Evaluation of an Opt-Out Protocol for Antibiotic De-Escalation in Patients With Suspected Sepsis: A Multicenter, Randomized, Controlled Trial

Abstract

Background

Sepsis guidelines recommend daily review to de-escalate or stop antibiotics in appropriate patients. This randomized, controlled trial evaluated an opt-out protocol to decrease unnecessary antibiotics in patients with suspected sepsis.

Methods

We evaluated non–intensive care adults on broad-spectrum antibiotics despite negative blood cultures at 10 US hospitals from September 2018 through May 2020. A 23-item safety check excluded patients with ongoing signs of systemic infection, concerning or inadequate microbiologic data, or high-risk conditions. Eligible patients were randomized to the opt-out protocol vs usual care. Primary outcome was post-enrollment antibacterial days of therapy (DOT). Clinicians caring for intervention patients were contacted to encourage antibiotic discontinuation using opt-out language. If continued, clinicians discussed the rationale for continuing antibiotics and de-escalation plans. To evaluate those with zero post-enrollment DOT, hurdle models provided 2 measures: odds ratio of antibiotic continuation and ratio of mean DOT among those who continued antibiotics.

Results

Among 9606 patients screened, 767 (8%) were enrolled. Intervention patients had 32% lower odds of antibiotic continuation (79% vs 84%; odds ratio, 0.68; 95% confidence interval [CI], .47–.98). DOT among those who continued antibiotics were similar (ratio of means, 1.06; 95% CI, .88–1.26). Fewer intervention patients were exposed to extended-spectrum antibiotics (36% vs 44%). Common reasons for continuing antibiotics were treatment of localized infection (76%) and belief that stopping antibiotics was unsafe (31%). Thirty-day safety events were similar.

Conclusions

An antibiotic opt-out protocol that targeted patients with suspected sepsis resulted in more antibiotic discontinuations, similar DOT when antibiotics were continued, and no evidence of harm.

Clinical Trials Registration

NCT03517007.

antimicrobial stewardship, sepsis, SEP-1, de-escalation, antibiotic

Early recognition and initiation of evidence-based interventions to improve sepsis care and outcomes have been a focus of national efforts. Specifically, the Centers for Medicare & Medicaid’s SEP-1 Core Measure includes a requirement for early administration of broad-spectrum antibiotics for patients with suspected sepsis [1]. Sepsis, however, is a challenging clinical diagnosis. Hypotension, lactic acidosis, and/or organ dysfunction may be caused by noninfectious etiologies or mixed with other pathologies, creating a confusing clinical picture [2].

While prompt use of antibiotics specifically for patients with septic shock is beneficial, antibiotic exposures can also have negative effects such as acquired drug-resistant infections, Clostridioides difficile infection, and adverse drug events. The Surviving Sepsis guidelines therefore recommend daily review to assess if antibiotics can be stopped or narrowed, a process called antibiotic de-escalation. In practice, patients with antibiotics started empirically for suspected sepsis often have broader regimens and longer antibiotic courses than necessary, even when an alternative diagnosis is more likely [3]. Importantly, missing from the SEP-1 measure is a balancing component to promote antibiotic de-escalation, especially among patients in whom infection has been ruled out or who do not have a sepsis syndrome.

Hospital antimicrobial stewardship programs aim to improve patient safety by promoting judicious antibiotic use and decreasing the risks associated with unnecessary antibiotic exposures [4]. Suspected sepsis is a clinical scenario where stewarding antibiotics is particularly challenging, especially as hospitals respond to the regulatory pressures of SEP-1 [5, 6].

Opt-out interventions provide a nudge to reevaluate or consider alternative clinical decisions, while the treating clinician retains ultimate decision-making capacity. This strategy has previously been used as a reminder for screening or as a prompt for evaluation. Assessments for antibiotic de-escalation may be best timed after pertinent negative culture data return and when clinicians can assess clinical response to antibiotic treatment and/or other interventions.

We designed an opt-out protocol for patients with suspected sepsis. The protocol aimed to identify low-risk patients in whom infection was not confirmed or who recovered quickly and were therefore candidates for antibiotic de-escalation. In this multicenter, randomized, controlled trial, we aimed to quantify the effect of the opt-out intervention on post-enrollment days of antibiotic therapy among selected patients with suspected sepsis.

