Patients

A total of 666 patients were included in the preintervention period, 1357 in the mandatory OPAT review period and 1208 in the sequential post GN-BSI guideline period. Clinical and demographic characteristics are shown in Table 1. Age, sex, and organism distribution were similar between groups. Differences between time periods were observed in ethnicity, UTI as source of infection, Charlson comorbidity index, and PITT bacteremia score. E coli was the most common pathogen (>70%) and UTI was a source of infection for >70% of GN-BSI in all groups. Ceftriaxone resistance was observed in less than 10% of cases in all time periods.

Table 1. Clinical characteristics by time period

Note. OPAT, outpatient parenteral antibiotic therapy; GN, Gram negative.

The percentage of patients receiving IV antibiotics at discharge was 22.7% during the preintervention period, 10.7% during the OPAT review period, and 9.2% during the postguideline period (P < 0.001) (Table 2). In the ITS sensitivity analysis, implementation of the OPAT review intervention was associated with an initial 6.8% decrease (95% CI [−12.2, −1.4) in discharge IV antibiotics across the system without a significant change in subsequent rate per month (0.27%; 95% CI−0.06, 0.62). No differences were observed in the initial level change (1.25%; 95% CI−0.27%, 5.2%) or subsequent change in rate per month (1.49%; 95% CI−0.19%, 0.31%) associated with dissemination of the GN-BSI guideline (Figure 1). No significant autocorrelation in the model was detected (p-value = 0.94 at 24 lags).

Table 2. Outcomes by time period

Note. OPAT, outpatient parenteral antibiotic therapy; GN, Gram negative.

Figure 1. Interrupted time series analysis of discharge IV therapy at system level.

Note: OPAT, outpatient parenteral antibiotic therapy; GN, Gram negative.

In aggregate, the median length of therapy was 13.5 days (IQR 11, 15) during the preimplementation period, 11.8 days (IQR 9, 14) during the OPAT review period, and 10.7 days (8.6, 13.6) during the GN-BSI guideline period (P < 0.001). The sequential introductions of the mandatory OPAT review and GN-BSI guidelines were not associated with significant estimated changes in level (−0.25 days [95% CI−1.49, 0.98]; 0.03 days [95% CI−0.92, 0.99], respectively) or estimated changes in slope in the ITS sensitivity analysis (0.04 days per month [95% CI−0.04, 0.12]; 0.02 [95% CI−0.04, 0.08], respectively) (Figure 2). No significant autocorrelation in the model was detected (p-value = 0.62 at 24 lags).

Figure 2. Interrupted time series analysis of median length of treatment at system level.

Note: OPAT, outpatient parenteral antibiotic therapy; GN, Gram negative.

The percentage of patients receiving < 8 days of therapy was 8.3%, 17.6%, 20.9% (p < 0.001) in the preintervention, mandatory OPAT review and sequential GN-BSI periods respectively, while the percentage of patients receiving < 10 days of antibiotic therapy in these periods was 19.1%, 37.6%, and 45.0% (p < 0.001).

Changes in discharge oral antibiotic selection were observed between periods (Table 3). Notable observations included increases in tmp/smx (5.1%, 12.5%, and 12.5% in each period, respectively; P < 0.001) and cephalexin (3.3%, 12.5%, 16.1%; p < 0.001) prescriptions. We also observed reductions in fluoroquinolones (49.1%, 45.4%, 42.5%; p 0.02) and cefuroxime prescriptions (8.3%, 2.4%, 1.5%; p < 0.001). Recommended dosing adherence for discharge oral beta-lactam prescriptions was 2.8% in the preimplementation period, 19.5% in the OPAT review period, and 33.5% in the postguideline period (p < 0.001).

Table 3. Discharge antibiotics

Note.

^a Dosing assessed regardless of renal function.

OPAT, outpatient parenteral antibiotic therapy; GN, Gram negative.

No differences between groups were observed in hospital LOS, 14-day readmissions, 14-day ED revisits, 14-day recurrence of bacteremia, 14-day new occurrence of bacteremia, 30-day all-cause mortality, 30-day C difficile testing, or 30-day C difficile infection (Table 2).

Subgroup analyses

There was substantial variation between hospital groups in their use of OPAT and length of therapy at baseline and in response to interventions over time as shown in Figures 3 and 4. These differences appear substantially smaller after the interventions.

Figure 3. Column chart of discharge IV therapy at hospital group level.

Note: OPAT, outpatient parenteral antibiotic therapy; GN, Gram negative; IV, intravenous.

Bars represent the percentage of patients at each hospital that received discharge IV antibiotics.

Figure 4. Boxplot of median length of therapy at hospital group level.

Note: OPAT, outpatient parenteral antibiotic therapy; GN, Gram negative.

