A clinical trial comparing physician prompting with an unprompted automated electronic checklist to reduce empirical...

To determine whether face-to-face prompting of critical care physicians reduces empirical antibiotic utilization compared to an unprompted electronic checklist embedded within the electronic health record.

Random allocation design.

Medical ICU with high-intensity intensivist coverage at a tertiary care urban medical center.

Two hundred ninety-six critically ill patients treated with at least 1 day of empirical antibiotics.

For one medical ICU team, face-to-face prompting of critical care physicians if they did not address empirical antibiotic utilization during a patient's daily rounds. On a separate medical ICU team, attendings and fellows were trained once to complete an electronic health record-embedded checklist daily for each patient, including a question asking whether listed empirical antibiotics could be discontinued.

Prompting led to a more than four-fold increase in discontinuing or narrowing of empirical antibiotics compared to use of the electronic checklist. Prompted group patients had a lower proportion of patient-days on which empirical antibiotics were administered compared to electronic checklist group patients (63.1% vs 70.0%, p = 0.002). Mean proportion of antibiotic-days on which empirical antibiotics were used was also lower in the prompted group, although not statistically significant (0.78 [0.27] vs 0.83 [0.27], p = 0.093). Each additional day of empirical antibiotics predicted higher risk-adjusted mortality (odds ratio, 1.14; 95% CI, 1.05-1.23). Risk-adjusted ICU length of stay and hospital mortality were not significantly different between the two groups.

Face-to-face prompting was superior to an unprompted electronic health record-based checklist at reducing empirical antibiotic utilization. Sustained culture change may have contributed to the electronic checklist having similar empirical antibiotic utilization to a prompted group in the same medical ICU 2 years prior. Future studies should investigate the integration of an automated prompting mechanism with a more generalizable electronic health record-based checklist.