METHODS

Study Design

De-escalating Empiric Treatment: Opting-OUt of Rx for Selected Patients with Suspected Sepsis (DETOURS) Trial was a patient-level, randomized, controlled trial performed in 10 acute care hospitals in the United States from September 2018 through May 2020. The protocol included 5 steps that all occurred on the same day: 1. eligibility screen, 2. safety check, 3. randomization, 4. opt-out procedure, and 5. guided de-escalation discussion (Supplement: DETOURS Protocol). Study teams included clinical pharmacists and/or physicians engaged in antimicrobial stewardship activities at their institution. Personnel who applied selection criteria and interacted directly with clinicians varied depending on local resources, existing antibiotic stewardship practices, and local implementation strategy (Supplementary Table 1).

Definitions, Randomization, and Intervention

Adult patients in nonintensive care units were screened for suspected sepsis using unit census lists that included antibiotic orders and blood culture data. Suspected sepsis was defined as having blood cultures that showed “no growth” at 48 to 96 hours after collection and active orders for broad-spectrum antibiotics [7]. Patients with blood culture results that indicated a likely contaminant (eg, single set with coagulase-negative Staphylococcus) and no central line were also eligible. Next, the study team reviewed charts to apply a 23-item safety checklist developed and vetted by an expert panel to exclude patients with ongoing signs and symptoms of systemic infection, concerning or inadequate microbiology data, or high-risk conditions [8]. Patients who passed the safety check were randomized in a 1:1 ratio using a predetermined schema stratified by hospital to receive either usual care (control) or the opt-out discussion (intervention).

For the intervention, a study team member contacted the primary clinician responsible for antibiotic decision-making. For most sites, clinical pharmacists or infectious diseases–trained pharmacists led the discussion (Supplementary Table 1). The protocol provided the following language for the verbal opt-out discussion: “[This patient] has passed the initial safety screen for de-escalation of antibiotics. Antibiotics will be stopped per protocol unless you opt-out.” If the clinician agreed to stop antibiotics, orders were changed to reflect this decision. If the clinician opted out and determined antibiotics should be continued, the study team member initiated a guided de-escalation discussion and asked the following questions: Why should antibiotics be continued in this patient?, What is the patient’s infection diagnosis?, Can you narrow the breadth of antibacterial coverage to the most likely pathogens?, and If the patient remains stable and no new clinical data emerge to suggest a different diagnosis, do you have an empiric de-escalation and duration-of-therapy plan?

Verbal interaction and response from the primary clinician were required for patients to be considered per protocol. Individual patients were enrolled only once. Study personnel were blinded to randomization status until step 3, but no further blinding was feasible due to required personal interactions.

Outcomes

The primary outcome was post-enrollment days of antibacterial therapy (DOT), defined as the number of calendar days of antibacterial use during hospitalization plus the intended days prescribed at discharge. Individual agent DOT were counted separately and summed per patient [9]. DOT counts began the day after enrollment up to 30 days. Discharge prescriptions were assumed to begin the day after discharge. Patients in whom antibiotics were stopped on the day of enrollment were assigned zero DOT. Secondary outcomes included 30-day safety events: C. difficile infection, deep vein thrombosis, readmission, new hemodialysis requirement, transfer to intensive care, death, new peripherally inserted central catheter line placement, and length of stay. Patients were assessed for relapse of suspected sepsis defined as restart or escalation of antibiotic therapy coupled with repeat blood culture collection and as reinitiation of inpatient antibiotics after >48 hours of no inpatient antibiotics. Secondary antibiotic use outcomes included total length of therapy (days of antibiotics in which agents are not separately counted), inpatient DOT, post-discharge DOT, and DOT by agent rank. Agent rank was based on breadth of antibiotic spectrum and priority for antimicrobial stewardship programs, as previously described (Protocol Appendix 2) [7]. The percent of patients with antibiotic de-escalation was defined as a reduction in agent rank and/or number of antibiotics comparing the enrollment day to 5 days after enrollment [7].

Statistical Analyses

Sample size was estimated using a simulation study based on 6 months of sample data from 3 hospitals (Supplement: DETOURS Statistical Analysis Plan). Target enrollment was set at 762, providing 87% power to calculate a 25% decrease in mean DOT.

