Methods

We conducted a mixed methods study. The study took place at a single academic hospital with on-site ID consultation. A multidisciplinary team of ID physicians, ID pharmacists, and information system specialists used a conceptual model of how clinicians interact with EHR to design SABER (Supplemental Figure S1). ^{Reference Sittig and Singh9} The model influenced three features: 1) minimal firing, 2) clear, concise language, and 3) embedment of the consult order. In particular, a 36-hour delay between positive cultures and firing promoted clinician autonomy and patient-specific decision-making while avoiding alert fatigue. When it went live, clinicians were not specifically informed about the alert, and the antimicrobial stewardship team ceased manual monitoring and intervention for positive S. aureus blood cultures.

We conducted a pre-post quantitative study to determine whether time to ID consultation for SAB differed after implementing SABER. The pre- and post-intervention phases ran from November 2018 to October 2019 and February 2020 to February 2021, respectively. Hospitalized, adult patients were included in our analysis if they had a blood culture drawn at our facility positive for S. aureus during the pre- or post-intervention phase. Patients were excluded if they had an ID consult prior to blood culture positivity, died within 48 hours of a blood culture positivity, or transferred to palliative care within 48 hours of admission. All data was collected via chart abstraction.

Our independent variable was whether patients received care in the pre- or post-intervention phase. Our primary outcome was time to ID consultation, defined as the time from initial blood culture positivity to signed ID note, when recommendations were available to improve care. Table 1 lists secondary outcome variables—care metrics and clinical outcomes—and covariates, all of which were included in the multivariate modeling due to their clinical relevance.

Table 1.

Patient characteristics, quality of care indicators, and patient outcomes before and after implementing Staphylococcus aureus Bacteremia Electronic Reminder

Data are presented as n (%) unless otherwise indicated.

Abbreviations: ICU, intensive care unit; IVDU, intravenous drug use; MRSA, methicillin resistant S. aureus; SAB, S. aureus bacteremia.

For statistical analyses, we constructed Kaplan-Meier curves depicting time to ID consultation. Marginal differences were compared using the log-rank test. We built a multivariate Cox proportional hazard model comparing time to ID consultation in the pre- and post-intervention groups. Differences in secondary outcomes between the pre- and post-intervention groups were assessed with one-way ANOVA or t-tests. We used Fischer’s exact test to estimate differences in in-hospital mortality.

We assessed clinician acceptance of SABER using a conceptual model of how clinicians interact with the EHR. ^{Reference Sittig and Singh9} We purposively recruited primary team members with different roles for whom the alert fired. We attempted to recruit clinicians who opted not to consult but were unsuccessful. Each participant verbally consented and completed a 45-minute, one-on-one, audio-recorded interview. The interviewer (J.P.) was not a clinician, minimizing concerns about the interview’s influence on subsequent clinical collaborations. Interviews were conducted using a semi-structured guide rooted in the conceptual model. Participants received $100 for participation. Audio recordings were transcribed verbatim and de-identified before rapid qualitative inquiry. ^{Reference Keating, McKinley and Dolan10} The seasoned analyst (J.K.) debriefed with the interviewer to discuss emergent themes and quotes, enhancing credibility. The University of Wisconsin Health Sciences Institutional Review Board exempted this study (#2019–0888 and #2019–1467).