Biography: Dr. Namuun Clifford is a Mongolian American nurse practitioner and a postdoctoral fellow in the National Clinician Scholars Program at UCSF. Her research advances equitable precision digital health to personalize chronic disease prevention and management.
Background and Problem Statement: Cardiovascular diseases such as hypertension and heart failure remain leading causes of death and disability in the United States, disproportionately affecting underserved populations. Digital health interventions (DHIs) offer promise for expanding access to care and supporting self-management, yet the field has prioritized personalized intervention delivery while underinvesting in understanding who engages, how engagement unfolds, and what shapes it. Without this understanding, uniform deployment of DHIs risks widening rather than narrowing cardiovascular health disparities. Nursing science, with its strong history of self-management research, whole-person care, and equity-centered scholarship, is uniquely positioned to lead this work.
Purpose/Objective(s): Grounded in the Digital Health Equity and AIM-ACT frameworks, three studies advance precision digital health for CVD management:
1) Identify trends and gaps in precision digital health for hypertension through a scoping review.
2) Identify psychosocial-behavioral phenotypes of digital engagement among adults with heart failure (HF) and examine their association with longitudinal HF outcomes.
3) Characterize longitudinal trajectories of multi-device engagement among adults with HF and identify predictors of trajectory membership.
Methods
Aim 1 used Arksey and O'Malley's scoping review framework to synthesize 46 studies on digital precision hypertension management, charting trends across phenotyping, prediction, and personalized interventions. Aims 2 and 3 analyzed data from a 6-month, multi-site, NIH-funded decentralized randomized controlled trial of adults with HF recruited across 20 U.S. states using activity trackers, smart scales, and ecological momentary assessment (EMA).
Aim 2 (N=175) applied an unsupervised machine learning method of k-medoids clustering with Gower distance, chosen for its robustness to outliers and compatibility with mixed data types, to 10 baseline psychological, social, and behavioral variables; cluster validity was established through silhouette analysis and bootstrap stability (mean Jaccard indices 0.91–0.93 across 500 resamples). Between-phenotype differences in functional status, quality of life, self-care behaviors, and HF knowledge were examined over six months using linear mixed-effects models, with pattern-mixture and tipping-point sensitivity analyses for missing data.
Aim 3 (N=184) applied growth mixture models with random intercepts and slopes to weekly multi-device sync data over 24 weeks, deriving device-specific engagement trajectories that were cross-classified to identify multi-device engagement patterns. Multinomial logistic regression examined baseline predictors organized by Digital Health Equity framework domains, including demographic, clinical, physical and digital environment, and sociocultural factors. The Digital Health Equity and AIM-ACT frameworks guided variable selection, analytic design, and interpretation across all studies.
Together, the integration of unsupervised machine learning, longitudinal trajectory modeling, and multimodal data sources, including psychosocial surveys, EMA, and passive device-generated data, advances nursing methodology for studying digital engagement in cardiovascular disease self-management.
Results/Findings
The scoping review of 46 studies revealed that phenotyping remains the least developed precision approach and that social, digital, and environmental determinants are insufficiently integrated into existing precision health strategies.
Aim 2 identified three distinct psychosocial-behavioral phenotypes: Engaged Self-Regulators (42%), Activated Learners (31%), and Challenged Survivors (27%), distinguished by gradients of psychological distress, social and structural disadvantage, and device engagement. All phenotypes improved significantly over six months in functional status, quality of life, self-care behaviors, and HF knowledge; however, between-phenotype differences in functional status and quality of life were large and persistent, consistently 20–25 points higher for Engaged Self-Regulators.
Aim 3 identified three multi-device engagement trajectories: Sustained Engagement (33%), Differential Engagement (42%), and Early Disengagement (24%), with the steepest declines in the first weeks. HF duration was the only significant baseline predictor; baseline characteristics explained only modest variance, suggesting that early engagement behavior signals disengagement risk more than static factors.
Conclusion(s): Collectively, this dissertation demonstrates that digital engagement in cardiovascular disease management is multidimensional, dynamic, and shaped by the psychosocial and structural contexts in which people live and manage their health. Baseline psychosocial-behavioral phenotypes capture clinically meaningful heterogeneity present at enrollment and stable across six months, while longitudinal trajectory analysis identifies the first weeks of a digital intervention as a critical window for proactive risk identification. By integrating evidence synthesis, phenotyping, and longitudinal modeling, this dissertation provides both the initial conditions and dynamic signals needed to design digital interventions that adapt to patients, rather than asking patients to adapt to interventions.
Implications for Practice or System Change(s): These findings support a Phenotype-Informed Multimodal Adaptive Intervention (MADI) framework in which baseline psychosocial-behavioral phenotypes are assessed at enrollment to inform intervention assignment and level of support (slow timescale), while longitudinal engagement behavior guides just-in-time adaptive support (fast timescale). As a nursing-led model of care, the framework assigns each phenotype to a distinct intervention arm: a High-Touch Hybrid Intervention with nurse and social work support for Challenged Survivors, a Coached Hybrid Intervention with nurse coaching for Activated Learners, and a Self-Directed Digital Intervention for Engaged Self-Regulators. For nursing science, this dissertation extends self-management science—a cornerstone of nursing’s intellectual tradition—into the digital age, advancing nursing-led models of adaptive intervention design and methodology for studying engagement as a multidimensional, dynamic construct. For nursing practice, disengagement should be interpreted not as patient nonadherence, but as a clinically meaningful signal of patient burden, risk, and psychosocial context, warranting human-in-the-loop care escalation. For nursing education, curricula must integrate digital health literacy assessment, equity-informed intervention delivery, and foundational competencies in AI/ML. For health systems and policy, precision digital health must be designed as equity infrastructure to reduce, rather than widen, cardiovascular health disparities.