Besa Bauta, PhD, MPH, MSW
New York University, Silver School of Social Work
"Digital mental health interventions optimize engagement metrics such as logins and completion rates, assuming higher engagement improves outcomes. However, engagement alone does not consistently reduce distress. We applied data-driven clustering to identify behavioral phenotypes for precision digital mental health design.
Baseline data from adults enrolled in a digital behavioral toolkit trial (N = 181) were analyzed. Technology engagement was calculated as a mean of seven items and distress as a mean of six items. Standardized age, engagement, and distress were entered into k-means clustering. Elbow and silhouette diagnostics supported a four-cluster solution (mean silhouette = 0.30).
Four balanced phenotypes emerged (23–27% of the sample): Digitally Strained (younger, high engagement, high distress); Digitally Adaptive Older (older, high engagement, lower distress); Tech-Fluent and Resilient (younger, high engagement, low distress); and Disconnected and Distressed (older, low engagement, elevated distress). Clusters differed across age, engagement, and distress (p < .001).
High engagement appeared in high and low-distress groups, demonstrating that engagement alone does not buffer psychological burden. Converting engagement metrics into computational phenotypes revealed clinically meaningful heterogeneity that enables AI-informed tailored intervention pathways beyond age-based personalization, supporting adaptive onboarding and engagement–distress–aligned support. "
Baseline data from adults enrolled in a digital behavioral toolkit trial (N = 181) were analyzed. Technology engagement was calculated as a mean of seven items and distress as a mean of six items. Standardized age, engagement, and distress were entered into k-means clustering. Elbow and silhouette diagnostics supported a four-cluster solution (mean silhouette = 0.30).
Four balanced phenotypes emerged (23–27% of the sample): Digitally Strained (younger, high engagement, high distress); Digitally Adaptive Older (older, high engagement, lower distress); Tech-Fluent and Resilient (younger, high engagement, low distress); and Disconnected and Distressed (older, low engagement, elevated distress). Clusters differed across age, engagement, and distress (p < .001).
High engagement appeared in high and low-distress groups, demonstrating that engagement alone does not buffer psychological burden. Converting engagement metrics into computational phenotypes revealed clinically meaningful heterogeneity that enables AI-informed tailored intervention pathways beyond age-based personalization, supporting adaptive onboarding and engagement–distress–aligned support. "
