Poor eating habits, particularly low fruit and vegetable intake, is a growing, serious public health concern, particularly among young adults age 21-30, referred to as Generation Y (GenY). GenY's poor dietary practices are associated with the onset of obesity and many chronic diseases, such as type 2 diabetes, as well as declines in predicted health status and life expectancy. Thus, there is a need to develop effective interventions to improve GenY's eating habits. MENU GenY is a computer-based intervention to encourage increased fruit and vegetable intake among GenY. A critical component of MENU GenY is personalized eCoaching. eCoaches use email to deliver motivation-enhancing coaching to encourage healthy eating, grounded in the principles of Motivational Interviewing (MI), an evidence-based communication technique to increase intrinsic motivation and self-efficacy for behavior change. The MI model posits that counselor's use of ?MI-consistent? communication techniques are responsible for eliciting behavior change through patient ?change talk? (i.e., statements about one's own desire, ability, reasons, need for or commitment to behavior change). A growing body of empirical evidence links change talk to behavior change, but research identifying the specific provider behaviors that elicit patient change talk is limited to specific populations (mainly adults who abuse substances and a couple studies of adolescents). Identifying specific communication strategies linked to behavior change and integrating these strategies into communication-based interventions (e.g., brief, motivation-enhancing interventions delivered in a variety of settings or public health initiatives) can increase these interventions' potency. However, a significant barrier to this research is the qualitative methods traditionally used to analyze the communication process which are resource-intensive, requiring an iterative process of human (subjective) interpretation of text. Rapidly developing computational technologies, specifically machine learning combined with classification models, offer a unique opportunity to accelerate this process. Our research group has recently applied machine learning-based data mining models to similar communication data. We automated a simple communication code scheme to characterize patient communication and achieved accuracy comparable to human coders. The goals of this study are to leverage innovative computer science machine learning and classification models to fully automate the communication coding process and link patterns in eCoach-patient communication to increases in fruit and vegetable intake. We propose a secondary analysis of data collected for a NICHD randomized clinical trial (R01 HD067314). The sample is 160 members of GenY drawn from both urban and rural settings (Detroit metropolitan area and rural Pennsylvania) with outcomes measured at baseline and 3 months. Our validated approach will accelerate the pace of outcomes-oriented communication research and identify effective communication strategies linked to healthy eating. These findings will be used to tailor interventions and public health messages and develop automated eCoaching.
This research will use cutting edge computer science technology ? machine learning-based classification models ? to automate communication coding, a resource-intensive qualitative analysis process, and identify key communication strategies used by eCoaches linked to increased fruit and vegetable (FV) intake among young adults age 21-30. Identifying communication strategies linked to increased FV intake will identify empirically derived communication techniques that can be used to tailor interventions and health promotion messages to more effectively promote behavior change in this age group. The development of this computer science technology is the first step toward automating the eCoaching process.
| Hasan, Mehedi; Kotov, Alexander; Carcone, April Idalski et al. (2018) Predicting the Outcome of Patient-Provider Communication Sequences using Recurrent Neural Networks and Probabilistic Models. AMIA Jt Summits Transl Sci Proc 2017:64-73 |