The primary outcome was heavily zero-inflated; thus, traditional regression models did not fit the outcome distribution well. Therefore, DOT were assessed using a hurdle model with logistic regression to describe the process for zero counts and truncated negative binomial regression to describe the process for positive counts [10]. Therefore, model output included 2 comparisons of intervention to control: odds ratio of receiving at least 1 DOT and ratio of mean DOT among patients who had at least 1 DOT. Clinical and safety outcomes were assessed using summary statistics as well as a 6-level desirability-of-outcome ranking (DOOR) response adjusted for duration of antibiotic risk (RADAR) [11]. Exclusive DOOR categories were determined a priori. RADAR was calculated using the primary outcome. We calculated the probability of a better DOOR/RADAR with intervention compared with control for a randomly selected patient along with 95% confidence interval (CI) using the Wilcoxon rank sum test. Secondary antibiotic use outcomes were analyzed using hurdle models similar to the primary outcome or with descriptive statistics. Prespecified subgroup analyses of the primary outcome included comparisons by unit type (medical vs surgical or medical/surgical), hospital type (major academic medical center vs community), and hospital.

Data and Safety Monitoring

An independent safety monitor reviewed safety outcomes at enrollment milestones of 10%, 25%, and 50% to ensure no imbalance emerged during the study. DETOURS was approved by Duke University and other participating sites’ institutional review boards with a waiver of consent. Data collection and randomization steps used Research Electronic Data Capture and extracts from the electronic health record. Analyses were conducted using SAS, version 9.4 (SAS Institute, Inc, Cary, NC).

RESULTS

A total of 9606 patient admissions were screened; 9440 were evaluated by the safety check, and 767 (8%) were randomized and analyzed as intention-to-treat (Figure 1). Most patients did not pass the safety check, commonly due to receipt of antibiotics before blood culture collection (35%), a positive bacterial culture (26%), or new oxygen requirement (21%; Table 1). The independent monitor observed no imbalance in safety events at the prespecified enrollment milestones. Among 383 patients randomized to intervention, 25 (6.5%) did not complete the protocol: 14 were discharged or had antibiotics discontinued prior to the opt-out discussion and 11 did not have verbal interaction with the primary team.

Figure 1.

Flow of patients through the De-escalating Empiric Therapy: Opting Out of Rx for Suspected Sepsis Trial. The trial was performed in 10 US acute care hospitals from September 2018 through May 2020. Results of the opt-out discussions were available only for those in the intervention arm. Abbreviation: ICU, intensive care unit.

Table 1.

Safety Check Criteria Among 9440 Patients With Suspected Sepsis

n (%)	Academic Hospital	Community Hospital	Total Evaluated in Safety Check
	n = 5638	n = 3802	N = 9440
Passed safety check, met no criteria	409 (7)	358 (9)	767 (8)
Safety check criteria met
Antibiotics 48 hours prior to first blood culture	2234 (40)	1011 (28)	3245 (35)
Positive bacterial cultures in the previous 4 days	1383 (25)	1027 (28)	2410 (26)
New and persistent infiltrate on chest imaging in the last 4 days	1409 (25)	1007 (27)	2416 (26)
New or higher than baseline oxygen requirement	1061 (19)	926 (24)	1987 (21)
Fever (≥38.0°C) in the last 48 hours	1074 (19)	642 (17)	1716 (18)
White blood cell count >14 K/uL in the last 24 hours	838 (15)	641 (17)	1479 (16)
Actively taking immunosuppressant medications	878 (16)	307 (9)	1185 (13)
Diagnosis of bacteremia or bloodstream infection during this admission or from an outside hospital prior to transfer	686 (12)	353 (10)	1039 (11)
Incision and drainage procedure for infection in the last 7 days	475 (8)	317 (9)	792 (9)
Solid organ or bone marrow transplant	519 (9)	88 (2)	607 (7)
Diagnosis of osteomyelitis	372 (7)	177 (5)	549 (6)
Diagnosis of bronchiectasis	143 (3)	48 (1)	191 (2)
Diagnosis of neutropenia in the last 7 days	94 (2)	61 (2)	155 (2)
Diagnosis of cystic fibrosis	72 (1)	2 (1)	74 (1)
Diagnosis of asplenia or post-splenectomy	46 (1)	28 (1)	74 (1)
Diagnosis of human immunodeficiency virus/AIDS with documented CD4 count <200 cell/mm3 in the past 3 months	50 (1)	38 (1)	88 (1)
Diagnosis of endocarditis	68 (1)	31 (1)	99 (1)
1 out of 1 blood cultures with skin flora in the last 4 days	29 (1)	46 (1)	75 (1)
Diagnosis of empyema	31 (1)	12 (1)	43 (1)
Diagnosis of agammaglobulinemia	5 (1)	9 (1)	14 (1)
Currently pregnant	3 (1)	6 (1)	9 (1)
Diagnosis of bone marrow aplasia	4 (1)	2 (1)	6 (1)
Diagnosis of lung abscess	7 (1)	1 (1)	8 (1)

