Quantifying Empiric Antibiotic Use in US Children’s Hospitals

Data were obtained from the Pediatric Health Information System (PHIS), an administrative database containing demographic, pharmacy, procedure, and laboratory charge data for all discharges from 48 US children’s hospitals (Children’s Hospital Association, Overland Park, KS). Our analysis included 42 hospitals, with 3 hospitals excluded because of missing data and 3 hospitals excluded because of unexpectedly low numbers of blood culture charges, raising concerns for data quality issues.

Patients aged 3 months to 20 years who were discharged between July 1, 2016, and June 30, 2017, were included. We analyzed new episodes of suspected bacteremia, defined as a charge for a blood culture and a charge for a new parenteral broad-spectrum antibiotic occurring within 1 calendar day. Parenteral broad-spectrum antibiotics included third- and fourth-generation cephalosporins, piperacillin and tazobactam, carbapenems, fluoroquinolones, aztreonam, vancomycin, daptomycin, and linezolid. Patients receiving any of these antibiotics or an enteral fluoroquinolone or linezolid in the 3 days before blood culture collection were excluded. The day of blood culture collection was defined as day 0 (Fig 1A). A window of 1 day before or after the blood culture was necessary to capture blood culture collection and antibiotic administration in proximity to one another (eg, at 11:59 pm and 12:01 am) given that charges in PHIS are captured by calendar day. An episode of suspected bacteremia was classified as a rule-out only if the last continuous day of antibiotic charges after the blood culture charge occurred on day 0, +1, +2, or +3, with no charge for any antibiotic on day +4 (Fig 1B). Conversely, episodes with continuous antibiotic charges through day +4 after the blood culture were classified as treatment courses. Patients who received any antibiotic aside from those most commonly used for noninfectious conditions or prophylaxis (oral trimethoprim and sulfamethoxazole, penicillin, azithromycin, erythromycin, rifaximin, or neomycin) were classified as having received an antibiotic on that day. A cut point of up to day +3 was selected because a 48-hour duration of antibiotics may result in antibiotic charges over 3 calendar days. We performed 2 analyses. The first was restricted to patients who remained hospitalized on day +4. The second included patients discharged before day +4, and episodes of suspected bacteremia were classified as rule-out only if there was no charge for an enteral antibiotic on the day of or day before discharge. Additional variables collected from PHIS include baseline demographic and clinical data.

We performed a single-center validation in which PHIS data were compared with electronic health record data. Among the 2807 episodes meeting the PHIS suspected bacteremia definition, we reviewed 280 randomly selected charts (10%). Three patients with false-positive results who did not receive a parenteral broad-spectrum antibiotic within 1 calendar day of the blood culture were identified, all of whom received the intravenous form of vancomycin enterally. Among 2652 patients who did not meet the PHIS suspected bacteremia definition, we reviewed 530 randomly selected charts (20%) and identified no misclassification. Therefore, sensitivity and specificity of the suspected bacteremia definition were 100% and 99%, respectively. Among 1028 patients meeting the suspected bacteremia definition who were discharged on day +1, +2, or +3, we reviewed 204 randomly selected charts (20%). No charge for oral antibiotics on the day of discharge or day before discharge was 82% sensitive and 100% specific for identifying patients not discharged from the hospital on oral antibiotics.

The primary outcome was the proportion of all broad-spectrum antibiotic DOTs accounted for by episodes meeting the definition for rule-out only.

We identified 67 032 episodes of suspected bacteremia. Fourteen percent began on or after hospital day 3, and 21% occurred in the ICU. Ceftriaxone or cefotaxime was prescribed most often (58%), followed by vancomycin (35%; Table 1). The overall frequency of suspected bacteremia episodes was 11.4 per 100 discharges, with substantial hospital-level variation observed (range: 3.4–17 episodes per 100 discharges). The proportion of suspected bacteremia episodes in which vancomycin was prescribed ranged from 16% to 59% (Fig 2).

