Abstract
The weighted incidence syndromic combination antibiogram (WISCA) is an antimicrobial stewardship tool that utilizes electronic medical record data to provide real-time clinical decision support regarding empiric antibiotic prescription in the hospital setting. The aim of this study was to determine the impact of WISCA utilization for empiric antibiotic prescription on hospital length of stay (LOS).
We performed a crossover randomized controlled trial of the WISCA tool at 4 hospitals. Study participants included adult inpatients receiving empiric antibiotics for urinary tract infection (UTI), abdominal-biliary infection (ABI), pneumonia, or nonpurulent cellulitis. Antimicrobial stewardship (ASP) physicians utilized WISCA and clinical guidelines to provide empiric antibiotic recommendations. The primary outcome was LOS. Secondary outcomes included 30-day mortality, 30-day readmission, Clostridioides difficile infection, acquisition of multidrug-resistant gram-negative organism (MDRO), and antibiotics costs.
In total, 6849 participants enrolled in the study. There were no overall differences in outcomes among the intervention versus control groups. Participants with cellulitis in the intervention group had significantly shorter mean LOS compared to participants with cellulitis in the control group (coefficient estimate = 0.53 [−0.97, −0.09], P = .0186). For patients with community acquired pneumonia (CAP), the intervention group had significantly lower odds of 30-day mortality compared to the control group (adjusted odds ratio [aOR] .58, 95% confidence interval [CI], .396, .854, P = .02).
Use of WISCA was not associated with improved outcomes for UTI and ABI. Guidelines-based interventions were associated with decreased LOS for cellulitis and decreased mortality for CAP.
In the United States, more than 2.8 million antibiotic-resistant infections and 35 000 deaths due to antibiotic-resistant infections occur annually [1]. Key drivers of antibiotic resistance are the overuse of broad spectrum antibiotics and unnecessary antibiotics (eg, use of antibiotics to treat noninfectious syndromes or colonizing microorganisms) [1–4]. In total, 75% of hospital days involve antimicrobial therapy [5], and an estimated 30% of these antimicrobials are unnecessary [6–8]. Healthcare systems must work to reduce unnecessary and overly broad spectrum antibiotic use in the hospital setting to help combat the rise in antimicrobial resistance and Clostridioides difficile infection (CDI).
With the dramatic increase in electronic health record (EHR) utilization by US healthcare systems [9], EHR-based clinical decision support (CDS) systems are promising tools to curtail unnecessary antibiotic use. CDS systems are computer applications that provide guidance for clinicians in making diagnostic and therapeutic decisions for patient care. Electronic CDS tools have been successfully deployed in the inpatient setting for a variety of diseases, including diabetes, acute kidney injury, and heart disease [10–12]. The Infectious Diseases Society of America recommends that computerized CDS tools be incorporated into Antibiotic Stewardship Programs [13], and a number of antibiotic stewardship CDS tools have been developed and utilized with varying levels of success [14].
We previously created an inpatient antibiotic stewardship tool called the weighted incidence syndromic combination antibiogram (WISCA) [15]. WISCA utilizes EHR data to provide real-time CDS regarding empiric antibiotic prescription in the hospital setting by determining the likelihood that a given antibiotic regimen will cover all the organisms causing a patient’s infectious syndrome based on that patient’s unique personal characteristics. WISCA provides personalized antibiotic recommendations using EHR data available at the time of antibiotic prescription, including patients’ demographics, comorbidities, prior microbiologic culture results, previous antibiotics, and healthcare exposure.
The WISCA tool has previously been shown to increase the likelihood of antibiotic coverage [16], have the potential to reduce time to effective antibiotic coverage [17], and to identify narrower antibiotic choices than chosen by prescribing physicians [18, 19]. The aim of this study was to investigate the impact of WISCA tool utilization during active antimicrobial stewardship surveillance on patient outcomes of hospital length of stay (LOS), mortality, readmission, adverse events, and costs, via a randomized controlled trial.