Demographics and clinical characteristics were similar between the 2 arms (Table 2). Fewer patients had antibiotics continued post-enrollment in the intervention arm compared with the control (Table 3): 301 (78.6%) vs 324 (84.4%); odds ratio, 0.68; 95% CI, .47–.98; P = .04. Among those who had antibiotics continued, days of therapy (mean 10.4 vs 9.9; ratio of means, 1.06; 95% CI, .88–1.26) and length of therapy (mean 8.3 vs 7.5; ratio of means, 1.12; 95% CI, .95–1.33) were not different. DOT distributions revealed that intervention patients had peaks at specific durations rather than a sloping curve as in the control group (Figure 2, Supplementary Figure 1). Fewer intervention patients were exposed to rank 3 (extended-spectrum) and rank 4 (protected) antibiotics, but the difference was only statistically significant for the rank 3 agent group (Table 3).

Figure 2.

Post-enrollment days of therapy among intervention and control patients in the De-escalating Empiric Treatment: Opting-OUt of Rx for Selected Patients with Suspected Sepsis (DETOURS) Trial. This overlapping histogram indicates the primary outcome distribution for both control and intervention arms of the DETOURS Trial as the percent of patients from each group with the specified days of therapy. Patients were enrolled at 48–96 hours after blood culture (day 2–3 of antibiotics). Thus, peaks in the intervention group were observed at typical antibiotic durations (eg, 14 days antibiotic duration peaks at 12 days on the post-enrollment days of therapy). Post-enrollment days of therapy were calculated up to 30 days after randomization.

Table 2.

Characteristics of 767 Patients Randomized in the De-escalating Empiric Therapy: Opting Out of Rx for Suspected Sepsis Trial

Characteristic	Intervention N = 383	Control N = 384
Age, median (Q1, Q3), y	63 (49, 73)	66 (53, 76)
Female sex, N (%)	189 (49)	173 (45)
Race/Ethnicity, N (%)
ȃWhite	187 (49)	195 (51)
ȃBlack	149 (39)	139 (36)
ȃNative American	18 (5)	25 (7)
ȃAsian/Pacific Islander	5 (1)	3 (<1)
ȃHispanic	0	2 (<1)
ȃOther/Unknown	24 (6)	20 (5)
Prior hospitalization in the last 90 days, N (%)	123 (32)	113 (29)
Prior surgery in the last 30 days, N (%)	36 (9)	28 (7)
Elixhauser comorbidity score, median (Q1, Q3)	11 (5, 19)	11 (4, 20)
Unit type at enrollment, N (%)a
ȃMedical	217 (57)	223 (58)
ȃMedical/Surgical	51 (13)	57 (15)
ȃSurgical	84 (22)	76 (20)
ȃTelemetry	15 (3)	12 (3)
ȃOther	15 (5)	14 (4)
Total length of hospital stay, median (Q1, Q3), d	5 (4, 9)	6 (4, 10)
Length of stay prior to enrollment, median (Q1, Q3), d	3 (3, 4)	3 (3, 4)
Intensive care unit exposure prior to enrollment, N (%)	57 (15)	52 (14)
International Classification of Diseases, Tenth Revision, Clinical Modification, infection diagnosis, N (%)b
ȃNone	94 (25)	106 (28)
ȃ>1 infection	102 (27)	103 (27)
ȃUrinary tract infection	54 (14)	51 (13)
ȃBloodstream/Septicemia	16 (4)	27 (7)
ȃSkin and soft tissue infection	43 (11)	42 (11)
ȃIntraabdominal infection (eg, cholecystitis)	36 (9)	19 (5)
ȃPneumonia	19 (5)	20 (5)
ȃEar, nose, throat infection	13 (3)	8 (2)
ȃGastrointestinal tract (eg, infectious colitis)	3 (1)	3 (1)
ȃBone and joint infection	2 (<1)	1 (<1)
ȃCentral nervous system infection	0	3 (1)
ȃGenital/Sexually transmitted infection	1 (<1)	1 (<1)

Missing data on unit type for 2 control patients and 1 intervention patient.

International Classification of Diseases, Tenth Revision, Clinical Modification, codes are determined at discharge, so data may have been influenced by the intervention.

Table 3.

