|Home | About | Journals | Submit | Contact Us | Français|
Antibiotics are often given for inflammatory bowel disease (IBD) exacerbations, but their use among pediatric inpatients has not been assessed. We aimed to validate administrative data for identifying hospitalizations for IBD exacerbation and to characterize antibiotic use for IBD exacerbations across children's hospitals.
To validate administrative data for identifying IBD exacerbation, we reviewed charts of 409 patients with IBD at 3 US tertiary care children's hospitals. Using the case definition with optimal test characteristics, we identified 3450 children with 5063 hospitalizations for IBD exacerbation at 36 children's hospitals between January 1, 2007 and December 31, 2009, excluding those with diagnosis codes for specific bacterial infections. We estimated predicted and expected hospital-specific antibiotic utilization rates using mixed-effects logistic regression, adjusting for patient- and hospital-level factors.
Administrative codes for receipt of intravenous steroids or endoscopy provided 79% positive predictive value and 71% sensitivity for identifying hospitalizations for IBD exacerbation. Antibiotics were administered for ≥2 of the first 3 hospital days during 40.7% of IBD exacerbations in US children's hospitals; however, the proportion of patients receiving antibiotics varied significantly across hospitals from 27% to 71% (P < .001), despite adjustment for several patient- and hospital-level variables. Among those given antibiotics, the 3 most common regimens were metronidazole alone (26.9%), metronidazole with ciprofloxacin (10.3%), and ampicillin with gentamicin and metronidazole (7.0%).
Significant variability exists in antibiotic use for children hospitalized with IBD exacerbation, which is unexplained by disease severity or hospital volume. Further study should determine the optimal antibiotic therapy for this condition.
Inflammatory bowel disease (IBD) is a chronic and potentially debilitating intestinal inflammatory condition without clear cause, and it is estimated to affect 1 to 1.5 million people in the United States. Of those affected, approximately 100 000 are children, and childhood IBD incidence has doubled over the last decade [1–4]. In addition, patients with IBD consume significant healthcare resources and often require hospitalization for disease exacerbation management .
IBD pathogenesis remains incompletely understood, but many investigators hypothesize that IBD develops in the setting of an abnormal host response to either commensal or pathogenic gut flora . Antibiotic therapy for IBD exacerbations—intended to suppress host inflammation by targeting these gut flora—has been examined in small, randomized trials of adults with either Crohn's Disease [7–10] or ulcerative colitis [11–14] and meta-analyses [15, 16]. These trials have not convincingly identified antibiotic regimens effective in ameliorating IBD symptoms. Therefore, the role of antibiotics in IBD treatment remains unclear [17–19].
Significant variation exists in antibiotic use during initial outpatient management of children with IBD, with 10% to 70% of children receiving antibiotics within the first 3 months after diagnosis . Likewise, extreme variation exists among both academic IBD specialists and community gastroenterologists in the frequency of antibiotic use for various IBD-related conditions . Although variation in other elements of IBD care has been examined, antibiotic use specifically during hospitalizations for pediatric IBD exacerbation has not been assessed [22, 23].
In the setting of increasing bacterial antimicrobial resistance, a limited pipeline of new antimicrobials in development, and few data to support the efficacy of antibiotic use for IBD exacerbations, an important first step in improving the quality of antibiotic prescribing is examining current practice patterns . Therefore, we aimed to validate the identification of IBD exacerbation with a large, pediatric administrative database and to characterize antibiotic use in children hospitalized with IBD exacerbation across US children's hospitals, hypothesizing that between-hospital variation beyond expectation would exist.
The Pediatric Health Information System (PHIS) is an administrative database that contains comprehensive inpatient data from freestanding children's hospitals across the United States. Participating hospitals are affiliated with the Child Health Corporation of America (Shawnee Mission, KS), a business alliance of children's hospitals. Participating hospitals provide de-identified data that are subjected to rigorous reliability and validity checks before being incorporated into the database. PHIS data include demographic information (age, gender, ethnicity, and primary expected payer), medication and procedure information, and discharge-level information (length of stay, disposition, diagnoses, and procedures). PHIS has previously been used to examine inpatient pediatric practice variation [25, 26].