METHODS
Development of WISCA Tool
The WISCA tool was developed for 6 infectious syndromes: urinary tract infection (UTI), abdominal biliary infection (ABI), nonpurulent cellulitis, community acquired pneumonia (CAP), aspiration pneumonia, and nursing home-associated pneumonia. Detailed information on the development of the WISCA tool for UTI and ABI has been previously described [15]. Briefly, microbiological data were collected for all adult patients diagnosed with UTI or ABI during an inpatient hospitalization at NorthShore University HealthSystem between 1 January 2006 and 31 August 2014. We selected 18 antibiotic regimen combinations for UTI and 22 antibiotic combinations for ABI (Supplementary Appendix Table 1). Logistic regression models were created to model the likelihood that each given antibiotic regimen would “cover” all organisms isolated from cultures for each clinical syndrome (ie, UTI or ABI). Variables included in the models as predictors included age, sex, healthcare location, history of methicillin-resistant Staphylococcus aureus (MRSA) or vancomycin-resistant enterococci (VRE), prior antibiotic exposure, previous healthcare encounters, nursing home residence, and comorbidities. For each antibiotic regimen, the WISCA tool also displayed the relative cost, how “broad spectrum” the coverage was, and the simplicity of the regimen.
Because nonpurulent cellulitis and pneumonia do not usually have associated microbiologic cultures to determine the causative pathogen, we developed first-line recommendations for antibiotic regimens based on expert consensus from a group of infectious disease (ID) physicians and available Infectious Diseases Society of America (IDSA) guidelines [20–22]. For cellulitis, cefazolin was the first-choice recommendation. For pneumonia, recommendations were as follows: levofloxacin or ceftriaxone plus azithromycin for CAP, ampicillin-sulbactam for aspiration pneumonia, and vancomycin plus ceftazidime for nursing home-associated pneumonia.
Overview of Clinical Trial
To investigate the impact of WISCA on patient outcomes, we implemented a randomized controlled trial with crossover design at four NorthShore hospitals from 1 July 2015 through 30 June 2018. The intervention was randomized at the hospital level, with 2 hospitals assigned to the intervention group and 2 hospitals assigned to the control group for the initial 18 months; hospital assignments were switched for the subsequent 18 months.
The primary outcome of the clinical trial was hospital LOS. Secondary outcomes included 30-day readmission, 30-day mortality, antibiotic charges, CDI within 180 days, and acquisition of a multidrug-resistant gram-negative organism (MDRO) within 180 days. To detect a difference in mean LOS of 0.1 day, the study required a sample size of 1570 patients in each of the intervention and control groups to have 80% power to detect a difference with α = 0.05 (σ = 1).
Patients were included in the study if they were ≥18 years old and presented with 1 of 6 clinical syndromes: (1) ABI, (2) UTI, (3) nonpurulent cellulitis, (4) CAP, (5) aspiration pneumonia, (6) nursing home-associated pneumonia (defined as pneumonia in a patient who had been admitted from a nursing home or other long-term healthcare facility within the 48 hours prior to diagnosis). Throughout the course of the study, an Antimicrobial Stewardship Program (ASP) physician reviewed the charts of all patients started on antibiotics within 24 hours at study hospitals to determine if they met one of the clinical syndromes. The review process was facilitated by a real-time EHR-based report of all antibiotics started among inpatients. Patients were excluded from the study if they (1) had an unclear focus of infection, (2) had multiple suspected sources of infection, (3) did not have a bacterial infection based on the ASP physician’s chart review, (4) were receiving pathogen-directed rather than empiric antibiotic therapy (ie, the pathogen causing the infection had already been identified by a microbiologic test), or (5) had an ID physician consulting on their case.
For patients with ABI or UTI, the ASP physician utilized WISCA to determine the most appropriate antibiotic regimen for the patient. This determination took into account several factors embedded in the WISCA tool including (1) percentage likelihood of coverage; (2) breadth of coverage spectrum with more “narrow spectrum” preferred; (3) cost of antibiotics; and (4) ease of administration of antibiotic (eg, 1 medication preferred over several and lower frequency of dosing preferred). When considering percentage likelihood of coverage, a minimum of 75% likelihood of coverage was required, with >90% likelihood of coverage for patients who were immunosuppressed or in the intensive care unit. Patient allergies and comorbidities were also taken into account when selecting regimens. If the ASP physician could not make a recommendation based on chart review alone, they communicated with the primary provider to gain more information about the patient and/or suggested a formal ID consult.
For cellulitis and pneumonia, ASP physicians recommended the guideline-based regimens described above. Although ASP physicians generally adhered to the guidelines, the recommended regimen was subject to the discretion of the ASP physician, taking into account particular patient characteristics, allergies, comorbidities, and contraindications (eg, fluoroquinolones were not recommended for patients with prolonged QT interval).
Intervention and Control Groups
For patients in the intervention group, the ASP physician performed prospective audit and feedback (within 24 hours of antibiotic start) via page or phone call to the primary provider and via written documentation in the EHR. For patients in the control group, the ASP physician recorded the recommended antibiotic in the study database but did not communicate the recommendation to the patient’s provider unless they believed the regimen being used represented a threat to the patient’s health.
