OBJECTIVE: To measure the association between inpatient bronchiolitis prevalence (IBP) and the delivery of unnecessary tests and treatments to patients hospitalized with bronchiolitis.
METHODS: A multicenter, retrospective, cohort study was performed using the Pediatric Hospital Information System database. All patients 2 months to 2 years of age hospitalized with bronchiolitis during 2004–2008 at participating pediatric hospitals were included. Main outcome measures were the probability of receiving potentially unnecessary care for bronchiolitis, including steroids, intravenously administered antibiotics, chest or neck radiographs, and any laboratory tests during hospitalization.
RESULTS: During winter months, with each 1% absolute increase in IBP, patients were less likely to receive steroids (incidence rate ratio: 0.968 [95% confidence interval: 0.960–0.976]; P < .001), radiographs (incidence rate ratio: 0.988 [95% confidence interval: 0.984–0.992]; P < .001), and laboratory tests (incidence rate ratio: 0.992 [95% confidence interval: 0.988–0.995]; P < .001). During summer months, similar associations were observed for steroids and radiographs. No association between IBP and antibiotic use was observed during either time period.
CONCLUSIONS: The frequency with which several types of unnecessary care were delivered to patients with bronchiolitis seemed to decrease with increasing IBP. This finding suggests that an association exists between contextual information and care delivery during the management of acute illness, and it highlights the importance of such information for delivery of high-quality health care.
- relative risk regression
- laboratory tests
- length of stay
- United States
WHAT'S KNOWN ON THIS SUBJECT:
Inpatient care for patients with bronchiolitis is highly variable and often without measurable benefit. Contextual factors, including the prevalence of a condition in the community, might affect care delivery.
WHAT THIS STUDY ADDS:
The adjusted risk of receiving several types of unnecessary care seems to be lower when a hospital is caring for greater numbers of patients with bronchiolitis. Hospitals should consider disseminating information about the prevalence of bronchiolitis to care providers.
Bronchiolitis, which is a common cause of acute pediatric hospitalization,1 represents an interesting target for studies of in-hospital pediatric care delivery because of the widespread knowledge that many of the tests and treatments commonly ordered for bronchiolitis do not decrease the severity or duration of illness.2,–,7 Therefore, efforts to improve the quality of care for bronchiolitis have focused on decreasing utilization through implementation of prespecified “pathways” of care delivery.8,–,10 The growing body of evidence that supportive care provides patients with outcomes equal to those provided by more intense care has led to the publication of a prominent clinical guideline that reinforces the lack of utility of routine use of laboratory testing, steroids, radiography, and antibiotics for patients with uncomplicated viral bronchiolitis.11
If high-quality care for bronchiolitis is defined in part on the basis of what physicians choose not to do, why do patients continue to receive unnecessary care? The clinical presentation of viral bronchiolitis includes fever, tachypnea, and respiratory distress, all of which also can be symptoms of more-serious bacterial illnesses. Although diagnosis of a viral infection decreases the probability of other illnesses, it does not exclude other causes of illness among infants.12 Bronchiolitis is caused most commonly by respiratory syncytial virus, which occurs in seasonal epidemics,13 but the viruses that cause bronchiolitis are present in communities throughout the year, to varying degrees.14 To accept a clinical hypothesis that a patient has bronchiolitis and to provide high-quality, evidence-based care, physicians must integrate a patient's clinical presentation with contextual information that includes the “pretest” probability of illness. Failure to incorporate such information might lead to an inappropriately low degree of faith in a clinical diagnosis, which might contribute to overuse of diagnostic tests and empirical therapeutic measures.
In this study, we hypothesized that an important determinant of care delivery to patients with eventual diagnoses of bronchiolitis would be the inpatient bronchiolitis prevalence (IBP) in the hospital in which their care is being delivered. A priori, we postulated that an elevated IBP would increase physician/care team confidence that a patient's illness was caused by bronchiolitis and therefore would decrease the likelihood that the patient would receive care that is generally thought to provide little or no benefit for bronchiolitis, although it is useful for other illnesses such as asthma or bacterial pneumonia. In addition, we hypothesized that hospitalization during periods of increased IBP would be associated with decreased inpatient charges and length of stay, both mediated by a decrease in unnecessary resource utilization.
