, hospitalizations for alcohol-attributable conditions, all-cause hospitalizations, and all-cause mortality. The R21 will also achieve milestones demonstrating readiness for the R33 phase. During the 3-year R33 phase, longitudinal analyses will test whether the AUDIT-C and DSM-5 checklist are associated with subsequent health outcomes and whether within-person changes in AUDIT-C scores over time are associated with changes in health outcomes. To our awareness, this work will be the first to evaluate a patient-reported AUD symptom checklist in any routine PC setting and the first to evaluate the AUDIT-C in any non-veteran routine PC setting. This research will improve clinical care by helping providers and patients understand the information provided by these scaled patient-reported measures and their associations with important adverse health outcomes. The work will also provide a foundation for future studies that use patient-reported EHR- based measures in pragmatic studies, implementation trials, and quality improvement initiatives.