Data Analysis
For patients enrolled in the study, we collected demographics, clinical information, and data regarding the primary outcome of hospital LOS as well as secondary outcomes of mortality, 30-day readmission, CDI, acquisition of MDRO, and antibiotic costs. For patients with multiple hospital admissions with infectious syndromes of interest during the study period, only data from the first admission were included in the analysis.
For categorical variables, χ 2 testing was performed and for continuous variables, comparisons were made using Student t test. We utilized a multivariable linear regression model to assess the effect of the intervention on the primary outcome of length of stay, controlling for baseline patient characteristics (ie, age, body mass index [BMI], sex, patient type, Charlson score, and Apache II score) as well as cluster and period. The coefficient estimates derived from the adjusted linear regression model were used to assess the impact of being in the intervention group on the length of stay. Similarly, we derived a multivariable logistic regression model to assess the effect of the intervention on the secondary outcome of 30-day mortality, adjusting for baseline patient characteristics, cluster, and period.
We performed subanalyses assessing the impact of being in the intervention group on outcomes stratified by infectious syndrome. We also compared outcomes among patients for whom the ASP physician recommended a change in the antibiotic regimen in the intervention group (in which the recommendation to change antibiotics was communicated to the primary team) vs. control group (in which the recommendation to change antibiotics was documented in the study database but not communicated to the primary team).
For patients in the intervention group, we measured the percentage of time that recommendations from the ASP physician were followed. Recommendations were considered followed if the patient was prescribed the recommended or a very similar antibiotic regimen within 24 hours of the recommendation. We compared outcomes for patients in whom recommendations were followed vs. patients for whom recommendations were not followed.
Statistical analyses were performed using R. This study was approved by the NorthShore University HealthSystem Institutional Review Board (EH13-3168).
RESULTS
Nearly 80 000 hospitalized patients were initiated on antibiotics during the 3 years of study, and a total of 6849 patients with the defined infectious syndromes were enrolled in the study (Figure 1). Of these, 32.33% (2214/6849) had ABI, 26.33% (1803/6849) had UTI, 24.88% (1704/6849) had CAP, 7.11% (487/6849) had cellulitis, 5.93% (406/6849) had aspiration pneumonia, and 3.43% (235/6849) had nursing home-associated pneumonia. Table 1 displays baseline demographics of study participants. Participants in the intervention group were older (mean age 71.06 vs 69.62; P = .0016), less likely to be Black (4.24% vs 5.52%; P = .0163), more likely to have Medicare insurance (60.28% vs 56.46%; P = .0014), and less likely to be observation status (14.31% vs 16.93%; P = .0032) than those in the control group. Otherwise, there were no significant differences in baseline characteristics between intervention and control groups.
Characteristics Among Control Versus Intervention Groups
| Control N = 3314 | Intervention N = 3535 | P-value | |
|---|---|---|---|
| Mean (SD) or N (%) | Mean (SD) or N (%) | ||
| Patient characteristics | |||
| Patient age (years) | 69.6 (19.1) | 71.7 (18.6) | .0016 |
| Female sex | 1828 (55.16%) | 1980 (56.01%) | .4938 |
| BMI | 27.63 (6.82) | 27.68 (7.01) | .7612 |
| APACHE II Score | 11.69 (5.56) | 11.95 (5.47) | .0614 |
| Charlson comorbidity index | 1.28 (1.67) | 1.30 (1.67) | .7062 |
| Ethnicity | |||