Data Sources and Patients
Data for this study were obtained from the Pediatric Health Information System (PHIS), an administrative database that contains inpatient, emergency department, ambulatory surgery, and observation-status data from 43 not-for-profit, tertiary care, pediatric hospitals in the United States. These hospitals are affiliated with the Child Health Corporation of America (CHCA) (Shawnee Mission, KS), a business alliance of children's hospitals. Data quality and reliability are ensured through a joint effort between the CHCA and participating hospitals. The data warehouse function for the PHIS database is managed by Thomson Reuters (Ann Arbor, MI). Data are deidentified at the time of data submission, and data are subjected to a number of reliability and validity checks before being included in the database. We examined demographic and treatment data for all children between 2 months and 2 years of age who were discharged after an inpatient stay with a primary diagnosis code of bronchiolitis (International Classification of Diseases, Ninth Revision [ICD-9], code 466.1) or bronchiolitis attributable to respiratory syncytial virus (ICD-9 code 466.11). For patients who are admitted to the hospital from an emergency department at the same institution, data on care delivered in the emergency department, including tests, radiographs, and medications, are included in inpatient PHIS records. We analyzed hospitalizations with dates of discharge between January 1, 2004, and December 31, 2008, although not all CHCA hospitals contributed data to the PHIS during the entire period. Data for patients younger than 2 months were not included in our data set because of the protocols for diagnosis and management of fever in this age group.15,16 We excluded patients who were hospitalized for >7 days as being unrepresentative of patients with uncomplicated bronchiolitis. If an individual was hospitalized more than once during the study period, only the first hospitalization was analyzed.
For estimation of the census at each hospital, all patients who were hospitalized at each CHCA hospital on a particular calendar day were enumerated. The proportion of each hospital's daily census represented by patients who eventually were discharged with a primary diagnosis of bronchiolitis was recorded as that day's IBP. We selected prevalence rather than absolute case number or statistical deviation from a normative state as our measure of hospital-wide bronchiolitis “intensity” because we think that prevalence represents a simple, unbiased estimate of hospital resources dedicated to the care of bronchiolitis. To prevent endogeneity, the IBP assigned to each patient was the IBP on the day before his or her admission.
The following variables were included as covariates in our regression models: age (in months), gender, length of stay (except in models where length of stay was an outcome), and season of admission (admissions during November through March were compared with admissions during other months). We included season of admission as an effect modifier in our models because of our hypothesis that the relationship between IBP and care delivery would differ during bronchiolitis season, when physicians might be more likely to think that a respiratory illness had a viral cause. We also constructed a measure of comorbid conditions modeled after that used by Lorch et al.17 For this binary variable, we considered the presence or absence of any of a group of prespecified comorbid conditions, including metabolic diseases, cancer, congenital disorders, seizures, cerebral palsy, cardiac disease, renal disease, and a tracheostomy, gastrostomy, or other “artificial opening,” as listed in secondary discharge diagnoses obtained from the PHIS database. All covariates were selected a priori as potential confounders before study analyses began; no additional variable selection procedure was conducted.
Outcomes and Modeling
We prespecified 4 binary care delivery outcomes of interest that might occur during hospitalization, namely, receipt of intravenously administered antibiotics, receipt of any corticosteroid, any laboratory testing, and any plain radiograph of the chest or neck. In previous studies, some of these therapies and diagnostic tests were shown to be both overused and variable across hospitals.18,19 During analyses, we refined the laboratory testing outcome by removing tests for viral antigens, which can be used for patient cohorting at institutions with shared rooms and therefore might be outside physician control. We also examined hospital charges (adjusted on the basis of the wage/price index for the hospital's location) and length of stay as outcomes. Charges were modeled by using the γ model with a logarithmic link20 and length of stay by using linear regression. Because our care delivery outcomes were common for the population under study, we modeled binary outcomes by using relative risk regression, which is a statistical technique that makes use of a Poisson model to calculate incidence rate ratios that can be interpreted accurately as relative risks.21 Each hospital was identified in models by using a dummy variable, to model the fixed effect of hospital-specific practices on outcomes. Generalized estimating equations were used to account for within-hospital clustering of data. All statistical analyses were conducted with R 2.10.1 (R Foundation for Statistical Computing, Vienna, Austria).
This study protocol was approved by the institutional review board at Seattle Children's Hospital. Publication of PHIS data was approved by the CHCA.
Our data set included information on 49 417 hospital discharges at 40 hospitals; 5.2% of those patients were subsequently readmitted and discharged again with a diagnosis of bronchiolitis during the study period. Figure 1A depicts IBP during the years covered by our data set and demonstrates clear seasonal increases in mean across-hospital IBP. Table 1 provides demographic information regarding the patients in our data set. The mean length of stay (2.54 days) was longer than the median (2 days), which suggests that the sample was skewed toward shorter lengths of stay. Eight percent of discharges in the sample had ≥1 diagnosis code consistent with a chronic comorbidity. All 4 of our markers of potential overutilization were common in the data set, with rates ranging from 26% to 70% in the sample. Overall, 23% of patients in the data set were billed for viral antigen tests, but these tests were excluded from our analysis of laboratory testing. The majority (87%) of all patients in our sample were admitted between October and March, which reflects the seasonality of bronchiolitis. During the months of October through March, the mean IBP in the data set was 2.8% (SD: 2.18%; range: 0%–14.3%). Hospitals exhibited substantial within- and across-hospital variability in IBP, and this variability seemed to be greater in the winter months (Fig 1B).