We used the PHIS database to identify all children ≤18 years with an International Classification of Diseases-9-Clinical Modification (ICD-9-CM) diagnosis of IBD (555.0-2, 555.9, 556.0-3, or 556.5-9) who were hospitalized between January 1, 2007 and December 31, 2009 at 3 hospitals: Children's Hospital of Philadelphia, Children's Mercy Hospital and Clinics in Kansas City, and Seattle Children's Hospital.
We performed chart review on a randomly selected cohort of 25% of these patients to determine whether the hospitalization was for IBD exacerbation. Subjects met our definition of IBD exacerbation if they had IBD without an infection, either as a cause of their symptoms or as a primary reason for hospitalization (eg, intestinal symptoms secondary to Clostridium difficile or hospitalization for pneumonia; however, we did not exclude intra-abdominal abscess, peri-anal abscess, and fistulae as “other infections” because these symptoms are manifestations of IBD exacerbation), and had any of the following: increased symptoms leading to hospitalization (increased pain, blood in stool, weight loss, etc) without alternate explanation; endoscopy- or biopsy-proven exacerbation; or need for increased immunosuppressive medications at discharge based on clinical or diagnostic findings during hospitalization.
We examined the test characteristics of several administrative data fields (including medication, procedure, and diagnosis data), both alone and in combination, for their ability to capture hospitalizations for IBD exacerbations identified by chart review.
After validating the use of PHIS for identifying IBD exacerbation, we performed a retrospective cohort study to assess inpatient variation in antibiotic therapy for children hospitalized with IBD exacerbation.
Our cohort was selected from all children ≤18 years who were discharged from 1 of 43 PHIS hospitals between January 1, 2007 and December 31, 2009 with an ICD-9-CM discharge diagnosis code for IBD. Because we were interested in maximizing both positive predictive value (PPV) and sensitivity to create a homogeneous and inclusive cohort of patients with IBD exacerbation, we used the algorithm that captured those subjects who had received intravenous (IV) steroids or endoscopy, which had a PPV of 79% and sensitivity 71% for identifying subjects with IBD exacerbation.
To capture antibiotics prescribed for treatment of IBD exacerbation, we targeted antibiotics given to subjects (1) at admission (to avoid capturing therapy for nosocomial infections) and (2) without an alternative diagnosis warranting antibiotic therapy. Thus, antibiotic exposure was defined as receipt of a systemic antibacterial agent for ≥2 of the first 3 days of hospitalization in children without ICD-9-CM codes for C difficile infection (008.45); septic arthritis (711.0, 711.9, 003.23, 098.50, 036.82); bacterial meningitis (320.X); upper respiratory tract infections (461.X, 462, 465.9); pneumonia (481, 482.X, 483.0, 485, 486); urinary tract infection (599.0); osteomyelitis (730.0-2); bacteremia (790.7); catheter-related bacteremia (999.31); soft tissue infections other than peri-anal abscess (680.X, 681.X, 682.X, 683); or Helicobacter pylori infection (041.86). We confirmed that we had excluded the vast majority of infections warranting antibiotic use by reviewing for potential exclusion all ICD-9-CM codes accounting for >0.1% of all diagnoses in the cohort. We also excluded admissions to hospitals with incomplete pharmacy data or with fewer than 25 IBD discharges over the 3-year study period, because these would be too few episodes to expect consistent therapeutic regimens.
Several variables were identified for adjustment in the multivariable model. To assess disease severity at hospital admission, we captured use of other standard IBD medicines (azathioprine, cyclosporine, methotrexate, steroids, and biologic therapies such as infliximab) within the first 3 days of hospitalization. We defined abdominal imaging as abdominal computed tomography, magnetic resonance imaging, ultrasound, or upper gastrointestinal series. Race was coded as a dichotomous variable (white vs non-white). We defined government insurance as Medicare, Medicaid, or any other government-sponsored insurance. PHIS provides a severity level variable based on the All Patient Refined-Diagnosis Related Groups score, which was assigned using age, sex, diagnoses, procedures, and discharge status, was coded by quartile, and is associated with risk of future readmission .