| Hispanic/Latino | 176 (5.31%) | 163 (4.61%) | .2011 |
| Non-Hispanic | 3130 (94.45%) | 3366 (95.22%) | .1650 |
| Unknown | 8 (0.24%) | 6 (0.17%) | .6976 |
| Race | |||
| Black | 183 (5.52%) | 150 (4.24%) | .0163 |
| Asian | 145 (4.38%) | 158 (4.47%) | .8960 |
| White | 2451 (73.96%) | 2636 (74.57%) | .5829 |
| Other or unknown | 535 (16.14%) | 591 (16.72%) | .5426 |
| Insurance | |||
| Medicaid | 243 (7.33%) | 251 (7.10%) | .7457 |
| Medicare or Medicare Advantage | 1871 (56.46%) | 2131 (60.28%) | .0014 |
| Private | 1154 (34.82%) | 1117 (31.60%) | .0050 |
| Self pay or other | 46 (1.39%) | 36 (1.02%) | .1955 |
| Hospitalization | |||
| Observation status | 561 (16.93%) | 506 (14.31%) | .0032 |
| ICU admission | 439 (13.25%) | 441 (12.48%) | .3589 |
| Control N = 3314 | Intervention N = 3535 | P-value | |
|---|---|---|---|
| Mean (SD) or N (%) | Mean (SD) or N (%) | ||
| Patient characteristics | |||
| Patient age (years) | 69.6 (19.1) | 71.7 (18.6) | .0016 |
| Female sex | 1828 (55.16%) | 1980 (56.01%) | .4938 |
| BMI | 27.63 (6.82) | 27.68 (7.01) | .7612 |
| APACHE II Score | 11.69 (5.56) | 11.95 (5.47) | .0614 |
| Charlson comorbidity index | 1.28 (1.67) | 1.30 (1.67) | .7062 |
| Ethnicity | |||
| Hispanic/Latino | 176 (5.31%) | 163 (4.61%) | .2011 |
| Non-Hispanic | 3130 (94.45%) | 3366 (95.22%) | .1650 |
| Unknown | 8 (0.24%) | 6 (0.17%) | .6976 |
| Race | |||
| Black | 183 (5.52%) | 150 (4.24%) | .0163 |
| Asian | 145 (4.38%) | 158 (4.47%) | .8960 |
| White | 2451 (73.96%) | 2636 (74.57%) | .5829 |
| Other or unknown | 535 (16.14%) | 591 (16.72%) | .5426 |
| Insurance | |||
| Medicaid | 243 (7.33%) | 251 (7.10%) | .7457 |
| Medicare or Medicare Advantage | 1871 (56.46%) | 2131 (60.28%) | .0014 |
| Private | 1154 (34.82%) | 1117 (31.60%) | .0050 |
| Self pay or other | 46 (1.39%) | 36 (1.02%) | .1955 |
| Hospitalization | |||
| Observation status | 561 (16.93%) | 506 (14.31%) | .0032 |
| ICU admission | 439 (13.25%) | 441 (12.48%) | .3589 |
Abbreviations: BMI, body mass index; ICU, intensive care unit; SD, standard deviation.
Characteristics Among Control Versus Intervention Groups
| Control N = 3314 | Intervention N = 3535 | P-value | |
|---|---|---|---|
| Mean (SD) or N (%) | Mean (SD) or N (%) | ||
| Patient characteristics | |||
| Patient age (years) | 69.6 (19.1) | 71.7 (18.6) | .0016 |
| Female sex | 1828 (55.16%) | 1980 (56.01%) | .4938 |
| BMI | 27.63 (6.82) | 27.68 (7.01) | .7612 |
| APACHE II Score | 11.69 (5.56) | 11.95 (5.47) | .0614 |
| Charlson comorbidity index | 1.28 (1.67) | 1.30 (1.67) | .7062 |
| Ethnicity | |||
| Hispanic/Latino | 176 (5.31%) | 163 (4.61%) | .2011 |
| Non-Hispanic | 3130 (94.45%) | 3366 (95.22%) | .1650 |
| Unknown | 8 (0.24%) | 6 (0.17%) | .6976 |
| Race | |||
| Black | 183 (5.52%) | 150 (4.24%) | .0163 |
| Asian | 145 (4.38%) | 158 (4.47%) | .8960 |
| White | 2451 (73.96%) | 2636 (74.57%) | .5829 |
| Other or unknown | 535 (16.14%) | 591 (16.72%) | .5426 |
| Insurance | |||
| Medicaid | 243 (7.33%) | 251 (7.10%) | .7457 |
| Medicare or Medicare Advantage | 1871 (56.46%) | 2131 (60.28%) | .0014 |
| Private | 1154 (34.82%) | 1117 (31.60%) | .0050 |
| Self pay or other | 46 (1.39%) | 36 (1.02%) | .1955 |
| Hospitalization | |||
| Observation status | 561 (16.93%) | 506 (14.31%) | .0032 |
| ICU admission | 439 (13.25%) | 441 (12.48%) | .3589 |
| Control N = 3314 | Intervention N = 3535 | P-value | |
|---|---|---|---|
| Mean (SD) or N (%) | Mean (SD) or N (%) | ||
| Patient characteristics | |||
| Patient age (years) | 69.6 (19.1) | 71.7 (18.6) | .0016 |
| Female sex | 1828 (55.16%) | 1980 (56.01%) | .4938 |
| BMI | 27.63 (6.82) | 27.68 (7.01) | .7612 |
| APACHE II Score | 11.69 (5.56) | 11.95 (5.47) | .0614 |
| Charlson comorbidity index | 1.28 (1.67) | 1.30 (1.67) | .7062 |
| Ethnicity | |||