By using multivariate models for each type of care under study, we calculated the incidence rate ratio associated with a 1% increase in absolute IBP, using season of admission as an effect modifier (Table 2). The probabilities of receiving steroids, radiographs, and laboratory tests all were observed to decrease with increasing IBP during winter months, and the probabilities of receiving steroids and radiographs also were observed to decrease with increasing IBP during summer months. To illustrate the implications of the incident rate ratios derived from our models for a patient's absolute probability of receiving each type of care, we prepared Fig 2, which depicts predicted changes in probability for a 6-month-old male patient without comorbidities who is hospitalized for 3 days during the winter at an “average” hospital in our data set during bronchiolitis season. Our model examining inpatient charges revealed a small but statistically significant decrease in the length of stay; adjusted relative inpatient charges accrued by patients with bronchiolitis as IBP increased during winter months (charge ratio estimate: 0.994 [95% confidence interval: 0.991–0.997]; P < .001). No association between IBP and the mean length of stay was observed.
We found that increased IBP was associated with improved quality of care for children hospitalized with bronchiolitis across multiple pediatric hospitals in the United States. For 2 of these types of care (steroid administration and radiography), this association seemed to hold during both winter and summer months. This association persisted despite adjustment for both patient- and hospital-specific factors that might alter care delivery. Importantly, given currently published guidelines for the care of patients with uncomplicated cases of bronchiolitis, care seemed to improve as IBP increased, which suggests that systems responsible for determining care delivery for bronchiolitis are enhanced by IBP or some closely associated factor. As a practical matter, our results suggest that children who are hospitalized with a condition during periods when the hospital has more children with that condition on its wards might receive better care. Our findings also recapitulate volume-outcome studies at the institutional level and suggest that, during periods of higher IBP, hospitals might function more like “specialty” hospitals and achieve better outcomes, as has been shown for adult care.22,23
Although we concede that our use of administrative observational data places limits on our ability to demonstrate causality, we think that the best explanation for our findings is an association between IBP and physicians' or care teams' willingness to manage bronchiolitis on purely clinical grounds. The clinical definition of bronchiolitis is a confluence of concerning symptoms, many of which (eg, fever, hypoxemia, and respiratory distress) can suggest the possibility of more-serious illnesses. For care providers to withhold tests and treatments from a sick child, systems must somehow support the assumption that the child will benefit from less care. We hypothesize–but cannot prove–that care providers in hospitals “know” their own IBP, and that care delivery is altered as a result. On the level of a physician ordering a test or treatment, IBP might serve as a proxy measure for the community prevalence of bronchiolitis, reinforcing the likelihood that a wheezing febrile child does not need steroids.
It is important to note that physicians who make a rational decision to factor IBP into decision-making when IBP is high make an equally rational decision to consider IBP when it is low. Bayes' theorem suggests that a patient's probability of having bronchiolitis is a function of the community prevalence (pretest probability) of the illness; if IBP is an accurate proxy measure for community prevalence, than a patient who presents with cough, fever, and distress at a time of low IBP is more deserving of chest radiographs and empiric antibiotic treatment than is a patient who presents during an outbreak of respiratory syncytial virus. Without a standard method for diagnosis, hospitals and care providers must live with uncertainty, bounded by their desire to provide safe effective care.
The relationship between clinical confidence and the appropriate withholding of unnecessary tests and treatments is of paramount importance during an era of constrained health care funding. For bronchiolitis, this is of particular concern because of the dramatic increase in hospitalizations observed in recent decades.1 As an example, hospitals that order more repeated complete blood counts for patients with bronchiolitis have higher mean charges for bronchiolitis hospitalizations.19 Our finding that charges adjusted for length of stay decreased with increasing IBP, an effect that is most likely mediated by decreased provision of unnecessary care, suggests that increased clinical confidence can lead to increased value (quality for a given cost) of care. As performance-based health care reimbursement takes effect, however, it is unlikely that utilization measures will be sufficiently granular to allow for adjustment on the basis of physicians' confidence in their diagnoses.
The most well-tested approach to improving care delivery for bronchiolitis is the “care pathway.” Such pathways have been demonstrated to decrease antibiotic overuse,8 hospital readmission,9 viral antigen testing,24 bronchodilator use,25,26 and inpatient costs.25 We do not know what proportion of patients in our data set were cared for with the use of this type of care guideline or pathway or when such a hospital-wide pathway was implemented at a given hospital during the 5 years of data we analyzed. Although studies exploring variability in health care delivery have attempted to adjust for the presence of care pathways,27 we do not consider the presence or absence of such systems to be a confounder of our results, because such systems would be unlikely to be causally related to IBP. Our results, however, have implications for the effectiveness of those pathways; for such systems to affect utilization effectively, physicians must interact with them before a diagnosis has been made and treatment initiated. Studies of systems that provide point-of-care evidence and guidance to physicians caring for patients with bronchiolitis have shown promising results, with respect to both ambulatory and inpatient overutilization.28,29 We suggest that our results support providing up-to-date information to physicians regarding the prevalence of bronchiolitis in their communities, because physicians might be responsive to such information when making care decisions. Pediatric hospitals that operate high-volume outpatient clinics and emergency departments could disseminate information to community physicians regarding the volume of bronchiolitis cases they are observing, which would improve awareness of bronchiolitis prevalence and appropriate care delivery outside the hospital.
