Trust Calibration in Healthcare AI Under Incomplete Information: A Thematic Narrative Review and Conceptual Framework

Trust in healthcare artificial intelligence (AI) is beneficial only when it is warranted by the system's actual reliability, uncertainty and limitations. This paper presents a focused thematic narrative review of research on trust calibration in AI-assisted healthcare decision-making and develops a literature-derived conceptual framework. Targeted searches across health, computing and human-computer interaction databases produced a final reference set of 20 publications. Seven themes were synthesized: appropriate trust rather than high trust; the influence of output design; the double-edged role of explanations; distinctions between confidence and uncertainty signals; warning and verification prompts; measurement beyond self-report; and decision-making under incomplete information. The synthesis shows that explanations and confidence indicators may improve understanding or case-specific calibration, but can also increase unwarranted reliance and do not consistently improve decision quality. The proposed framework organizes the evidence into interrelated contextual, design, indicator, reference and outcome components, linked through an alignment judgment that compares the user's response with what the recommendation actually warrants. Appropriate reliance, rather than greater trust, is positioned as the central design and evaluation objective for healthcare AI.