| Hispanic/Latino | 176 (5.31%) | 163 (4.61%) | .2011 |
| Non-Hispanic | 3130 (94.45%) | 3366 (95.22%) | .1650 |
| Unknown | 8 (0.24%) | 6 (0.17%) | .6976 |
| Race | |||
| Black | 183 (5.52%) | 150 (4.24%) | .0163 |
| Asian | 145 (4.38%) | 158 (4.47%) | .8960 |
| White | 2451 (73.96%) | 2636 (74.57%) | .5829 |
| Other or unknown | 535 (16.14%) | 591 (16.72%) | .5426 |
| Insurance | |||
| Medicaid | 243 (7.33%) | 251 (7.10%) | .7457 |
| Medicare or Medicare Advantage | 1871 (56.46%) | 2131 (60.28%) | .0014 |
| Private | 1154 (34.82%) | 1117 (31.60%) | .0050 |
| Self pay or other | 46 (1.39%) | 36 (1.02%) | .1955 |
| Hospitalization | |||
| Observation status | 561 (16.93%) | 506 (14.31%) | .0032 |
| ICU admission | 439 (13.25%) | 441 (12.48%) | .3589 |
Abbreviations: BMI, body mass index; ICU, intensive care unit; SD, standard deviation.
Participant flow diagram. Abbreviations: ABI, abdominal-biliary infection; ID, infectious diseases; UTI, urinary tract infection; WISCA, weighted incidence syndromic combination antibiogram.
Among the intervention group, the ASP physician agreed with the initial empiric antibiotic regimen in 63.1% (2232/3535) of cases, recommended a change in antibiotics in 31.2% (1240/3535) of cases, and recommended an ID consult (ie, unable to make a recommendation based on chart review alone) in 1.4% (50/3535) of cases. For 4.2% (149/3535) of cases, data were missing. The most frequently recommended antibiotic regimens for UTI were ceftriaxone (50%) and piperacillin-tazobactam (17%). The most frequently recommended antibiotic regimens for ABI were ampicillin-sulbactam (41%) and piperacillin-tazobactam (35%). For all syndromes, the most commonly prescribed antibiotics in the intervention group were piperacillin-tazobactam (14%), ampicillin-sulbactam (9%), and levofloxacin (9%). The most commonly prescribed antibiotic regimens in the control group were ceftriaxone (13%), ampicillin-sulbactam (13%), and piperacillin-tazobactam (11%).
There were no overall differences in the primary outcome of LOS or in secondary outcomes of 30-day mortality, 30-day readmission rate, CDI, MDRO acquisition, or antibiotic charges among those in the intervention versus control groups (Table 2). In the multivariable linear regression model for LOS, the intervention was not associated with a significant difference in LOS (coefficient estimate = −0.056, 95% confidence interval [CI]. −.254, .142, P = .5803). The multivariable logistic regression model showed no difference in 30-day mortality between the intervention and control groups (adjusted odds ratio [aOR] 0.998, 95% CI, .821, 1.214, P = .9864).
Outcomes Among Control Versus Intervention Groups
| Control N = 3314 | Intervention N = 3535 | P- value | |
|---|---|---|---|
| Mean (SD) or N (%) | Mean (SD) or N (%) | ||
| Length of stay (days) | 4.54 (4.42) | 4.50 (4.39) | .6899 |
| 30-day mortality | 178 (5.37%) | 194 (5.49%) | .8730 |
| 30-day readmission | 344 (10.38%) | 374 (10.58%) | .8180 |
| Antibiotic charges (dollars) | 546.75 (607.92) | 548.72 (604.76) | .8931 |
| C. difficile infection within 180 days | 151 (4.56%) | 165 (4.67%) | .8717 |
| New onset MDRO within 180 days | 55 (1.66%) | 52 (1.47%) | .5950 |
| Control N = 3314 | Intervention N = 3535 | P- value | |
|---|---|---|---|
| Mean (SD) or N (%) | Mean (SD) or N (%) | ||
| Length of stay (days) | 4.54 (4.42) | 4.50 (4.39) | .6899 |
| 30-day mortality | 178 (5.37%) | 194 (5.49%) | .8730 |
| 30-day readmission | 344 (10.38%) | 374 (10.58%) | .8180 |
| Antibiotic charges (dollars) | 546.75 (607.92) | 548.72 (604.76) | .8931 |
| C. difficile infection within 180 days | 151 (4.56%) | 165 (4.67%) | .8717 |
| New onset MDRO within 180 days | 55 (1.66%) | 52 (1.47%) | .5950 |
Abbreviations: MDRO, multidrug-resistant gram-negative organism; SD, standard deviation.
