Guiding Clostridioides difficile Infection Prevention Efforts in a Hospital Setting With AI.

Importance: Increasingly, artificial intelligence (AI) is being used to develop models that can identify patients at high risk for adverse outcomes. However, the clinical impact of these models remains largely unrealized. Objective: To evaluate the association of an AI-guided infection prevention bundle with Clostridioides difficile infection (CDI) incidence in a hospital setting. Design, Setting, and Participants: This prospective, single-center quality improvement study evaluated adult inpatient hospitalizations before (September 1, 2021, to August 31, 2022) and after (January 1, 2023, to December 31, 2023) AI implementation. Data analysis was performed from January to August 2024. Intervention: A previously validated institution-specific AI model for CDI risk prediction was integrated into clinical workflows at the study site. The model was used to guide infection prevention practices for reducing pathogen exposure through enhanced hand hygiene and reducing host susceptibility through antimicrobial stewardship. Main Outcomes and Measures: The primary outcome was CDI incidence rate. Secondary outcomes included antimicrobial use and qualitative assessments of bundle implementation. Results: Pre-AI and post-AI samples included 39 046 (21 645 [55.4%] female; median [IQR] age, 58 [36-70] years) and 40 515 (22 575 [55.7%] female; median [IQR] age, 58 [37-70] years) hospitalizations, respectively. After adjusting for differences in clinical characteristics, there was no significant reduction in CDI incidence (pre-AI period: 5.76 per 10 000 patient-days vs post-AI period: 5.65 per 10 000 patient-days; absolute difference, -0.11; 95% CI, -1.43 to 1.18; P = .85). Relative reductions greater than 10% in normalized antimicrobial days were seen for piperacillin-tazobactam (-9.64; 95% CI, -12.93 to -6.28; P < .001) and clindamycin (-1.04; 95% CI, -1.60 to -0.47; P = .03), especially for high-risk patients alerted by AI (relative reduction for piperacillin-tazobactam, 16.8%; 95% CI, 8.0%-24.6%). On the basis of qualitative assessments via semistructured interviews and field observations, the study found that health care staff's experiences with AI-guided workflows varied. In particular, the enhanced hand hygiene protocols were met with poor adherence, whereas pharmacists consistently engaged with the alerts. Conclusions and Relevance: In this quality improvement study, the implementation of an AI-guided infection prevention bundle was not associated with a significant reduction in the already low CDI incidence rate at the study site, but it was associated with reduced CDI-associated antimicrobial use. The results highlight the potential of AI in supporting antimicrobial stewardship. Barriers to implementation, including infrastructure, staff knowledge, and workflow integration, need to be addressed in future applications.