Abstract
Automated microbial identification and antibiotic susceptibility testing systems (automated ID/AST systems) are widely used in clinical microbiology, yet ensuring standardized and accurate report generation remains challenging. Large language models, such as ChatGPT, offer potential for improving report consistency and objectivity through structured prompt engineering. This study evaluates the effectiveness of ChatGPT in generating standardized microbiology reports for automated ID/AST systems, compared to clinical microbiologists (CM). ChatGPT was provided with structured prompts based on clinical & laboratory standards institute (CLSI) guidelines to generate automated ID/AST systems reports. A prompt engineering framework was developed to enhance AI-generated reports. Performance was assessed across five dimensions: accuracy, relevance, objectivity, completeness, and clarity. Eight clinical cases were analyzed, comparing reports from three groups: CM, ChatGPT before prompt training (ChatGPT_BT), and ChatGPT after prompt training (ChatGPT_AT). The ChatGPT_BT group demonstrated higher relevance and completeness compared to the CM' group (p < 0.0001 and p < 0.0001). After training, the ChatGPT_AT group produced reports with significantly improved quality across all five dimensions (p < 0.001, p < 0.0001, p < 0.0001, p < 0.0001 and p < 0.0001). Moreover, the ChatGPT_AT group showed notable improvements in relevance, objectivity, completeness, and clarity compared to the ChatGPT_BT (p < 0.001, p < 0.0001, p < 0.0001 and p < 0.05), with no significant difference in accuracy (p ≥ 0.05). ChatGPT, when guided by a structured prompt engineering process, shows significant potential in assisting CM by enhancing the objectivity, clarity, completeness, relevance, and accuracy of automated ID/AST systems reports.