Key takeaways:
- An AI-guided an infection prevention bundle didn’t considerably scale back incidence of C. difficile an infection in a hospital.
- Nonetheless, the bundle did scale back use of antimicrobials that would enhance danger for an infection.
An AI-guided an infection prevention software didn’t considerably scale back incidence of Clostridioides difficile an infection in a hospital setting, but it surely did scale back use of antimicrobials that enhance danger for an infection, in line with examine outcomes.
Jenna Wiens
“Such danger fashions, that are developed utilizing machine studying, might be built-in into medical workflows to assist information antimicrobial stewardship,” Jenna Wiens, PhD, examine writer and affiliate professor of laptop science and engineering on the College of Michigan, informed Healio. “Nonetheless, it’s necessary to contemplate how the mannequin is built-in, since massive modifications to workflows might have restricted impact.”
Knowledge derived from Tang S, et al. JAMA Netw Open. 2025;doi:10.1001/jamanetworkopen.2025.15213.
The CDC named C. difficile one of many prime 5 pressing threats in 2019, Wiens stated, and it stays “stubbornly prevalent in hospitals.”
Wiens and colleagues had developed an AI-based mannequin utilizing knowledge from digital well being information that would flag sufferers in danger for C. difficile a mean of 5 days prematurely of an infection. Nonetheless, they didn’t know whether or not the mannequin may have an effect on affected person outcomes in medical apply.
“We sought to combine this danger mannequin to assist information an infection prevention efforts at Michigan Medication,” Wiens stated.
In a potential, high quality enchancment examine, the researchers assessed grownup inpatient hospitalizations earlier than (September 2021 to August 2022) and after (January 2023 to December 2023) implementation of the AI software.
An infection prevention efforts, which focused high-risk hospitalizations, included encouraging soap-and-water handwashing to scale back pathogen publicity and antimicrobial oversight to scale back host susceptibility.
These efforts had been carried out by way of greatest apply advisory alerts, medicine critiques by inpatient pharmacists and well being report critiques by examine physicians.
Incidence of C. difficile served as the first final result, and antimicrobial use and implementation of the bundle served as secondary outcomes.
The pre-AI pattern included 39,046 hospitalizations (median age, 58 years; interquartile vary, 36-70; 55.4% ladies), and the post-AI pattern included 40,415 hospitalizations (median age, 58 years; interquartile vary, 37-70; 55.7% ladies).
Outcomes confirmed 127 circumstances of C. difficile earlier than AI implementation and 148 circumstances after, for an adjusted incidence charge of 5.76 (95% CI, 4.87-6.69) per 10,000 patient-days vs. 5.65 (95% CI, 4.78-6.56) per 10,000 patient-days (absolute distinction, –0.11; 95% CI, –1.43 to 1.18) — a distinction that didn’t attain statistical significance.
Nonetheless, the researchers did observe reductions in use of ampicillin-sulbactam (–2.82; 95% CI, –4.59 to –1.03), piperacillin-tazobactam (–9.64; 95% CI, –12.93 to –6.28) and clindamycin (–1.04; 95% CI, –1.6 to –0.47), particularly for high-risk hospitalizations notified by way of AI alert (relative discount for piperacillin-tazobactam, 16.8%; 95% CI, 8%-24.6%).
“We measured a discount within the incidence of C. difficile infections, but it surely was not statistically vital,” Wiens stated. “Nonetheless, we did discover a statistically vital discount in the usage of antimicrobials that may enhance the chance of [C. difficile infections].”
Semi-structured interviews with well being care workers, in addition to area observations, revealed “restricted and doubtlessly insufficient use” of some prevention methods, the researchers wrote.
Particularly, the hand-hygiene effort yielded poor adherence, and of 17 inpatient practitioners interviewed, six stated they ignored the alerts and 10 couldn’t recall the content material delivered. Conversely, six of the seven pharmacists interviewed didn’t take into account the implementation bundle disruptive and all had been attentive to the AI-guided alerts.
Wiens famous that the examine was restricted by the shortcoming to conduct a randomized medical trial in a single-hospital setting, because the major final result was C. difficile an infection.
“The switch of sufferers and workers throughout models basically eradicated the ‘management’ group,” she stated. “As well as, we didn’t measure a statistically vital distinction within the incidence of C. difficile. A examine with an extended remark interval could also be wanted.”
For extra info:
Jenna Wiens, PhD, might be reached at wiensj@umich.edu.
Sources/Disclosures
Collapse
Disclosures:
Wiens reviews no related monetary disclosures. Please see the examine for all different authors’ related monetary disclosures.
