OBJECTIVES

: Scalable, accessible interventions to increase physical activity in underserved patients are needed. We pilot-tested an mHealth intervention to help ethnic-minority patients set weekly walking goals, ultimately aiming to create an automated virtual walking coach.

METHODS

: Sedentary adults ages 21-65 years with body mass index (BMI) >25 kg/m² and a smartphone were recruited from a minority-serving academic primary care clinic. Patients received a Fitbit and 8 weeks of two-way text-based coaching. Using texts, patients set weekly specific, measurable, attainable, relevant, timely (SMART) step goals with reminders and encouragement from their coach. In addition to descriptive statistics, a linear mixed-effects model assessed overall trends in daily steps. Linear regression assessed trends within patients. Bivariate linear mixed-effects models examined associations between patient characteristics and daily steps.

RESULTS

: The analytic sample consisted of 1,323 days with step data from 28 patients with multiple comorbidities. Twenty-one (75.0%) patients were female, 23 (82.1%) were African American, 4 (14.3%) were Hispanic, and 1 (3.6%) was White. Average baseline age was 47.3 years (SD=9.9), weight was 237.3 pounds (SD=58.6), and BMI was 39.3 kg/m² (SD=9.3). Daily average number of steps was significantly higher in Week 8 than in Week 1 (8,336 [SD=3,913] vs. 7,018 [SD=3,704]; p=0.04). Each day in the intervention was associated with a non-significant average increase of 9 steps (p=0.17). Seven (25.0%) patients had a significant increase in daily steps during the intervention, 2 (7.1%) had a significant decrease, and 19 (67.9%) had no significant change. In bivariate analyses, variables significantly associated with increased steps were younger age, lower BMI, Hispanic and White vs. African American ethnicity, higher self-reported health, and employment vs. unemployment.

CONCLUSIONS

: Weekly facilitated text-based goal-setting showed early promise towards increasing steps in vulnerable patients. Future work will use imitation learning, sentiment analysis, dialogue modeling, and behavior change theory towards developing an automated virtual walking coach.