Outcomes Among Control Versus Intervention Groups
| Control N = 3314 | Intervention N = 3535 | P- value | |
|---|---|---|---|
| Mean (SD) or N (%) | Mean (SD) or N (%) | ||
| Length of stay (days) | 4.54 (4.42) | 4.50 (4.39) | .6899 |
| 30-day mortality | 178 (5.37%) | 194 (5.49%) | .8730 |
| 30-day readmission | 344 (10.38%) | 374 (10.58%) | .8180 |
| Antibiotic charges (dollars) | 546.75 (607.92) | 548.72 (604.76) | .8931 |
| C. difficile infection within 180 days | 151 (4.56%) | 165 (4.67%) | .8717 |
| New onset MDRO within 180 days | 55 (1.66%) | 52 (1.47%) | .5950 |
| Control N = 3314 | Intervention N = 3535 | P- value | |
|---|---|---|---|
| Mean (SD) or N (%) | Mean (SD) or N (%) | ||
| Length of stay (days) | 4.54 (4.42) | 4.50 (4.39) | .6899 |
| 30-day mortality | 178 (5.37%) | 194 (5.49%) | .8730 |
| 30-day readmission | 344 (10.38%) | 374 (10.58%) | .8180 |
| Antibiotic charges (dollars) | 546.75 (607.92) | 548.72 (604.76) | .8931 |
| C. difficile infection within 180 days | 151 (4.56%) | 165 (4.67%) | .8717 |
| New onset MDRO within 180 days | 55 (1.66%) | 52 (1.47%) | .5950 |
Abbreviations: MDRO, multidrug-resistant gram-negative organism; SD, standard deviation.
We performed a subgroup analysis among participants with hypotension (defined as systolic blood pressure <90 mmHg) in the 24 hours prior to enrollment in the study because early effective antibiotic therapy may be more likely to be associated with improved outcomes in this group of patients. However, there were no differences in outcomes of LOS (coefficient estimate = 0.28, 95% CI, −.55, 1.11, P = .506) or 30-day mortality (aOR 1.15, 95% CI, .77, 1.71, P = .5621), among intervention vs. control patients with hypotension. Among the subgroup of patients for whom the ASP physician recommended a change in the antibiotic regimen, there were also no significant differences in the intervention vs. control groups for LOS (coefficient estimate = 0.237, 95% CI, −.075, .55, P = .137) or 30-day mortality (aOR 1.10, 95% CI, .79, 1.50, P = .632).
Among intervention group patients where a recommendation was made to change antibiotics, the recommendation was followed 53.8% (667/1240) of the time. There were no significant differences in LOS or 30-day mortality among intervention participants in which the recommendation to change antibiotics was followed versus not followed (data not shown).
In the subgroup analyses within each infectious syndrome, several statistically significant differences in outcomes among intervention versus control groups were found. In the multivariable model for the outcome of length of stay, patients with cellulitis in the intervention group had significantly shorter mean length of stay compared to participants with cellulitis in the control group (coefficient estimate = 0.53, 95% CI, −.97, −.09, P = .0186), but there was no difference in length of stay for participants with other syndromes (Table 3). In the multivariable model for the secondary outcome of 30-day mortality, being in the intervention group was associated with significantly decreased odds of mortality for participants with CAP (aOR = 0.582 95% CI, .396, .854, P = .0204) but was not associated with a difference in odds of mortality for other syndromes (Table 3).
