Design and operational feasibility of an AI-enabled, physician-led clinical judgment layer for healthcare navigation platform (Preprint)

BACKGROUND Patients frequently report difficulty understanding diagnoses, treatment options, and recommended next steps within fragmented healthcare systems, contributing to anxiety, decisional conflict, and incomplete follow-through.¹,² Time-limited encounters, increasing diagnostic complexity, and care across multiple settings further challenge patient comprehension and engagement. While AI-enabled digital health tools have expanded access to medical information, many systems function primarily as information delivery or automation tools and lack explicit mechanisms for physician-led clinical judgment or integration of patient-specific records.³,⁴ Advances in artificial intelligence support medical document aggregation and organization,⁵ yet unclear delineation between algorithmic processing and physician accountability has raised concerns regarding safety, trust, and clinical responsibility.⁶,⁸ OBJECTIVE To assess the design, implementation, and operational feasibility of an AI-enabled healthcare navigation platform structured to separate AI-based information aggregation from physician-led clinical judgment, interpretation, and patient guidance in real-world use. METHODS We conducted a prospective pilot service evaluation of an AI-enabled, physician-led healthcare navigation platform. Fifty participants completed at least one structured navigation encounter over three months. AI components were limited to aggregation, organization, and summarization of medical records, while licensed physicians provided all interpretation, recommendations, and guidance.⁹,¹⁰ Outcomes were analyzed descriptively to characterize feasibility-relevant signals, including utilization, workflow completion, patient-reported clarity and decision confidence, follow-through on physician-recommended actions, patient experience measures, and qualitative feedback, without hypothesis testing or causal attribution. RESULTS Utilization patterns reflected sustained engagement and acceptable data completeness for feasibility assessment. Mean patient-reported clarity scores were higher following navigation (8.3, SD 1.0) compared with pre-navigation assessment (3.6, SD 1.4). Observed score differences were broadly distributed across participants. Engagement included repeated navigation encounters and medical document review. Physician-recommended follow-through actions were common, with timely completion reported. Patient experience ratings indicated acceptability of the physician-led model. CONCLUSIONS This pilot evaluation provides descriptive evidence supporting the implementation feasibility of an AI-enabled, physician-led clinical judgment layer for healthcare navigation. Observed signals reflect system use and workflow execution rather than clinical effectiveness and inform the design of future controlled evaluations.¹¹ CLINICALTRIAL This work was conducted as a prospective pilot service evaluation and was not registered as a clinical trial.¹¹