Discussion

We report a 25% immediate reduction in blood-culture contamination rates after ISDD implementation. This reduction persisted, and was further improved upon, during 20 months of follow-up data. Given the size of our study (>190,000 blood cultures) and the duration of follow-up, these findings provide strong supportive evidence that implementation of an ISDD sustainably reduces blood culture contamination rates.

Interestingly, we observed a transient increase in inpatient vancomycin use immediately after the ISDD implementation despite a decrease in the blood-culture contamination rate. This increase was not sustained, and the observed trend change in vancomycin use was small (<1% per month), suggesting a small gradual effect of the intervention on total inpatient vancomycin use. We interpret this finding as showing that blood-culture contamination is a minor driver of total inpatient vancomycin use and that an immediate level change in vancomycin use in response to ISDD implementation is unlikely. A more accurate metric would be vancomycin use in the first 3 days of hospitalization; however, we did not have access to these data. Future studies of ISDDs should evaluate their impact on other outcomes such as length of stay and unnecessary testing (eg, repeat blood cultures).

In contrast to our study, Nielson et al ^{Reference Nielsen, Nguyen and Wahl2} reported a significant decrease in vancomycin DOT following ISDD implementation. These researchers analyzed the impact of 2 interventions on vancomycin DOT. The first intervention used nucleic acid amplification testing (NAAT)–based microbial identification for blood cultures, and the second intervention, instituted 8 months later, used ISDD implementation. These researchers reported no significant effect of NAAT-based identification on vancomycin DOT, but they did find a significant effect with ISDD implementation. However, ITS analysis was not used, and an alternate hypothesis is that NAAT-based identification required a learning period before the effects on vancomycin usage were realized.

One strength of the ITS design is the ability to model pre- and postintervention trends, which increases confidence in estimating an intervention’s effect. ^{Reference Kontopantelis, Doran, Springate, Buchan and Reeves8} This methodology is well suited to before-and-after quasi-experimental studies, which are commonly used to evaluate quality improvement initiatives. Additional strengths include the large number of cultures, inclusion of a control group, and the extended postintervention period. The latter allowed us to demonstrate that the reduction in contamination rates associated with ISDD implementation is sustainable beyond a “rollout” period.

This study had several limitations. We were unable to track how often the ISDD was used in the intervention hospital’s ED, which may have led to the underestimation of the effect of the intervention. At both hospitals, feedback on contamination rates was given to unit leadership and individual providers, but we were unable to measure or to control for how this informed action. Additionally, the changing patient case mix could have contributed to the observed differences in contamination rates or the lack of observed difference in vancomycin usage. However, the use of a nearby control hospital with a similar patient and provider population helped mitigate this limitation.

In conclusion, we report the implementation of an ISDD associated with a substantial, sustainable decrease in blood-culture contamination rates but without a major effect on vancomycin DOT. ISDDs have the potential to improve patient care by reducing unnecessary downstream testing that results from blood-culture contamination. Further research is needed to couple this intervention with antibiotic stewardship, to assess the impact on patient-centered outcomes (eg, length of stay), and to quantify the potential financial benefits to facilitate wider adoption of these devices.