Association of Intervention With Outcomes of Length of Stay and Mortality in Multivariable Models
| Association of Intervention With Length of Staya | Association of Intervention With Mortalityb | |||
|---|---|---|---|---|
| Syndrome | Coefficient Estimate [95% CI] | P-value | Adjusted Odds Ratio [95% CI] | P-value |
| All patients | −.056 [−.254, .142] | .5803 | .998 [.821, 1.214] | .9864 |
| UTI | .144 [−.256, .543] | .4809 | 1.494 [.968, 2.305] | .1284 |
| ABI | .063 [−.287, .413] | .7242 | .906 [.58, 1.416] | .7172 |
| Cellulitis | −.53 [−.97, −.09] | .0186 | .45 [.079, 2.567] | .4507 |
| CAP | −.163 [−.517, .192] | .3687 | .582 [.396, .854] | .0204 |
| Aspiration pneumonia | −.414 [−1.574, .746] | .4846 | 1.695 [.996, 2.886] | .1026 |
| Nursing home pneumonia | −.07 [−1.31, 1.17] | .9121 | 1.333 [.614, 2.894] | .5420 |
| Association of Intervention With Length of Staya | Association of Intervention With Mortalityb | |||
|---|---|---|---|---|
| Syndrome | Coefficient Estimate [95% CI] | P-value | Adjusted Odds Ratio [95% CI] | P-value |
| All patients | −.056 [−.254, .142] | .5803 | .998 [.821, 1.214] | .9864 |
| UTI | .144 [−.256, .543] | .4809 | 1.494 [.968, 2.305] | .1284 |
| ABI | .063 [−.287, .413] | .7242 | .906 [.58, 1.416] | .7172 |
| Cellulitis | −.53 [−.97, −.09] | .0186 | .45 [.079, 2.567] | .4507 |
| CAP | −.163 [−.517, .192] | .3687 | .582 [.396, .854] | .0204 |
| Aspiration pneumonia | −.414 [−1.574, .746] | .4846 | 1.695 [.996, 2.886] | .1026 |
| Nursing home pneumonia | −.07 [−1.31, 1.17] | .9121 | 1.333 [.614, 2.894] | .5420 |
Abbreviations: ABI, multidrug-resistant gram-negative organism; CAP, community acquired pneumonia; CI, confidence interval; UTI, urinary tract infection.
aA linear regression model was derived adjusting for baseline characteristics, cluster, and period effects.
bA logistic regression model was derived adjusting for baseline characteristics, cluster, and period effects.
Association of Intervention With Outcomes of Length of Stay and Mortality in Multivariable Models
| Association of Intervention With Length of Staya | Association of Intervention With Mortalityb | |||
|---|---|---|---|---|
| Syndrome | Coefficient Estimate [95% CI] | P-value | Adjusted Odds Ratio [95% CI] | P-value |
| All patients | −.056 [−.254, .142] | .5803 | .998 [.821, 1.214] | .9864 |
| UTI | .144 [−.256, .543] | .4809 | 1.494 [.968, 2.305] | .1284 |
| ABI | .063 [−.287, .413] | .7242 | .906 [.58, 1.416] | .7172 |
| Cellulitis | −.53 [−.97, −.09] | .0186 | .45 [.079, 2.567] | .4507 |
| CAP | −.163 [−.517, .192] | .3687 | .582 [.396, .854] | .0204 |
| Aspiration pneumonia | −.414 [−1.574, .746] | .4846 | 1.695 [.996, 2.886] | .1026 |
| Nursing home pneumonia | −.07 [−1.31, 1.17] | .9121 | 1.333 [.614, 2.894] | .5420 |
| Association of Intervention With Length of Staya | Association of Intervention With Mortalityb | |||
|---|---|---|---|---|
| Syndrome | Coefficient Estimate [95% CI] | P-value | Adjusted Odds Ratio [95% CI] | P-value |
| All patients | −.056 [−.254, .142] | .5803 | .998 [.821, 1.214] | .9864 |
| UTI | .144 [−.256, .543] | .4809 | 1.494 [.968, 2.305] | .1284 |
| ABI | .063 [−.287, .413] | .7242 | .906 [.58, 1.416] | .7172 |
| Cellulitis | −.53 [−.97, −.09] | .0186 | .45 [.079, 2.567] | .4507 |
| CAP | −.163 [−.517, .192] | .3687 | .582 [.396, .854] | .0204 |
| Aspiration pneumonia | −.414 [−1.574, .746] | .4846 | 1.695 [.996, 2.886] | .1026 |
| Nursing home pneumonia | −.07 [−1.31, 1.17] | .9121 | 1.333 [.614, 2.894] | .5420 |
Abbreviations: ABI, multidrug-resistant gram-negative organism; CAP, community acquired pneumonia; CI, confidence interval; UTI, urinary tract infection.
aA linear regression model was derived adjusting for baseline characteristics, cluster, and period effects.
bA logistic regression model was derived adjusting for baseline characteristics, cluster, and period effects.
DISCUSSION
We performed a very large (nearly 7000 patients), and the first randomized controlled trial of the WISCA antibiotic stewardship CDS tool. WISCA was deployed as a unique tool utilizing both predictive modeling and clinical guidelines to make recommendations for empiric antibiotics. The WISCA predictive model was not associated with improved outcomes for UTI and ABI, but guidelines-based antibiotic recommendations were associated with decreased LOS for cellulitis and decreased mortality for CAP.
