A Tailoring Algorithm to Optimize Behavior Change Janet Brigham, Harold S. Javitz, Ruth E. Krasnow, Lisa M. Jack, Gary E. Swan Center for Health Sciences, SRI International janet.brigham@sri.com Abstract Effective computerized tailoring can enhance the impact of health interventions. Long-term success rates can be improved with prospective tailoring that builds on evidence-based research. A new algorithm, developed with data from smoking cessation clinical trials and published practice guidelines, is designed to predict the likelihood of abstinence. The algorithm prioritizes the content of a stop-smoking intervention individually for each user and indicates the potential effect of utilizing various stop-smoking medications and stop-smoking approaches. Thus, it has the potential to guide a smoker through the cessation process by dynamically optimizing the likelihood of success. Importantly, the algorithm predicts that even a daily smoker may be able to substantially improve the likelihood of quitting and staying quit both by using stop-smoking techniques and medications and by addressing emotional and cognitive issues that sustain smoking. 1. Background 1.1. Tailoring in health interventions Applying tailoring (also called personalization or customization) to health interventions is challenging for many reasons, including the realities that health behaviors are complex and involve multiple etiologies, numerous variables, and various pathways to success. Similarly, identifying and applying effective tailoring is an exacting process because so many situations and behaviors can impact health consequences, including treatment adherence, physical activity, nutrition, management of chronic conditions, and varying levels of risk. Tailored health interventions could enable powerful treatments to successfully help individuals and populations at risk [1]. Tailoring has been effective in diverse settings, addressing varied behaviors and populations [2-5]. Tailoring could have the potential to be a “disruptive” technique [6] that could actualize the potential of computerized interventions [7] and perhaps revolutionize health interventions. Several factors can make tailoring more effective. A Cochrane Review [2] determined that interventions that are tailored to identify barriers to behavior change prospectively can improve outcomes. Dynamically adaptive, computer-based tailored interventions are overall more effective than single-assessment inter- ventions [1]. Theory-based tailoring adds additional potential power to a tailoring approach. However, theory can be applied only if enough variables are explored to make theory operationalizable. Evidence indicates that theory-based interventions are more effective than those without a theoretical grounding. Also important is the use of multiple determinants (factors that influence the nature or outcome) and multiple levels of those determinants [8]. For example, high perceived stress could affect the outcome of a smoking cessation attempt. Since stress perception is scored as ordinal- level data, it can be scored at multiple levels, or in other words, is a determinant with multiple levels. Tailoring often has been based on basic demo- graphics and on linear criteria such as the Stages of Change (Transtheoretical Model) [9]. The Transtheo- retical Model has been applied to numerous health behaviors and can be helpful in explaining why people at high risk might not be prepared to make a significant behavior change, but its linear nature does not necess- arily describe the often nonlinear process of making those changes. Which is to say, while it may be instruc- tive in defining an individual’s status, it may have limited predictive power. Although the Transtheoretical Model depicts stages in the change process in what can be a dynamic approach, it has neither multiple determinants nor multiple levels within determinants. 1.2 Computer-tailored interventions Employing a computerized approach to tailoring allows the tailoring to be based on combinations of variables representing needs, risk factors, psychosocial factors, and biological factors. The necessary assess- ment and decision steps have become increasingly auto- mated with the advent of computer-driven treatments, 2014 47th Hawaii International Conference on System Science 978-1-4799-2504-9/14 $31.00 © 2014 IEEE DOI 10.1109/HICSS.2014.334 2645