Antibiotic Use and Safety Outcomes Among Patients in the De-escalating Empiric Therapy: Opting Out of Rx for Suspected Sepsis Trial

Outcome	Intervention N = 383	Control N = 384	Odds Ratio or Ratio of Meansa (95% CI), Intervention vs Control
Post-enrollment antibiotic use
Post-enrollment days of therapy, mean (SD)	8.2 (9.9)	8.3 (10.2)	…
Post-enrollment days of therapy, median (Q1, Q3)	5 (1, 12)	5 (2, 11.5)	…
Antibiotics received (non-zero days of therapy), N (%)	301 (78.6%)	324 (84.4%)	0.68 (.47–.98)
ȃ Mean (SD) days of therapy among those with antibiotics continued	10.4 (10.1)	9.9 (10.4)	1.06 (.88–1.26)
ȃ Mean (SD) length of therapy among those with antibiotics continued	8.3 (7.8)	7.5 (7.1)	1.12 (.95–1.33)
Received rank 3 (extended-spectrum) antibiotics	138 (36.0%)	167 (43.5%)	0.73 (.55–.98)
ȃ Mean (SD) days of therapy among those with rank 3 agents continued	7.6 (7.9)	6.9 (7.9)	1.14 (.84–1.55)
Received rank 4 (protected) antibiotics	13 (3.4%)	16 (4.2%)	0.81 (.38–1.70)
ȃ Mean (SD) days of therapy among those with rank 4 agents continued	5.1 (5.1)	7.1 (7.2)	0.65 (.23–1.85)
De-escalation: Reduction in number and/or rank of antibiotics by day 5 after blood culture	219 (57.5%)	202 (52.9%)	…
Safety outcomes
Sum of major safety events	133	157	…
ȃ Readmission	61 (15.9%)	57 (14.8%)	…
ȃ Relapse of suspected sepsis	30 (7.8%)	30 (7.8%)	…
ȃ Clostridioides difficile infection	4 (1.0%)	7 (1.8%)	…
ȃ Deep venous thrombosis	1 (0.3%)	6 (1.6%)	…
ȃ ICU admission	26 (6.8%)	33 (8.6%)	…
ȃ Hemodialysisb	1 (0.3%)	8 (2.1%)	…
ȃ Death	10 (2.6%)	16 (4.2%)	…
Peripherally inserted central catheter line placement	11 (2.9%)	11 (2.9%)	…
Post-randomization length of stay, median (interquartile range)	2 (1, 6)	2 (1, 6)	…
Reinitiation of inpatient antibiotic therapy after >48 hours of no antibiotics	16 (4.2%)	16 (4.2%)	…
30-day DOOR, N (%)c			Probability of a better DOOR/RADARd (95% CI), Intervention vs Control
ȃ1: alive	301 (78.6%)	289 (75.3%)	0.52 (.48–.56)
ȃ2: readmission, relapse of suspected sepsis, C. difficile infection, or deep venous thrombosis	31 (8.1%)	33 (8.6%)	…
ȃ3: ≥2 of items in DOOR = level 2	16 (4.2%)	18 (4.7%)	…
ȃ4: subsequent ICU admission or hemodialysis	25 (6.5%)	25 (6.5%)	…
ȃ5: subsequent ICU admission and hemodialysis	0 (0.0%)	3 (0.8%)	…
ȃ6: death	10 (2.6%)	16 (4.2%)	…

Abbreviations: CI, confidence interval; DOOR, desirability-of-outcome ranking; ICU, intensive care unit; RADAR, response adjusted for duration of antibiotic risk; SD, standard deviation.

Hurdle models produced 2 ratios comparing intervention to control: odds ratio of at least 1 day of antibiotic therapy among all patients and the ratio of mean days of antibiotic therapy only among those who received antibiotics.

Missing data for 6 (4 control, 2 intervention).

Patients were assigned to a single DOOR based on all outcomes experienced. For example, if a patient required ICU admission, then hemodialysis, and then died within the 30-day follow-up period, they would be assigned a rank of 6.

Probability estimate based on Wilcoxon rank sum test comparing intervention to control patients.

Safety outcomes demonstrated minor differences with fewer events occurring among intervention patients for most outcomes (Table 3) including C. difficile infection (4 vs 7), deep vein thrombosis (1 vs 6), intensive care unit admission (26 vs 33), hemodialysis (1 vs 8), and death (10 vs 16). Some patients experienced more than 1 adverse event. Thus, patients were assigned to exclusive DOOR categories. The probability of a better DOOR/RADAR was 52% (95% CI, 48%–56%) comparing intervention to control patients, demonstrating no difference in global patient outcomes.