We described the distribution of demographic variables, covariates, and antibiotic use in the cohort using means, proportions, medians, and interquartile range (IQR). We assessed whether observed variation in antibiotic use for IBD might be attributed to patient case-mix differences across hospitals, and whether variation beyond that expected by different case mixes might exist.
To test the hypothesis of no difference across hospitals, we estimated hospital-specific rates of antibiotic use based on each hospital's mix of patient characteristics using 2 models: one assuming no difference and the other allowing for difference among hospitals . These 2 sets of rates were compared. In both models, the random effect represents the deviation of a hospital's antibiotic use from that expected for a hospital with the same patient characteristics and is assumed to follow a common normal distribution for all hospitals. Thus, the across-hospital variation is summarized at an aggregate level rather than at individual hospital levels, and the inferences are applicable to the broader collection of hospitals. We defined outlier hospitals as those in which antibiotic use was more than 2 standard deviations from the overall cohort mean. We examined univariate associations between all patient-level (ie, specific to a particular subject) and hospital-level (ie, specific to a particular hospital) factors and the risk of receiving antibiotics. We initially included all factors with P < .15 for their association with antibiotic use in our multivariate model, and we retained only those that remained significant in multivariate analysis.
Stata 11.0 (StataCorp, College Station, TX) was used for all analyses; a two-tailed P value <.05 was considered significant. This study was approved by the Institutional Review Board at each of the 3 participating institutions.
The validation subcohort included 409 children with ICD-9-CM codes for IBD hospitalized at 1 of 3 participating institutions; 401 (98%) had chart reviews consistent with IBD, and 258 (63%) of the hospitalizations were for IBD exacerbation.
Of the 409 subjects reviewed, the median age was 14 (IQR, 12-16); 195 (48%) received IV steroids (methylprednisolone, hydrocortisone, or prednisolone); 162 (40%) received other immunosuppression (adalimumab, azathioprine, certolizumab, infliximab, cyclosporine, mercaptopurine, methotrexate, or other); 104 (25%) underwent upper or lower endoscopy; 87 (21%) received a diagnosis for any infection; 84 (21%) underwent surgery; and 45 (11%) received abdominal imaging.
The algorithms that maximized PPV and sensitivity are included in Table 1. Other algorithms involving combinations of other immunosuppressive use, abdominal imaging, absence of infection, and surgery did not provide improved PPV or sensitivity over those described in Table 1.
From 10 641 PHIS admissions with ICD-9-CM codes for IBD at 43 hospitals during the 3-year study period, 6388 episodes met our algorithm definition of IBD exacerbation (ie, included either IV steroid receipt or endoscopy). From these, 673 were excluded for containing an ICD-9-CM code for specific infections, and another 652 were excluded because they occurred at hospitals either with incomplete pharmacy data or with fewer than 25 discharges for IBD during the study period. Thus, our final cohort included 5063 hospitalizations by 3450 unique patients in 36 hospitals. The median annual number of IBD exacerbation admissions per hospital was 53 (IQR, 39-70).
The median age at first admission during the study was 14.6 years (IQR, 11.9-16.5), and 51.9% of the cohort members were male. The median number of admissions per person was 1 (IQR, 1-2), with 95% of patients having ≤3 admissions. There were 1667 readmissions, of which 426 (25.6%) occurred within 30 days of the previous discharge (30-day readmissions). Median length of stay was 5 days (IQR, 3-8). More than 98% of admissions resulted in a discharge home; 4 (<0.1%) children died in the hospital (Table 2). The remaining children were transferred to another facility, left against medical advice, or had missing disposition data (approximately 1% total).
