From Behavioral Theory to Chairside Decision Support: Development of ELIGE AI, a Hybrid Rule-Based and Large Language Model Tool for the Behavioral Management of Dental Caries

Background: Dental caries is the most prevalent noncommunicable disease worldwide. Within the DCIEH framework it is a dysbiosis of the dental biofilm microbiome that is regulated by biology, modulated by behavior and conditioned by the environment. It is also a trans-complex disease whose outcome emerges from the interaction of biological, behavioral and socio-environmental factors. Behavioral sciences offer mature theories to explain and change oral health behaviors, but a persistent gap separates these theories from routine chairside care. In 2024 we proposed ELIGE/CHOOSE, a theoretical framework that integrates Social Cognitive Theory (SCT) and the Capability-Opportunity-Motivation-Behavior (COM-B) model into three clinical phases: Identify, Articulate and Plan.Objective: To report the systematic development of ELIGE AI, a clinical decision support tool that operationalizes the ELIGE/CHOOSE framework through an evidence-based, author-designed deterministic algorithm combined with a large language model (LLM).Methods: Development was guided by the Medical Research Council framework for complex interventions, the ORBIT model and the IDEAS framework for digital behavior change interventions. Reporting draws on TIDieR, DECIDE-AI and TRIPOD-LLM. Development proceeded in five phases: (1) theoretical integration of SCT, COM-B, the Transtheoretical Model, the Health Action Process Approach, motivational interviewing and Intervention Mapping; (2) translation into a stepwise clinical protocol; (3) formalization of the first author's clinical reasoning as an explicit decision algorithm, with every rule traceable to theory or evidence; (4) integration of an LLM restricted to bounded language tasks; and (5) iterative build-and-refine cycles across versions informed by clinical use.Results: ELIGE AI (v3.9) is a web application with seven screens across the three ELIGE/CHOOSE phases plus a follow-up module. In about 20 to 25 minutes it produces a specified behavior, a principal COM-B determinant with a displayed rationale, a graded goal ladder, Behaviour Change Technique Taxonomy v1 techniques matched to the determinant, rule-based referrals and personalized patient materials. At follow-up it re-measures the same constructs, behavior frequency and dental caries risk, and recommends the next step on the same behavioral route. The algorithm makes every clinical decision; the LLM only drafts and classifies; the clinician validates and can override every output.Conclusions: ELIGE AI shows how a published behavioral framework and its author's clinical reasoning can be encoded as a transparent algorithm and safely combined with generative AI. The tool now enters a three-stage evaluation of usability, feasibility and validity, and efficacy on dental caries outcomes.