Responses to opt-out discussions provided rationales for continuing antibiotics (Table 4). When opt-out discussions were held with a clinician on the medicine service, they more frequently agreed to stop antibiotics compared with surgery or infectious diseases. When the opt-out discussion involved a trainee, antibiotics were more often continued. Clinicians most frequently continued antibiotics for treatment of a localized infection (76%). The second most common reason for continuing antibiotics was belief that stopping antibiotics was unsafe (31%). Reported antibiotic indication was evenly distributed among urinary tract, intraabdominal, skin and soft tissue, and respiratory tract infections. Approximately two-thirds of clinicians did not verbalize a suspected or confirmed pathogen. Notably, 35 clinicians continued antibiotics for treatment of pneumonia, despite exclusion of patients with abnormal chest imaging or hypoxia from enrollment.

Table 4.

Opt-Out Discussion Results for Intervention Patients Who Completed the Protocol

Response		Stopped Antibiotics N = 59	Continued Antibiotics N = 299
Post-enrollment days of therapy, median (Q1, Q3)		0	6 (2–12)
Clinician service, N (%)a
	Medicine	56 (97)	234 (78)
	Surgery	2 (3)	31 (10)
	Infectious diseases	0	5 (2)
	Specialty surgery	0	29 (10)
Clinician type involved in discussion, N (%)a
	Physician	46 (79)	204 (70)
	Trainee physician	5 (9)	66 (23)
	Nurse practitioner	5 (9)	14 (5)
	Physician’s assistant	2 (3)	9 (3)
Rationale for continuing antibiotics (multiple responses possible)
	Treatment of localized infection		227 (76)
	Pending clinical data		61 (20)
	Inadequate initial culture or diagnostic workup		35 (12)
	Believe that stopping antibiotics is unsafe, not otherwise specified		93 (31)
	Clinical uncertainty		36 (12)
	Defer antibiotic decision-making to consultant		30 (10)
	Perceived administrative need for antibiotics (eg, to justify ongoing admission)		23 (8)
	Other		2 (<1)
Reported indication for antibiotics (single response)
Unknown source, but believe infection is present			31 of 299 (10)
Urinary tract infection			75 of 299 (27)
	Cystitis		52
	Pyelonephritis		16
	Unspecified		7
Intraabdominal infection			70 of 299 (25)
	Abscess, perforation, or diverticulitis		21
	Peritonitis, including Spontaneous bacterial peritonitis		13
	Colitis (not Clostridioides difficile)		13
	Biliary/Liver		12
	Appendicitis		2
	Other or unspecified		9
Skin and soft tissue infection			56 of 299 (20)
	Cellulitis		31
	Infected wound/ulcer		15
	Abscess		2
	Other or unspecified		8
Respiratory infection			55 of 299 (19)
	PneumoniaȃCommunity-acquired		29
	Hospital-acquired or healthcare-associated		6
	Exacerbation of chronic obstructive pulmonary disease		18
	Upper respiratory tract or unspecified		2
Otherb			12 of 299 (4)
Clinical data pending at time of discussion, among those indicating pending data (multiple responses possible)
	Pending radiology diagnostic imaging		6 of 61
	Pending surgical procedure		3 of 61
	Pending consultant input		3 of 61
	Pending microbiology test/culture		9 of 61
	Unspecified		39 of 61
Can verbalize a confirmed or suspected pathogena			101 of 299 (33.8)
Can verbalize de-escalation plan including durationa			212 of 299 (70.9)

Missing responses occurred in <2% of participants.

“Other” includes bone/joint (5), central nervous system (3), suspected culture-negative endocarditis (1), dental (1), endophthalmitis (1), pelvic inflammatory disease (1). Three-hundred fifty-eight intervention patients completed the opt-out protocol.

Subgroup analyses by unit type, academic vs community hospitals, and by hospital did not reveal significant differences (Supplementary Tables 2 and 3).

DISCUSSION

To our knowledge, the DETOURS trial is the first patient-level, multicenter, randomized, controlled trial to evaluate an antibiotic stewardship intervention. Participants were selected using narrowly defined safety check criteria in order to optimize both acceptance of the intervention as well as safety. The opt-out intervention resulted in significant decline in antibiotic continuations and decreased exposures to broad-spectrum antibiotics. Distributions of days of therapy among those in whom antibiotics were continued were not different.