Antibiotics were prescribed for ≥2 of the first 3 days of hospitalization to 40.7% of children hospitalized with IBD exacerbation. Three specific antibiotic regimens were used in ≥5% of all subjects who received antibiotics: metronidazole alone (26.9%), metronidazole with ciprofloxacin (10.3%), or ampicillin with gentamicin and metronidazole (7.0%). The remaining 55.7% of subjects who received antibiotics received 1 of 273 unique regimens, each of which was used in <5% of all patients. A similar distribution was found when excluding those with an intestinal abscess or who had undergone a surgical procedure (including abscess drainage).
Increasing age and concomitant use of cyclosporine, methotrexate, and IV steroids were significantly associated with increased likelihood of antibiotic exposure, whereas subjects who received azathioprine were less likely to receive antibiotics (Table 3).
Patient-level factors specific to a hospitalization episode (hospitalization-level factors) associated with severity of illness (severity level, admission to the intensive care unit, ventilator use, total parenteral nutrition use, and performance of abdominal imaging) were each strongly associated with increased likelihood of antibiotic exposure. Likewise, children readmitted after a prior IBD hospitalization were >30% more likely to receive antibiotics than first-time admissions during the study period. Patients with concomitant diagnoses for intestinal or peri-anal abscess or for drainage procedures and other surgeries were also significantly more likely to receive antibiotics (odds ratio, 7.10; 95% confidence interval, 5.83-8.65; P < .001).
Use of antibiotics for patients with IBD varied across hospitals from 23.1% to 70.9% (Table 4), which did not change significantly after adjusting for demographic factors (age, sex), IBD-related factors (use of total parenteral nutrition, use of methotrexate or cyclosporine, upper or lower endoscopy during admission, and use of abdominal imaging), and case severity factors (severity level, intensive care unit admission, presence of surgical procedures during the admission, and presence of an abscess during the admission). Insurance status, calendar year, 30-day readmissions, need for mechanical ventilation, use of biologics (eg, infliximab), IV steroid use, or azathioprine use were not retained in the multivariable model because they were not significantly associated with antibiotic receipt when the other variables were included.
After adjustment, variability in antibiotic use ranged from 26.6% to 71.4% (P < .001; Figure 1), and 15 (42%) of the 36 hospitals included in the study were outliers, with 8 hospitals using significantly more antibiotics and 7 hospitals using significantly less than the group mean.
This degree of adjusted across-hospital variation remained even when 30-day readmissions (P < .001), encounters with identified abscesses or surgeries (P < .001), or both (P < .001) were excluded. Significant variation (P < .001) also remained when examining only patients who received both IV steroids and endoscopy and without an ICD-9-CM code for infection (n = 1251), a group with a 98% PPV and 97% specificity for IBD exacerbation based on our chart validation. Lastly, adjusted, across-hospital variation also remained significant when examining only patients prescribed 1 of the 3 most common antibiotic combinations: metronidazole (P < .001), metronidazole with ciprofloxacin (P < .001), or ampicillin with gentamicin and metronidazole (P < .001).
Our large, retrospective cohort study of children with IBD admitted to tertiary care pediatric hospitals has demonstrated that (1) administrative data can be used to identify pediatric hospitalizations for IBD exacerbation, (2) 40% of children hospitalized for IBD exacerbation without concomitant bacterial infection receive antibiotics, and (3) there is significant institutional variation in both the frequency and choice of antibiotic therapy for children hospitalized for IBD exacerbation.
The proportion of patients with IBD exacerbation who were prescribed antibiotics varied widely across hospitals, with some hospitals using antibiotics in approximately 25% of patients and others in almost 75%. This degree of variability remained despite adjustment for several patient-level factors and hospital-level factors and, thus, was not explained by patient severity of illness or the number of IBD patients treated at individual institutions.
Because additional variations in clinical presentation might have driven antibiotic use, we analyzed multiple subcohorts. Neither the exclusion of 30-day readmissions for IBD (to avoid selection of initial treatment failures) nor the exclusion of those with defined abscesses and surgical procedures affected our results. We analyzed only antibiotics given for ≥2 of the first 3 hospital days, to capture antibiotics begun at admission for IBD while avoiding both antibiotics begun later in the hospitalization for nosocomial infections and antibiotics ordered and discontinued quickly (and possibly given inadvertently). When antibiotics were prescribed, the specific regimens chosen varied significantly across hospitals. The majority of children hospitalized with IBD exacerbation and who received antibiotics were given an uncommon regimen, one used in <5% of all such encounters.