From 67 032 episodes of suspected bacteremia, 40 107 (60%) patients remained hospitalized on day +4. Among these 40 107 patients, 28 494 (71%) were classified as having received an antibiotic treatment course and 11 613 (29%) were classified as having undergone a rule-out only. Antibiotics administered for a rule-out only accounted for 34 925 DOTs of a total of 755 681 broad-spectrum antibiotic DOTs (4.6%). When all 67 032 episodes were included (retaining 26 925 episodes in which discharge occurred before day +4) and a charge for an oral antibiotic on the day of or day before discharge was used to classify patients, a total of 34 909 (52%) were classified as having received an antibiotic treatment course and 32 123 patients (48%) were classified as having undergone a rule-out only (Supplemental Fig 3). Antibiotics administered for a rule-out only accounted for 91 618 DOTs of a total of 755 681 inpatient broad-spectrum antibiotic DOTs (12%).

We estimated that nearly 50% of children with suspected bacteremia underwent a rule-out only. Although short in duration, these rule-outs occur frequently and account for an estimated 12% of all broad-spectrum antibiotic DOTs in freestanding children’s hospitals. In addition, because antibiotics used for this indication are administered for 48 to 72 hours or less, they are often not reviewed by antimicrobial stewardship programs using prospective audit with feedback, highlighting the importance of other interventions to optimize antibiotic use in this scenario.⁴ 

One key intervention may be reducing empirical anti–methicillin-resistant Staphylococcus aureus (anti-MRSA) therapy given that MRSA is infrequently identified in most populations and is declining in prevalence nationally.^5–7  We nevertheless identified a nearly fourfold difference in the rate of vancomycin use across 42 children’s hospitals, variation unlikely to be explained fully by differences in case mix, infection type, stewardship practices, or local antibiogram (eg, MRSA prevalence). A second opportunity relates to the urgency with which antibiotics are started; although rapid broad-spectrum antibiotic administration is appropriate for patients with septic shock, generalizing this recommendation to clinically stable patients risks overtreatment and a missed opportunity to identify either a focal bacterial infection (for which a narrower-spectrum antibiotic may be appropriate) or a noninfectious condition.^2,8  Finally, with continuous blood culture monitoring systems, bacteremia may effectively be ruled out within 24 hours, such that shortening the duration of an antibiotic rule-out may be feasible.^9–12 

Our study has several limitations. First, we used billing rather than clinical data to identify patients with suspected bacteremia and to quantify antibiotic use. However, previous studies, as well as our local validation, confirmed the accuracy of these administrative data.¹³  Second, our assessment of whether a bacterial infection was present was based on the prescribed duration of antibiotics rather than clinical or microbiologic data, which are not available in PHIS. Although this is a significant limitation of this work, a duration of antibiotics of ≥4 consecutive days is a duration beyond the period of time required for bacteremia to be ruled out, if not for a documented bacterial infection. A related limitation is that patients treated with antibiotics on nonconsecutive days (eg, patients with renal failure) may be misclassified as undergoing a rule-out only; however, we expect such patients to represent a minority of the cohort. Third, PHIS includes data only from freestanding children’s hospitals, so generalizability to community hospitals is limited given differences in case mix across settings. Finally, PHIS does not capture outpatient prescriptions, so we were unable to definitively classify the 40% of patients discharged before day +4 as having undergone a rule-out only versus a treatment course, nor can we confirm that patients did not receive outpatient antibiotics in the 3 days before day 0. Our single-center validation revealed that oral antibiotic administration on the day of discharge or the day before discharge is a reasonable surrogate for continued oral antibiotic therapy, although of note, this does not reflect patients who were sent home on parenteral therapy.

A significant volume of inpatient broad-spectrum antibiotic DOTs are prescribed to children undergoing an antibiotic rule-out only. Antibiotic choice, particularly empirical use of vancomycin, may represent novel stewardship targets.