Studies of antimicrobial stewardship interventions in inpatient settings rarely involve designs with patient-level randomization [12]. National guidelines for implementing an antimicrobial stewardship program make strong recommendations for core strategies such as preauthorization or prospective audit and feedback [4]. Evidence for these core strategies, however, was graded moderate due to nonrandomized, quasi-experimental, or crossover study designs. DETOURS is larger than other controlled trials of interventions to reduce unnecessary antibiotic use, with implementation across 10 hospitals. The most recent Cochrane review of studies to improve antibiotic prescribing in hospitalized patients included 58 randomized, controlled studies of 221 total reviewed studies [13]. Only 12 randomized, controlled trials reported antibiotic days, with size ranging from 13 to 503 participants, and most were conducted at single institutions. Our study demonstrates that stewardship interventions can be evaluated using patient-level, randomized methodology.

The DETOURS intervention used verbal opt-out language as a communication strategy and impacted immediate decisions to stop antibiotics. Established strategies performed after initiation of antibiotics, such as prospective audit and feedback and “handshake” stewardship rounds, emphasize relationships with clinicians; ongoing, repeated educational opportunities; and methods of persuasion [14–19]. The success of these personalized interventions is highly dependent on the time, skills, influence, and emotional intelligence of an antibiotic steward to develop trust with a treating clinician. In contrast, this study used screening tools to identify eligible patients, then standardized language and questions. The opt-out language set expectations for antibiotic discontinuation based on patient-specific review and screening criteria. Thus, it sped up decisions to stop antibiotics among patients who ultimately may have had antibiotics stopped at a later time point in the absence of an intervention. The single time point of intervention may explain why there was an impact on immediate decisions but not on subsequent antibiotic days. Prior evaluations of single, time-based self-evaluations of antibiotics by the treating clinician, such as an antibiotic “time-out,” have also shown little effect on antibiotic days [20–23]. Though not large in effect, antibiotic selection after the DETOURS intervention was narrower in spectrum and durations appeared more standard. To achieve larger effects, future refinements of the opt-out intervention might include additional persuasive communication techniques and/or multiple points of contact to directly address drivers of antibiotic use.

DETOURS was designed to identify a low-risk population of “sepsis rule-outs,” or patients quickly determined not to have sepsis syndromes or systemic infection. Safety outcomes were similar, proving that a de-escalation intervention did not result in adverse advents or sepsis relapse. The safety check resulted in a highly selected population and low proportion of enrolled patients. Given the small number of eligible patients, it is conceivable that the protocol was overly conservative. Antibiotic stewardship strategies must address trade-offs; a safety-focused screening approach may be more acceptable for primary clinicians but result in a smaller impact on antibiotic use. Further, while the screening strategy ensured safety, it is unclear if that safety screening was reassuring enough to adjust clinicians’ antibiotic decision-making. When asked, clinicians still cited safety as the second most common reason to continue antibiotics.

The rationales for antibiotic continuation in this study revealed recognized challenges in antimicrobial stewardship and sepsis care, particularly around diagnostic uncertainty and/or error. The “Ds” of antibiotic stewardship include choice of drug, dose, de-escalation, and duration, but diagnosis is truly the most challenging “D” [24]. Diagnosis, however, is not listed as a distinct target for optimization in the consensus definition of an antibiotic stewardship program [4, 25]. Clinicians may be satisfied with an indication of “sepsis” without a specific source or assume a source without supportive clinical evidence. In this study, “pneumonia” was a reported indication to continue antibiotics, despite exclusion of patients with radiographic infiltrates or ongoing hypoxia [26]. Thirty percent of clinicians who determined antibiotics should be continued were not able to voice anticipatory decision-making around antibiotic de-escalation and duration. We believe the intervention effect was limited by the presence of clinical uncertainty and risk-averse thought processes, which may be especially acute when the label of “sepsis” has been invoked. Standardized questions did not directly address these drivers. These results reinforce a need to improve and support clinical reasoning and accuracy of infection diagnosis as well as address antibiotic decisions made in scenarios of uncertainty. Further, we must develop communication and training strategies to address a skewed interpretation of the benefit vs risk of antibiotics. Indeed, many clinicians still consider “antibiotics just in case” to be safer than “watchful waiting” despite our growing understanding of antibiotic risks and the urgency of antibiotic resistance.