There are several limitations to our study that warrant comment. First, given the observational nature of our study, causality cannot be established, although the most plausible explanation seems to support it. Second, although we adjusted for severity of illness, our measures were limited by our use of administrative billing data. For this limitation to lead to bias in our results, however, children with more-severe illnesses or comorbidities would have to have been more likely to be admitted during times of high bronchiolitis prevalence (which is plausible) and also to have been less likely to receive one of our selected process-of-care measures (which seems implausible).
Our use of administrative data is complicated by the nature of discharge diagnoses, which are often made on the basis of negative findings of test-and-treat approaches to medicine. Some of the patients who contributed to our daily measure of IBP might have carried working diagnoses such as “possible pneumonia” or “rule out sepsis.” This would bias our measure of prevalence toward higher values, but such patients included in our care delivery measures would also bias the probability of receiving care upward. We found the opposite, which suggests that this putative bias is not an adequate explanation for our results. In addition, the clinical significance of the small changes in bronchiolitis care delivery we observed deserves consideration. We would argue that, given the large number of children admitted to the hospital with viral bronchiolitis on an annual basis, even small improvements in care delivery might generate systematic cost savings and improved patient and family experiences. Whether such improvements are worth the costs of systems that support them is a question for further research. Furthermore, our creation of a single model to explore the relationship between IBP and care delivery across multiple hospitals represents a simplification of the complex relationships that exist at each institution. Our observation of an association between IBP and care delivery in the cohort does not imply that such an association exists at every hospital in the cohort. Rather, the balance of the evidence supports the association in the population, and more research is needed to explore the systems that support or impede this association.
Finally, although we used data from a large number of hospitals, the extent to which patient populations and care delivery at CHCA hospitals are representative of other pediatric hospitals or nonpediatric hospitals that care for children is not known. We would surmise, however, that community hospitals exhibit even greater variability and deliver more unnecessary care than pediatric hospitals. Such hospitals might demonstrate less responsiveness to IBP than we observed, but the variability in their practices might make them more amenable to improvements in quality brought about by the institution of care pathways and the provision of contextual information regarding bronchiolitis prevalence. Improving the care of bronchiolitis at nonpediatric hospitals, where many cases of bronchiolitis are managed, remains a vital goal.
Patients with bronchiolitis who are hospitalized during periods of elevated IBP seem to be less likely to receive unnecessary care. We think that this finding represents a novel population-level observation of physician behavior. Systems that seek to penalize overutilization for bronchiolitis should be designed to account for the empiric and uncertain nature of the diagnosis of bronchiolitis and should ensure that potentially necessary care is not discouraged. Hospitals should consider the effects of unit- and hospital-level factors on physician decision-making and should engineer care systems that provide evidence to physicians at the time decisions are made.
- Accepted July 1, 2011.
- Address correspondence to William C. VanCleve, MD, Department of Anesthesia and Pain Medicine, University of Washington, 1959 NE Pacific Street, Box 356540, Seattle, WA 98195-6540. E-mail:
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
- PHIS —
- Pediatric Health Information System
- CHCA —
- Child Health Corporation of America
- IBP —
- inpatient bronchiolitis prevalence
- Friis B,
- Andersen P,
- Brenøe E,
- et al
- Wilson SD,
- Dahl BB,
- Wells RD
American Academy of Pediatrics, Subcommittee on Diagnosis and Management of Bronchiolitis. Diagnosis and management of bronchiolitis. Pediatrics. 2006;118(4):1774–1793
- Byington CL,
- Enriquez FR,
- Hoff C,
- et al
- Jaskiewicz JA,
- McCarthy CA,
- Richardson AC,
- et al
- Lorch SA,
- Millman AM,
- Zhang X,
- Even-Shoshan O,
- Silber JH
- Christakis DA,
- Cowan CA,
- Garrison MM,
- Molteni R,
- Marcuse E,
- Zerr DM
- Lumley T,
- Kronmal R,
- Ma S
- Perlstein PH,
- Kotagal UR,
- Bolling C,
- et al
- King WJ,
- Le Saux N,
- Sampson M,
- Gaboury I,
- Norris M,
- Moher D
- Copyright © 2011 by the American Academy of Pediatrics