Because differences in patient demographics, disease severity, or hospital characteristics did not drive variability in antibiotic use for IBD exacerbation, other factors (such as participation in an IBD collaborative research network or clinician preference) may have contributed. Furthermore, the role of antibiotics in IBD exacerbation treatment—and the patient population that might benefit from those antibiotics—remains unclear. The current degree of antibiotic use variability for patients with IBD exacerbation highlights 2 ways in which consensus and standardization regarding antibiotic use for this indication should improve patient care.
First, the spectra of activity of the regimens used for IBD exacerbations are not uniform. If a common intestinal pathogen contributes to IBD exacerbation pathogenesis, the efficacy of these different antibiotic regimens in ameliorating IBD exacerbations will be different based on their activity against that pathogen. Regimens with activity against the relevant pathogens should therefore improve symptoms faster than those without such activity, suggesting that some antibiotic regimens will more effectively treat IBD exacerbations. Second, less efficacious regimens may place children at unnecessary risk of developing antibiotic-resistant infections and antibiotic-related adverse events without the benefit of ameliorating their IBD exacerbations. Therefore, future studies should examine the comparative effectiveness of different antibiotic regimens and durations for pediatric IBD treatment to determine which are most effective in relieving symptoms.
Use of administrative data has several limitations, with misclassification of either outcome or exposure being the major risk. Use of ICD-9-CM codes to identify IBD patients has been validated in an outpatient pediatric sample, and we confirmed through our own multicenter validation in this study that 98% of the patients hospitalized with an ICD-9-CM code for IBD actually have IBD . Moreover, we demonstrated that an administrative data-based algorithm has moderate sensitivity and good PPV for identifying hospitalizations for IBD exacerbation—a common, specific event experienced by IBD patients that is often targeted for antibiotic therapy. Because an administrative data definition for IBD exacerbation should require subjects to have an IBD diagnosis code, our starting population was all children with an IBD diagnosis code. Whereas some IBD patients with exacerbation but without a diagnosis code for IBD may have existed and would have been missed by our algorithm, their frequency is likely to be quite low and would have decreased our reported sensitivity trivially.
To further avoid the inappropriate classification of antibiotic use for IBD exacerbation, we excluded patients with ICD-9-CM codes for bacterial infections to remove patients with alternative indications for antibiotic therapy. Although some of these antibiotics may have been continued as part of prior outpatient management, their inclusion would be nondifferential across hospitals and would therefore lead us to underestimate the true across-hospital variation in antibiotic prescribing. Lastly, although we were unable to use common IBD symptom severity measures (such as the Pediatric Crohn's Disease and Ulcerative Colitis Activity Indices), we adjusted for illness severity, illness using multiple other patient-level factors (including the use of concomitant IBD medicines, endoscopy, imaging, and surgery), and intensive care unit admission [30, 31].
In summary, using a validated mechanism to identify children with IBD exacerbation, we found significant variation in antibiotic use for children hospitalized with IBD exacerbations across pediatric tertiary care centers. This variation was not explained by severity of illness or other patient- and hospital-level factors. Future comparative effectiveness studies of different antibiotic regimens are warranted to optimize the role of antibiotic use for children hospitalized with IBD exacerbation.
All authors have contributed significantly to this work and have given final approval of the manuscript.
Financial support. This work was supported by National Research Service Award Institutional Training Grant T32-HD043021 (to MPK). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Dr. Kronman had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Potential conflicts of interest. TEZ has received research support from or submitted grants to Merck, Pfizer, and Viropharma for unrelated projects and has served on an advisory board for Pfizer, Merck, and Hemocue. He has also received honoraria for speaking from Merck and has served as an expert witness for Butler Snow LLP. DMZ has received research funding from Sage Products Inc. and Viguard Ltd. for unrelated projects. All other authors: No reported conflicts.