Estimating the impact of engagement with digital health interventions on patient outcomes in randomized trials.

Abstract

Guidance is needed on studying engagement and treatment effects in digital health interventions, including levels required for benefit. We evaluated multiple analytic approaches for understanding the association between engagement and clinical outcomes. We defined engagement as intervention participants' response rate to interactive text messages, and considered moderation, standard regression, mediation, and a modified instrumental variable (IV) analysis to investigate the relationship between engagement and clinical outcomes. We applied each approach to two randomized controlled trials featuring text message content in the intervention: REACH (Rapid Encouragement/Education and Communications for Health), which targeted diabetes, and VERB (Vanderbilt Emergency Room Bundle), which targeted hypertension. In REACH, the treatment effect on hemoglobin A1c was estimated to be -0.73% (95% CI: [-1.29, -0.21]; P = 0.008), and in VERB, the treatment effect on systolic blood pressure was estimated to be -10.1 mmHg (95% CI: [-17.7, -2.8]; P = 0.007). Only the IV analyses suggested an effect of engagement on outcomes; the difference in treatment effects between engagers and non-engagers was -0.29% to -0.51% in the REACH study and -1.08 to -3.25 mmHg in the VERB study. Standard regression and mediation have less power than a modified IV analysis, but the IV approach requires specification of assumptions. This is the first review of the strengths and limitations of various approaches to evaluating the impact of engagement on outcomes. Understanding the role of engagement in digital health interventions can help reveal when and how these interventions achieve desired outcomes.