We examined antimicrobial use and length of stay (LOS) before and after a change in ASP approach at the Hospital of the University of Pennsylvania, a 776-bed tertiary care academic medical center in Philadelphia. Approval was obtained from the University of Pennsylvania Institutional Review Board. All inpatients at least 18 years of age who received at least 1 dose of any oral or intravenous antibacterial or antifungal agent were included. To maintain independency of antimicrobial prescribing practices, only the first admission for each individual was included during the study period.

The hospital ASP was implemented in 1993, using formulary restriction with prior authorization.¹ Prior authorization was provided by infectious diseases–trained clinical pharmacists, infectious diseases fellows, and the infectious diseases consult service, as described previously.¹³ Through May 2009, antimicrobials that were broad spectrum, associated with resistant organisms, associated with serious adverse events, or costly were restricted, requiring prior authorization from the ASP team before dispensing from the pharmacy. This included commonly used broad-spectrum agents such as cefepime and piperacillin/tazobactam as well as antifungals. Vancomycin was available for 72 hours without prior authorization, but further use required ASP approval.

Requests for prior authorization increased over time, many for empiric broad-spectrum antibiotic regimens. To better utilize ASP resources, restriction categories were reassessed. In June 2009, the restriction was lifted for most antimicrobial agents on formulary, including the broad-spectrum agents cefepime, piperacillin/tazobactam, and vancomycin. For these, prospective audit with feedback was also introduced in June 2009. Audits occurred Monday through Friday by 3 infectious diseases–trained clinical pharmacists and included all patients in the hospital with an active order for any of those agents. Suggested modifications, if necessary, were communicated to the covering provider, and acceptance of the recommendations was at their discretion. Prior authorization, conducted in the same manner as before, remained for other broad-spectrum and/or costly agents (including aztreonam, ceftazidime, daptomycin, levofloxacin, linezolid, and meropenem) and for antifungal agents (Figure 1).

For this analysis, the preintervention period contained all patients admitted during the 24 months before the ASP change, June 2007 through May 2009. The postintervention period included admissions during the 24 months after the change, July 2009 through June 2011. June 2009 was excluded from both periods, as this was a transitional time for both prescribers and the ASP.

The primary outcome was antimicrobial doses administered measured by days of therapy (DOT), the preferred metric of antimicrobial use.¹⁴ A single DOT is recorded for each individual antimicrobial administered to a patient on a given day regardless of dose and frequency.¹⁵ A complementary measure of antimicrobial use, length of therapy (LOT), corresponds to each day a patient receives any systemic antimicrobial therapy, regardless of the number of agents or doses.¹⁶ For example, an individual who receives cefepime for 3 days contributes 3 DOT and 3 LOT, whereas a patient receiving both cefepime and vancomycin for 3 days contributes 6 DOT but 3 LOT. Although the primary outcome quantified composite use of all 6 broad-spectrum anti-gram-negative antibiotics on formulary (aztreonam, cefepime, ceftazidime, levofloxacin, piperacillin/tazobactam, and meropenem), subanalyses focused on cefepime and piperacillin/tazobactam, the specific anti-gram-negative agents for which the stewardship approach changed from prior authorization to prospective audit.

Additional end points evaluated included total hospital LOS and LOS after receipt of first dose of systemic antimicrobial, both measured in days for all patients who received at least 1 dose of an antimicrobial. Appropriateness of empiric antibiotic selection for the subset of patients with bloodstream infections (BSIs) caused by gram-negative organisms was also assessed (with the exception of the first year of the preintervention period because microbiology data were not available before May 2008). Empiric antibiotic regimens were evaluated by a blinded infectious diseases–trained clinical pharmacist on the following criteria: (1) the antibiotic(s) were active against the pathogen; (2) active antibiotic(s) were administered within 24 hours of the blood culture; (3) the antibiotic(s) used were of the narrowest spectrum available based on the results of susceptibility testing and the reviewers’ opinion; and (4) in the case of multiple antibiotic agents, redundant or overlapping spectra of activity were minimized.

To account for factors that could impact antimicrobial use and LOS other than the change in ASP strategy, we examined 3 additional antimicrobial groups.^17,18 These included (1) all systemic antibiotics and antifungals except the 3 drugs that were prospectively audited (referred to as nonaudited antimicrobials), to uncover trends in total antimicrobial use during the study period; (2) vancomycin, the most commonly used broad-spectrum anti-gram-positive agent at our institution; and (3) all systemic antifungal drugs, to provide a nonequivalent dependent variable. Nonequivalent dependent variables are outcome variables similar to the primary outcome and subject to the same alternative causes of an observed effect but are not directly affected by the intervention.