WISCA utilization may not have improved outcomes because empiric antibiotic choice may not be crucial for patients with ABI and UTI, the 2 syndromes that relied on WISCA predictive modeling in our study. The majority of ABI cases included appendicitis or cholecystitis in which the primary treatment is source control rather than antibiotics [23, 24]. For patients with UTI, urine culture results are often available within 24 hours, allowing for pathogen-directed antibiotic therapy. Therefore, initial empiric antibiotic choice may not be as important in UTI as in syndromes where cultures take longer to result (eg, bacteremia). Randhawa et al found that WISCA had the potential to more than double the likelihood of adequate empiric antibiotic coverage among intensive care unit (ICU) patients with ventilator-associated pneumonia and catheter-associated bloodstream infection, patients whose outcomes may be more highly impacted by early adequate antibiotic therapy [17].
Although there were no overall differences in outcomes, when stratified by individual infectious syndromes there were several notable findings. For patients with cellulitis, being in the intervention group was associated with significantly shorter length of stay compared to the control group. Shorter LOS may have occurred because patients were more likely to receive an appropriate antibiotic regimen early in their stay, or because the regimen was more easily transitioned from intravenous to oral antibiotics to allow for earlier hospital discharge. For patients with CAP, those in the intervention group had lower odds of mortality compared to patients in the control group. Similarly, Dean et al found that use of electronic clinical decision support tool with pneumonia treatment guidelines in the emergency department was associated with lower mortality for patients with CAP [25].
In our study, providers in the intervention group followed recommendations to change antibiotics 60% of the time. Prior literature shows wide variability in provider acceptance of CDS recommendations, ranging from as low as 4% to >90% [26–28]. Many factors influence providers’ decisions to follow or reject antibiotic stewardship CDS recommendations, including undocumented patient comorbidities or allergies, severity of patient illness, and concern for additional sources of infection [28]. In the current study, an ASP physician communicated the WISCA-guided antibiotic recommendation to the treating provider via page and/or documentation in the EHR. If the WISCA tool were embedded within the EHR for the primary provider to utilize directly, it is not known if acceptance of the recommendations would be higher or lower.
Our study has several limitations. We randomized our intervention at the hospital level with a cross over design. As a result, providers in hospitals that began in the intervention arm may have changed their antibiotic prescription practices to more closely align with guidelines after receiving recommendations from study ASP physicians. These practice changes may have persisted when the hospitals crossed over from intervention to control arms, thereby biasing the control arms in the second half of the study. However, in the multivariable models, we did not find a period effect. Furthermore, there was a slight imbalance in baseline patient characteristics in the intervention versus control groups, with patients in the intervention group slightly older and less likely to be “observation status” upon admission. This may have biased the intervention group toward higher likelihood of adverse outcomes such as mortality, which did not occur. Moreover, the multivariable models controlling for baseline patient characteristic and study period did not show a difference in outcomes for patients in control versus intervention groups. Finally, ASP physicians were not blinded to the enrollment status of the patient when determining recommendations. Therefore, ASP physicians may have been biased when making antibiotic recommendations for intervention patients because the intervention required documentation in the medical record and might influence patients’ treatment course.
In conclusion, we found that use of WISCA to predict likelihood of empiric antibiotic coverage for UTI and ABI was not associated with improved outcomes among inpatients. However, use of guideline-based antibiotic recommendations for cellulitis and CAP did lead to improved patient outcomes. Future research is needed to understand if WISCA-guided recommendations would improve outcomes for other infectious syndromes in which outcomes may be more closely tied to early effective empiric antibiotic therapy, for example, sepsis.
Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.
Notes
Acknowledgments. The authors thank the NorthShore University HealthSystem Infectious Diseases physicians who participated in the antimicrobial stewardship reviews, Parul Patel, MT (ASCP), who helped manage the project, the Information Technology group at NorthShore for their assistance with creating EHR reports to facilitate the project, and Susan Boehm, RN, who reviewed patient outcomes for the intervention group with no bacterial infection.
Financial support. This study was funded by the Agency for Healthcare Research and Quality (AHRQ), US Department of Health and Human Services, grant 5R01HS022283. The opinions are those of the authors and do not reflect the official position of AHRQ or the US Department of Health and Human Services.
Potential conflicts of interest. A. R. reports a patent for system and methods for providing syndrome-specific, weighted-incidence treatment regimen recommendations. All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.
