Abstract

Medication adherence remains a persistent challenge in chronic disease self-management, contributing to preventable complications, increased healthcare costs, and diminished quality of life. Although digital health interventions have historically relied on reminder systems and rule-based nudges, the rapid emergence of large language models (LLMs) introduces new possibilities for conversational, adaptive, and context-aware patient engagement. However, existing literature offers limited theoretical guidance for designing patient-facing generative systems, particularly in high-stakes healthcare contexts where unconstrained outputs may undermine user safety, autonomy, and trust. This study addresses this gap by asking: How can behavioral health theories be translated into architectural design constraints that shape LLM affordances for supportive medication adherence nudging? It advances the proposition that LLMs for medication adherence should be conceptualized as theory-guided, design-constrained socio-technical systems rather than neutral communication tools. Drawing on behavioral theories such as Self-Determination Theory and the Health Belief Model, alongside socio-technical systems perspectives, we argue that theoretical constructs must be translated into system boundary conditions that shape conversational behavior. We introduce the concept of design-constrained adaptivity, defined as a configuration in which generative personalization is structurally bounded by embedded role limits, content fidelity requirements, and transparency mechanisms. As an early-stage theoretical contribution, this work develops a conceptual framework that positions design constraints as the mediating layer between behavioral theory and LLM affordances. Rather than allowing open-ended generative freedom, the framework specifies how adaptive nudging can remain supportive without becoming coercive or clinically inappropriate. This research lays the foundation for a broader program on constrained generative systems in healthcare and outlines directions for future empirical validation, operationalization of design principles, and evaluation strategies.

Download

Share

COinS