All data were obtained from the Penn Data Store (PDS), a centralized data repository of information from all electronic clinical systems across the Penn Medicine Health System. PDS contains demographics, diagnoses, procedures, laboratory results, medication orders, and doses administered for all inpatient and outpatient visits. For this study, demographic, microbiology, medication orders, and administration records were extracted for inpatients admitted during the 4-year study period.

Patient characteristics were compared between the 2 time periods using the χ² or Fisher exact test for categorical variables and the Wilcoxon rank-sum test for continuous variables. Antimicrobial consumption was evaluated on a monthly basis. All included patients were assigned a month (1–49) according to admission date. DOT, LOT, and LOS for each patient were calculated, totaled for each month, and then standardized to 1,000 patient-days (DOT/1,000-PD, LOT/1,000-PD, and days/1,000-PD, respectively), using total patient-days for all admissions in that month irrespective of antimicrobial administration.¹⁵

Because antimicrobial consumption for a given month is likely related to that in previous months, we used interrupted time series analysis.^17,19 The generalized autoregressive conditional heteroskedasticity (GARCH) model was fitted for all DOT and LOS measures. GARCH allows for the assessment of autocorrelation, volatility, and the influence of an exogenous input, such as the change in ASP method.^20,21

For the treatment of gram-negative BSIs, the frequency of adequate empiric therapy, redundant spectra of activity, and excessively broad therapy were compared in the preintervention and postintervention groups using the χ² test. All analyses were performed with STATA statistical software (ver. 12.1; StataCorp, College Station, TX).

We identified 55,336 patients who received antimicrobial therapy from June 2007 through June 2011: 29,660 patients before the intervention and 25,676 after the intervention. Patients in both groups were similar with regard to baseline demographics collected at admission (Table 1).

During the preintervention period, use of broad-spectrum anti-gram-negative antibiotics was declining at a rate of −4.00 DOT/1,000-PD per month. However, during the postintervention period, use increased by 0.80 DOT/1,000-PD per month, indicating that the change in ASP was associated with a slope change of 4.80 DOT/1,000-PD per month (P < .001; Table 2, Figure 2), or an increase of 0.49% per month after the intervention. Similarly, after decreasing during the 2 years before the ASP change, use of cefepime and piperacillin/tazobactam significantly increased following the transition to prospective audit with feedback by 3.21 DOT/1,000-PD per month (P = .003; Table 2, Figure 2). This represented a 1.1% increase in use per month through the postintervention period. The use of the other 4 broad-spectrum anti-gram-negative antibiotics, for which prior authorization was required in both time periods, decreased before the intervention, with no significant increase after the intervention (slope change, 1.62 DOT/1,000-PD per month; P = .100; Table 2, Figure 2).

Overall use of all systemic antimicrobial agents significantly increased after the change in ASP method (P < .001; Table 2, Figure 3). Vancomycin use declined before the intervention but significantly increased after the intervention (P = .005; Table 2, Figure 3). While use of nonaudited antimicrobials also significantly increased after the change in ASP methods (P < .001), the slope during the postintervention period continued to decline at −1.87 DOT/1,000-PD per month (Table 2, Figure 3). The LOT of all systemic antimicrobials declined before the intervention by −2.30 LOT/1,000-PD per month, and, despite a significant increase in slope (P = .029), use continued to decrease after the intervention by −0.33 LOT/1,000-PD per month.

Antifungal agents, the nonequivalent dependent control variable, followed similar patterns of use as the primary variable. Before the intervention, use of these drugs decreased by −1.77 DOT/1,000-PD per month. After the intervention, use increased by 0.65 DOT/1,000-PD per month, representing a slope change of 2.42 DOT/1,000-PD per month (P = .001; Table 2, Figure 3).

Before the intervention, total hospital LOS for patients receiving antimicrobials was decreasing at a rate of −1.57 days/1,000-PD per month; however, LOS following the ASP transition increased by 1.94 days/1,000-PD (P = .016). When limiting the analyses to LOS after the first dose of a systemic antimicrobial, we found a similar reversal of trends (−1.42 days/1,000-PD per month before the intervention; +2.30 days/1,000-PD after the intervention; P = .004; Figure 4).

A total of 488 patients with gram-negative BSIs were identified from the available microbiology data. There were 165 BSI episodes available in 13 months of the preintervention period and 311 in 24 months of the postintervention period. There was no significant difference in the proportion of inappropriate empiric antibiotic regimen selection between the 2 groups (18.2% vs 18.0%; P = .962). Similarly, there were no significant differences in the proportion of redundant antibiotic regimen (11.5% vs 10.3%; P = .681) and in excessively broad antibiotic coverage (42.4% vs 37.3%; P = .275) before and after the intervention.