Artificial Intelligence (AI)-Driven Event Scheduling for Pathology Resident Rotations: A Novel Schema-Based Approach

Document Type

Conference Proceeding

Publication Date

3-2025

Publication Title

Laboratory Investigation

Abstract

Background: Event scheduling is critical in activity planning involving complex team dynamics and diverse task requirements. Traditional scheduling methods often struggle to balance multiple variables, constraints, and preferences and are time-consuming. This study explores a novel schema-based approach utilizing Artificial Intelligence Large Language Models (AI-LLMs) to generate comprehensive and flexible monthly schedules for pathology resident service. Design: A schema was developed to define critical elements in routine recurrent schedules, such as dates, team members, tasks, constraints, scheduling cycles, and rules. The schema was examined/confirmed for feasibility. Example schedules were provided as format-style templates for AI to adopt. An AI-LLM Claude Sonnet by Anthropic or ChatGPT-3.5 by OpenAI, was prompted to generate new schedules based on the schema and example schedules. AI-generated schedules were then evaluated by a human scheduler and a different AI-LLM for consistency with the schema, feasibility of execution, and flexibility in accommodating various operational scenarios. The method was tested with two different AI-LLMs (see Table note). Results: Each AI-generated schedule covers 30-31 days with five residents at three different PGY levels and 24 different tasks (see Table). AI-LLMs can generate complex operation schedules when provided with explicit, well-defined schema. The AI-generated schedules were consistent with the schema, firmly adhering to constraints and task requirements. However, flexibility in adapting to unforeseen changes varied, emphasizing the importance of schema precision and scenario anticipation. Some minor modifications may be inevitable to accommodate unpredictable individual needs that the scheme cannot include. Two different AI-LLMs (ChatGPT-3.5 by OpenAI and Claude-3.5 by Anthropic) were used. Schedules generated by Claude are presented. The AI-generated schedule covers 30-31 days with 5 residents at 3 different PGY levels (shown in the top row), 24 tasks abbreviated (e.g., BG¼ gross big specimen, etc.) in date rows. Conclusions: This study demonstrated using AI-LLMs to schedule complex resident activities with our schema-based method. This method can be adopted by other residency programs with the same training components required by the American Board of Pathology. Such an AI-driven solution to time-consuming recurrent event scheduling is expected to significantly impact many organizations/disciplines that depend on effective and accurate event scheduling.

Volume

105

Issue

3 Suppl

First Page

16

Comments

114th Annual Meeting of the United States and Canadian Academy of Pathology (USCAP), March 22-27, 2025, Boston, MA

Last Page

17

DOI

10.1016/j.labinv.2024.102727

Link to Full Text

Share

COinS