Limitations to this trial must be acknowledged. This multicenter study included both community and academic hospitals with varied baseline resources and stewardship practices (Supplementary Table 1), requiring a pragmatic approach. The variety in the study sample, however, also provides evidence that the intervention was feasibly applied in diverse settings. Rates of antibiotic stops varied by hospital, suggesting effects of site-specific implementation, baseline de-escalation practice, or personnel (Supplementary Table 3). Second, the safety screen required high levels of personnel time for chart reviews, which may or may not have led to significant impact for an individual patient and may limit feasibility for some programs. Efficiency in identifying real-time opportunities for antibiotic stewardship remains a significant barrier encountered in this intervention as well as other guideline-recommended stewardship strategies such as prospective audit and feedback, preauthorization, and handshake rounds [4, 16, 17]. Future strategies to automate screening processes using logic-based rules in electronic health records may help produce more targeted stewardship chart reviews. We did not perform a subjective assessment of antibiotic appropriateness, which is difficult to standardize among reviewers. Instead, the safety screen was designed to objectively identify a low-risk population most eligible for de-escalation. Criteria did not exclude some patients with localized, culture-negative infections and a legitimate indication for broad-spectrum antibiotics (eg, spontaneous bacterial peritonitis). Inclusion of such patients may have led to smaller effects. The study could not be fully blinded; thus, contamination may have led to learning over time and bias toward a null effect. Time-related effects are difficult to interpret due to the many changes in care delivery associated with the onset of the coronavirus disease 2019 pandemic (Supplementary Table 4).

This multicenter, randomized, controlled trial of an opt-out antibiotic stewardship intervention led to a one-third lower odds of antibiotic continuation among all patients but similar days of therapy among patients in whom antibiotics were continued. Future stewardship interventions must more directly address diagnostic uncertainty and perceptions of safety when addressing antibiotic decision-making for patients with suspected sepsis. These future strategies might include comparisons of communication techniques and communications training for antibiotic stewards who support front-line providers.

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 acknowledge the following contributors to the De-Escalation OpTing-OUt of Rx in Selected patients with Suspected Sepsis Trial: Gweneth Francis, Christine Zurawski, Neha Shah, Ebbing Lautenbach, Anne Jaskowiak, Naasha Talati, Keith Hamilton, Shawn Binkley, Stephen Saw, Steven Morgan, Vasilios Athans, Tiffany Lee, Amanda Binkley, Christo Cimino, Eric Locklear, and Alicia Nelson.

Disclaimer. The Centers for Disease Control and Prevention (CDC) had no role in study design; collection, analysis, and interpretation of data; writing of the report; and decision to submit for publication.

Financial support. This work was supported the CDC Prevention Epicenters Program (to D. J. A.; 1U54CK000164).

Author notes

Potential conflicts of interest. E. D. A. reports grants or contracts from the University of Maryland (paid to author), University of Chicago (paid to author), CDC Prevention Epicenter Program (paid to institution), Oxford University Clinical Research Unit (paid to author), CDC (paid to institution), and DASON Member Hospital Contracts (paid to institution); royalties or licenses from UpToDate (paid to author); consulting fees from the American College of Clinical Pharmacy (paid to author), Hospital Association of New York State (paid to author), Sarah Moreland Russel Consulting (paid to author), and HealthTrackRX (paid to author); and support for attending meetings and/or travel from the American Society of Microbiology (paid to author), Pew Charitable Trusts (paid to author), and Oxford University Clinical Research Unit (paid to institution). A. E. D. reports payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing, or educational events from Merck & Co (paid to author). M. Z. D. reports grants or contracts from the National Institutes of Health, GSK, Johnson & Johnson, and Contrafect (paid to institution); support for attending meetings and/or travel from GSK; and participation on a data and safety monitoring board or advisory board for GSK (paid to author). D. J. A. reports grants or contracts from Agency for Healthcare Research and Quality (to institution), royalties or licenses from UpToDate Online (paid to author), and other financial or nonfinancial interests from Infection Control Education for Major Sports, LLC (owner). M. K. reports grants or contracts from Agency for Healthcare Research and Quality (paid to institution) and the Massachusetts Department of Public Health (paid to institution) and royalties or licenses from UpToDate (paid to author). R. W. M. reports grants or contracts from the CDC (paid to institution) and the Agency for Healthcare Research and Quality (paid to institution), royalties or licenses from UpToDate, Inc. (paid to author), speaker honoraria for the North Carolina Statewide Program for Infection Control and Epidemiology (paid to author), support for attending meetings and/or travel from the Society for Healthcare Epidemiology of America, and is on the Society for Healthcare Epidemiology of America Board of Trustees. A. P. reports grants from Clinical and Translational Science Award (to Biostatistics, Epidemiology, and Research Design Core, within the Biostatistics and Bioinformatics Department at Duke University). J. C. P. reports serving on the advisory boards for Shionogi, Inc, and Gilead Sciences, Inc. 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.