Discussion

When new data emerge to change optimal therapy, it may take many years to change established behaviors. ^{Reference Morris, Wooding and Grant14} Here we report the evaluation of two sequential AS interventions for GN-BSI focused on reducing discharge IV treatment and total antibiotic duration in a large healthcare system. We observed a significant decrease in the percentage of patients receiving IV antibiotic therapy at discharge (from 22.7% to 9.2%). The ITS sensitivity analysis suggests that the OPAT review resulted in an immediate decrease in discharge IV antibiotic use. We also observed a significant decrease in the overall median length of antibiotic therapy (from 13.5 to 10.7 days), however the results of our ITS analysis suggest these observations were due to secular trends. Importantly, we did not observe any difference in safety outcomes before or after each intervention implementation. We observed substantial variation between hospitals in baseline rates of IV discharge antibiotics and duration of therapy, as well as in observed response to AS interventions.

Previous studies have suggested that OPAT review by AS/ID providers may have value in decreasing the number of inappropriate courses of OPAT, improving safety, and decreasing costs by identifying patients for whom oral antibiotics are an appropriate alternative. ^{Reference Sharma, Loomis and Brown15,Reference Heintz, Halilovic and Christensen16} Indeed, this was the experience in our system.

Although we observed reductions in IV antibiotic prescribing at discharge and overall length of therapy over the course of our AS initiatives, we are unable to confirm a temporal association between individual interventions and outcomes, outside OPAT review. It is possible that the more passive nature of the interventions targeting duration or the robust AS review of these patients throughout the study reduced the effect sizes of the tested interventions.

Additional opportunities within our system to further reduce total antibiotic duration exist. At the end of our postintervention period, median LOT was still 10 days. Possible reasons for prolonged durations of therapy include prescribers not counting in-hospital antibiotic duration in total antibiotic duration, discharge outpatient prescriptions with set interval durations of therapy for a full course of antibiotics requiring editing to shorten durations, older references recommending 10 days for pyelonephritis when using beta-lactams or TMP/SMX, inadvertent inclusion of complicated cases in our cohort, and lack of awareness of or disagreement with our guideline. ^{Reference Gupta, Hooton and Naber17} Our institutional UTI guideline active during these QI initiatives recommended 10 days of therapy for pyelonephritis when using TMP/SMX and cephalosporins, and may have contributed to this observation. This same concern has been raised by other institutions. ^{Reference DiPietro, Dubrovskaya and Marsh18} We hope the recently published national guideline for complicated UTI will improve adherence to 7-day durations given pyelonephritis is the most common source contributing to GN-BSI. ^{Reference Trautner19}

In AS, proven interventions often have differing results when applied to different settings. Within our large health system, we observed substantial variation across sites in baseline practices and in the observed outcomes after interventions. Deeper understanding of the reasons for variation in the willingness to implement new evidence-based practices and the behavioral strategies that are most effective remain key areas for future research. It is our experience that having a functional AS quality improvement (QI) team alone may be insufficient to drive change. Change readiness and change capacity are important factors that determine QI initiative success. ^{Reference Madu20–Reference Thampi, Szymczak and Leis23} Despite implementing coordinated AS interventions at the system level, we still observed variation in antibiotic prescribing patterns at the local hospital levels. Whether these differences were due to local culture and readiness, differences in engagement of key thought leaders, negotiation skills of the AS teams, or other factors, our observations highlight the importance of considering potential unique local dynamics when implementing QI interventions across a large health system.

Our findings should be considered in light of several limitations. First, findings from our evaluation of an institutional quality improvement initiative cannot be generalized to other settings where AS program structure and resources or clinical practice patterns may be different. Second, we relied on readily available electronic data including ICD10 diagnosis codes to approximate patients with uncomplicated GN-BSI. Due to potential misclassifications, patients with complicated GN-BSI may have been included, which may have had some influence on the observed outcomes in what we intended to be our target population. Third, observations may have likewise been influenced by the use of total duration of antibiotic therapy instead of duration of active antimicrobial therapy. However, given that ceftriaxone is our preferred empiric antimicrobial and ceftriaxone resistance was less than 10% in our cohort, this is unlikely to have changed median length of treatment. Fourth, due to the retrospective design, it is possible that some clinical outcomes may have been misclassified (eg, mortality, readmissions) although we would expect this to be similarly distributed over time. Fifth, extending our preintervention period may have more accurately estimated preintervention trends. Unfortunately, 2018 was the first-year post implementation of a new EHR throughout our health system. We selected 2018 as the beginning of our preintervention period to reduce the possibility of differences between EHRs; most importantly the increased ability to capture outpatient prescriptions in the new EHR.

After implementing two GN-BSI-focused AS initiatives in our large health system, we observed a shift in antibiotic prescribing patterns, to more frequent oral rather than IV antibiotics at discharge, and shorter overall antibiotic durations, without obvious changes in adverse outcomes. Observed prescribing patterns varied by hospital, highlighting the need to identify behavioral and cultural barriers to change.